The 'scylla sstable-summary' GDB command crashes with
'ValueError: Argument "count" should be greater than zero' when
inspecting ms-format (trie-based) sstables. This happens because
ms-format sstables don't populate the traditional summary structure,
leaving all fields zeroed out, which causes gdb.read_memory() to be
called with a zero count.
Fix by:
- Adding zero-length guards to sstring.to_hex() and sstring.as_bytes()
to return early when the data length is zero, consistent with the
existing guard in managed_bytes.get().
- Adding the same guard to scylla_sstable_summary.to_hex().
- Detecting ms-format sstables (version == 5) early in
scylla_sstable_summary.invoke() and printing an informative message
instead of attempting to read the unpopulated summary.
Fixes: SCYLLADB-1180
Closesscylladb/scylladb#29162
In test_index_requires_rf_rack_valid_keyspace, the create_table call
for a plain tablet-based table can fail with 'Unable to reach schema
agreement' after the server's 10s timeout is exceeded. This happens
when schema gossip propagation across the 4-node cluster takes longer
than expected after a sequence of rapid schema changes earlier in the
test.
Add a retry (up to 2 attempts) on schema agreement errors for this
specific create_table call rather than increasing the server-side
timeout.
Fixes: SCYLLADB-1135
Closesscylladb/scylladb#29132
Potential fix for https://github.com/scylladb/scylladb/security/code-scanning/139,
To fix the problem, explicitly restrict the `GITHUB_TOKEN` permissions
for this workflow/job so it has only what is needed. The script reads PR
data and repository info (which is covered by `contents: read`/default
read scopes) and posts a comment via `github.rest.issues.createComment`,
which requires `issues: write`. No other write scopes (e.g., `contents:
write`, `pull-requests: write`) are necessary.
The best fix without changing functionality is to add a `permissions`
block scoped to this job (or at the workflow root). Since we only see a
single job here, we’ll add it under `check-fixes-prefix`. Concretely, in
`.github/workflows/backport-pr-fixes-validation.yaml`, between the
`runs-on: ubuntu-latest` line (line 10) and `steps:` (line 11), add:
```yaml
permissions:
contents: read
issues: write
```
This keeps the token minimally privileged while still allowing the script
to create issue/PR comments.
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Closesscylladb/scylladb#27810
`read_barrier(session2)` was supposed to ensure `node2` has caught up on schema
before a CL=ALL write. But `patient_cql_connection(node2)` creates a
cluster-aware driver session `(TokenAwarePolicy(DCAwareRoundRobinPolicy()))`
that can route the barrier CQL statement to any node — not necessarily `node2`.
If the barrier runs on `node1` or `node3` (which already have the new schema),
it's a no-op, and `node2` remains stale, thus the observed `WriteFailure`.
The fix is to switch to `patient_exclusive_cql_connection(node2)`,
which uses `WhiteListRoundRobinPolicy([node2_ip])` to pin all CQL to `node2`.
This is already the established pattern used by other tests in the same file.
Fixes: SCYLLADB-1139
No need to backport yet, appeared only on master.
Closesscylladb/scylladb#29151
The test intentionally creates huge index pages.
But since 5e7fb08bf3,
the index reader allocates a block of memory for a whole index page,
instead of incrementally allocating small pieces during index parsing.
This giant allocation causes the test to fail spuriously in CI sometimes.
Fix this by disabling sstable compression on the test table,
which puts a hard cap of 2000 keys per index page.
Fixes: SCYLLADB-1152
Closesscylladb/scylladb#29152
Use add_new_sstable_and_update_cache() when attaching SSTables
downloaded by the node-scoped local loader.
This is the correct variant for new SSTables: it can unlink the
SSTable on failure to add it, and it can split the SSTable if a
tablet split is in progress. The older
add_sstable_and_update_cache() helper is intended for preexisting
SSTables that are already stable on disk.
Additionally, downloaded SSTables are now left unsealed (TemporaryTOC)
until they are successfully added to the table's SSTable set. The
download path (download_fully_contained_sstables) passes
leave_unsealed=true to create_stream_sink, and attach_sstable opens
the SSTable with unsealed_sstable=true and seals it only inside the
on_add callback — matching the pattern used by stream_blob.cc and
storage_service.cc for tablet streaming.
This prevents a data-resurrection hazard: previously, if the process
crashed between download and attach_sstable, or if attach_sstable
failed mid-loop, sealed (TOC) SSTables would remain in the table
directory and be reloaded by distributed_loader on restart. With
TemporaryTOC, sstable_directory automatically cleans them up on
restart instead.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-1085.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Closesscylladb/scylladb#29072
The native Scylla nodetool reports ECONNREFUSED as 'Connection refused',
not as 'ConnectException' (which is the Java nodetool format). Add
'Connection refused' to the valid_errors list so that transient
connection failures during concurrent decommission/bootstrap topology
changes are properly tolerated.
Fixes SCYLLADB-1167
Closesscylladb/scylladb#29156
When a `with_connect` operation timed out, the underlying connection
attempt continued to run in the reactor. This could lead to a crash
if the connection was established/rejected after the client object had
already been destroyed. This issue was observed during the teardown
phase of a upcoming high-availability test case.
This commit fixes the race condition by ensuring the connection attempt
is properly canceled on timeout.
Additionally, the explicit TLS handshake previously forced during the
connection is now deferred to the first I/O operation, which is the
default and preferred behavior.
Fixes: SCYLLADB-832
Backports to 2026.1 and 2025.4 are required, as this issue also exists on those branches and is causing CI flakiness.
Closesscylladb/scylladb#29031
* github.com:scylladb/scylladb:
vector_search: test: fix flaky test
vector_search: fix race condition on connection timeout
The code in upload_file std::move()-s vector of names into
merge_objects() method, then iterates over this vector to delete
objects. The iteration is apparently a no-op on moved-from vector.
The fix is to make merge_objects() helper get vector of names by const
reference -- the method doesn't modify the names collection, the caller
keeps one in stable storage.
Fixes#29060
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29061
Switch directories::do_verify_owner_and_mode() from lister::scan_dir() to
utils::directory_lister while preserving the previous hidden-entry
behavior.
Make do_verify_subpath use lister::filter_type directly so the
verification helper can pass it straight into directory_lister, and keep
a single yielding iteration loop for directory traversal.
Minus one scan_dir user twards scan_dir removal from code.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29064
The update_credentials_and_rearm() may get "empty" credentials from
_creds_provider_chain.get_aws_credentials() -- it doesn't throw, but
returns default-initialized value. In that case the expires_at will be
set to time_point::min, and it's probably not a good idea to arm the
refresh timer and, even worse idea, to subtract 1h from it.
Fixes#29056
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29057
The format string had two {} placeholders but three arguments, the
_upload_id one is skipped from formatting
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29053
The test_restore_primary_replica_same_domain and test_restore_primary_replica_different_domain tests have very much in common. Previously both tests were also split each into two, so we have four tests, and now we have two that can also be squashed, the lines-of-code savings still worth it.
This is the continuation of #28569
Tests improvement, not backporting
Closesscylladb/scylladb#28994
* github.com:scylladb/scylladb:
test: Replace a bunch of ternary operators with an if-else block
test: Squash test_restore_primary_replica_same|different_domain tests
test: Use the same regexp in test_restore_primary_replica_different|same_domain-s
The do_test_backup_abort() fetched the node's workdir and resolved cf_dir
solely to construct a unique-ish backup prefix:
prefix = f'{cf_dir}/backup'
The comment already acknowledged this was only "unique(ish)" — relying
on the UUID-derived cf_dir name as a uniqueness source is roundabout.
unique_name() is already imported and used for exactly this purpose
elsewhere in the file.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29030
The endpoint URL remains intact. Having it next to another toppartitions
endpoint (the /column_family/toppartitions one) is natural.
This endpoint only needs sharded<replica::database>&, grabs it from
http_context and doesn't use any other service. In column_family.cc the
database reference is already available as a parameter. Once more user
of http_context.db is gone.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Closesscylladb/scylladb#28996
The test in question uses several helpers from the backup sute, but it doesn't really need them -- the operations it want to perform can be performed with standard pylib methods. "While at it" also collect some dangling effectively unused local variables from this test (these were apparently left from backup tests this one was copied-and-reworked from)
Enhancing tests, not backporting
Closesscylladb/scylladb#29130
* github.com:scylladb/scylladb:
test/refresh: Simplify refresh invocation
test/refresh: Remove r_servers alias for servers
test/refresh: Replace check_mutation_replicas with a plain CQL SELECT
test/refresh: Inline keyspace/table/data setup in test_refresh_deletes_uploaded_sstables
test/refresh: Prepare indentation for new_test_keyspace in test_refresh_deletes_uploaded_sstables
test/refresh: Decouple test_refresh_deletes_uploaded_sstables from backup tests
test/refresh: Remove unused wait_for_cql_and_get_hosts import
Two issues found in the lister returned by gs_client_wrapper::make_object_lister()
Lister can report EOF too early in case filter is active, another one is potential vector out-of-bounds access
Fixes#29058
The code appeared in 2026.1, worth fixing it there as well
Closesscylladb/scylladb#29059
* github.com:scylladb/scylladb:
sstables: Fix object storage lister not resetting position in batch vector
sstables: Fix object storage lister skipping entries when filter is active
When it deadlocks, groups stop merging and compaction group merge
backlog will run-away.
Also, graceful shutdown will be blocked on it.
Found by flaky unit test
test_merge_chooses_best_replica_with_odd_count, which timed-out in 1
in 100 runs.
Reason for deadlock:
When storage groups are merged, the main compaction group of the new
storage group takes a compaction lock, which is appended to
_compaction_reenablers_for_merging, and released when the merge
completion fiber is done with the whole batch.
If we accumulate more than 1 merge cycle for the fiber, deadlock
occurs. Lock order will be this
Initial state:
cg0: main
cg1: main
cg2: main
cg3: main
After 1st merge:
cg0': main [locked], merging_groups=[cg0.main, cg1.main]
cg1': main [locked], merging_groups=[cg2.main, cg3.main]
After 2nd merge:
cg0'': main [locked], merging_groups=[cg0'.main [locked], cg0.main, cg1.main, cg1'.main [locked], cg2.main, cg3.main]
merge completion fiber will try to stop cg0'.main, which will be
blocked on compaction lock. which is held by the reenabler in
_compaction_reenablers_for_merging, hence deadlock.
The fix is to wait for background merge to finish before we start the
next merge. It's achieved by holding old erm in the background merge,
and doing a topology barrier from the merge finalizing transition.
Background merge is supposed to be a relatively quick operation, it's
stopping compaction groups. So may wait for active requests. It
shouldn't prolong the barrier indefinitely.
Tablet tests which trigger merge need to be adjusted to call the
barrier, otherwise they will be vulnerable to the deadlock.
Fixes SCYLLADB-928
Backport to >= 2025.4 because it's the earliest vulnerable due to f9021777d8.
Closesscylladb/scylladb#29007
* github.com:scylladb/scylladb:
tablets: Fix deadlock in background storage group merge fiber
replica: table: Propagate old erm to storage group merge
test: boost: tablets_test: Save tablet metadata when ACKing split resize decision
storage_service: Extract local_topology_barrier()
This patch series introduces a new documentation for exiting guardrails.
Moreover:
- Warning / failure messages of recently added write CL guardrails (SCYLLADB-259) are rephrased, so all guardrails have similar messages.
- Some new tests are added, to help verify the correctness of the documentation and avoid situations where the documentation and implementation diverge.
Fixes: [SCYLLADB-257](https://scylladb.atlassian.net/browse/SCYLLADB-257)
No backport, just new docs and tests.
[SCYLLADB-257]: https://scylladb.atlassian.net/browse/SCYLLADB-257?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQClosesscylladb/scylladb#29011
* github.com:scylladb/scylladb:
test: add new guardrail tests matching documentation scenarios
test: add metric assertions to guardrail replication strategy tests
test: use regex matching in guardrail replication strategy tests
test: extract ks_opts helper in test_guardrail_replication_strategy
docs: document CQL guardrails
cql: improve write consistency level guardrail messages
This PR enables the node_exporter systemd collector and configures the unit whitelist to include scylla-server.service and systemd-coredump services.
**Motivation**: We currently lack visibility into system-level service states, which is critical for diagnosing stability issues.
This configuration enables two specific use cases:
- Detecting Coredump Loops: We encounter scenarios where ScyllaDB enters a restart loop. To pinpoint SIGSEGV (coredumps) as the root cause, we need to track when the systemd-coredump service becomes active, indicating a dump is being processed.
- Identifying Startup Failures: We need to detect when the scylla-server unit enters a failed state. This is essential for catching unrecoverable errors (e.g., corrupted commitlogs or configuration bugs) that prevent the server from starting.
example of promql queries:
- `node_systemd_unit_state{name=~"systemd-coredump@.*", state="active"} == 1`
- `node_systemd_unit_state{name="scylla-server.service", state="failed"} == 1`
Closes#28402
reader_concurrency_semaphore::signal() guards against available
resources exceeding the initial limit after a signal, which would
indicate a bug such as double-returning resources. It reports the
issue via on_internal_error_noexcept and clamps resources back to
the initial values. However, before this commit there were no tests
that verified this behavior, so bugs like SCYLLADB-1014 went
undetected.
Add a test that artificially signals resources that were never
consumed and verifies that signal() detects the negative leak and
clamps available resources back to the initial limit.
Refs: SCYLLADB-1014
Fixes: SCYLLADB-1031
Closesscylladb/scylladb#28993
Since 7564a56dc8, all tables default to
repair-mode tombstone-gc, which is identical to immediate-mode for RF=1
tables. Consequently the tombstones written by the tests in this test
file are immediately collectible and with some unlucky timing, some of
them can be collected before the end of the test, failing the empty-page
prefix check because the empty pages prefix will be smaller than
expected based on the number of tombstones written.
Disable tombstone-gc to remove this source of flakyness.
Fixes: SCYLLADB-1062
Closesscylladb/scylladb#29077
The test has expectation w.r.t which write makes it to which nodes:
* inserts make it to all nodes
* delete makes it to all-1 (QUORUM) node
However, this was not expressed with CL, and the default CL=ONE allowed
for some nodes missing the writes and this violating the tests
expectations on what data is persent on which nodes. This resulted on
the test being flaky and failing on the data checks.
Use explicit CL for the ingestion to prevent this.
The improvements to the test introduced in
a8dd13731f was of great help in
investigating this: traces are now available and the check happens after
the data was dumped to logs.
Fixes: SCYLLADB-870
Fixes: SCYLLADB-812
Fixes: SCYLLADB-1102
Closesscylladb/scylladb#29128
Introduce an initial and experimental implementation of an alternative log-structured storage engine for key-value tables.
Main flows and components:
* The storage is composed of 32MB files, each file divided to segments of size 128k. We write to them sequentially records that contain a mutation and additional metadata. Records are written to a buffer first and then written to the active segment sequentially in 4k sized blocks.
* The primary index in memory maps keys to their location on disk. It is a B-tree per-table that is ordered by tokens, similar to a memtable.
* On reads we calculate the key and look it up in the primary index, then read the mutation from disk with a single disk IO.
* On writes we write the record to a buffer, wait for it to be written to disk, then update the index with the new location, and free the previous record.
* We track the used space in each segment. When overwriting a record, we increase the free space counter for the segment of the previous record that becomes dead. We store the segments in a histogram by usage.
* The compaction process takes segments with low utilization, reads them and writes the live records to new segments, and frees the old segments.
* Segments are initially "mixed" - we write to the active segment records from all tables and all tablets. The "separator" process rewrites records from mixed segments into new segments that are organized by compaction groups (tablets), and frees the mixed segments. Each write is written to the active segment and to a separator buffer of the compaction group, which is eventually flushed to a new segment in the compaction group.
Currently this mode is experimental and requires an experimental flag to be enabled.
Some things that are not supported yet are strong consistency, tablet migration, tablet split/merge, big mutations, tombstone gc, ttl.
to use, add to config:
```
enable_logstor: true
experimental_features:
- logstor
```
create a table:
```
CREATE TABLE ks.t(pk int PRIMARY KEY, a int, v text) WITH storage_engine = 'logstor';
```
INSERT, SELECT, DELETE work as expected
UPDATE not supported yet
no backport - new feature
Closesscylladb/scylladb#28706
* github.com:scylladb/scylladb:
logstor: trigger separator flush for buffers that hold old segments
docs/dev: add logstor documentation
logstor: recover segments into compaction groups
logstor: range read
logstor: change index to btree by token per table
logstor: move segments to replica::compaction_group
db: update dirty mem limits dynamically
logstor: track memory usage
logstor: logstor stats api
logstor: compaction buffer pool
logstor: separator: flush buffer when full
logstor: hold segment until index updates
logstor: truncate table
logstor: enable/disable compaction per table
logstor: separator buffer pool
test: logstor: add separator and compaction tests
logstor: segment and separator barrier
logstor: separator debt controller
logstor: compaction controller
logstor: recovery: recover mixed segments using separator
logstor: wait for pending reads in compaction
logstor: separator
logstor: compaction groups
logstor: cache files for read
logstor: recovery: initial
logstor: add segment generation
logstor: reserve segments for compaction
logstor: index: buckets
logstor: add buffer header
logstor: add group_id
logstor: record generation
logstor: generation utility
logstor: use RIPEMD-160 for index key
test: add test_logstor.py
api: add logstor compaction trigger endpoint
replica: add logstor to db
schema: add logstor cf property
logstor: initial commit
db: disable tablet balancing with logstor
db: add logstor experimental feature flag
Trie-based sstable indexes are supposed to be (hopefully) a better default than the old BIG indexes.
Make the new format a new default for new clusters by naming ms in the default scylla.yaml.
New functionality. No backport needed.
This PR is basically Michał's one https://github.com/scylladb/scylladb/pull/26377, Jakub's https://github.com/scylladb/scylladb/pull/27332 fixing `sstables_manager::get_highest_supported_format()` and one test fix.
Closesscylladb/scylladb#28960
* github.com:scylladb/scylladb:
db/config: announce ms format as highest supported
db/config: enable `ms` sstable format by default
cluster/dtest/bypass_cache_test: switch from highest_supported_sstable_format to chosen_sstable_format
api/system: add /system/chosen_sstable_version
test/cluster/dtest: reduce num_tokens to 16
take_snapshot return values were unused so drop them. do_refresh was a
thin wrapper around load_new_sstables that added no logic; inline it
directly into the gather expression.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The goal of test_refresh_deletes_uploaded_sstables is to verify that
sstables are removed from the upload directory after refresh. The replica
check was just a sanity guard; a simple SELECT of all keys is sufficient
and much lighter.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Replace create_dataset() with explicit keyspace creation via new_test_keyspace,
inline CREATE TABLE, and direct cql.run_async inserts — matching the pattern
used in do_test_streaming_scopes. This removes the last dependency on backup
helpers for dataset setup and makes the test self-contained.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Wrap the test body under if True: to pre-indent it, making the subsequent
patch that introduces new_test_keyspace a pure content change with no
whitespace noise.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Replace create_cluster() from object_store/test_backup.py with a plain
manager.servers_add(2) call. The test does not use object storage, so
there is no need to pull in the backup helper along with its config and
logging knobs.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This PR fixes the Installation page:
- Replaces `http `with `https `in the download command.
- Replaces the Open Source example from the Installation section for CentOS (we overlooked this example before).
Fixes https://github.com/scylladb/scylladb/issues/29087
Fixes https://github.com/scylladb/scylladb/issues/29087
This update affects all supported versions and should be backported as a bug fix.
Closesscylladb/scylladb#29088
* github.com:scylladb/scylladb:
doc: remove the Open Source Example from Installation
doc: replace http with https in the installation instructions
The default 'cassandra' superuser was removed from ScyllaDB, which
broke PGO training. exec_cql.py relied on username/password auth
('cassandra'/'cassandra') to execute setup CQL scripts like auth.cql
and counters.cql.
Switch exec_cql.py to connect via the Unix domain maintenance socket
instead. The maintenance socket bypasses authentication, no credentials
are needed. Additionally, create the 'cassandra' superuser via the
maintenance socket during the populate phase, so that cassandra-stress
keeps working. cassandra-stress hardcodes user=cassandra password=cassandra.
Changes:
- exec_cql.py: replace host/port/username/password arguments with a
single --socket argument; add connect_maintenance_socket() with
wait ready logic
- pgo.py: add maintenance_socket_path() helper; update
populate_auth_conns() and populate_counters() to pass the socket
path to exec_cql.py
Fixes SCYLLADB-1070
Closesscylladb/scylladb#29081
Add tests for RF guardrails (min/max warn/fail, RF=0 bypass,
threshold=-1 disable, ALTER KEYSPACE) and write consistency level
guardrails to cover all scenarios described in guardrails.rst.
Test runtime (dev):
test_guardrail_replication_strategy - 6s
test_guardrail_write_consistency_level - 5s
Refs: SCYLLADB-257
Replace loose substring assertions with regex-based matching against
the exact server message formats. Add regex constants for all
guardrail messages and rewrite create_ks_and_assert_warnings_and_errors()
to verify count and content of warnings and failures.
Refs: SCYLLADB-257
The Web Installer page includes instructions to install the old pre-2025.1 Enterprise versions,
which are no longer supported (since we released 2026.1).
This commit removes those redundant and misleading instructions.
Fixes https://github.com/scylladb/scylladb/issues/29099Closesscylladb/scylladb#29103
This patchset migrates: query_all_directly_granted, query_all,
get_attribute, query_attribute_for_all functions to use cache
instead of doing CQL queries. It also includes some preparatory
work which fixes cache update order and triggering.
Main motivation behind this is to make sure that all calls
from service_level_controller::auth_integration are cached,
which we achieve here.
Alternative implementation could move the whole auth_integration
data into auth cache but since auth_integration manages also lifetime
and contains service levels specific logic such solution would be
too complex for little (if any) gain.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-159
Backport: no, not a bug
Closesscylladb/scylladb#28791
* github.com:scylladb/scylladb:
auth: switch query_attribute_for_all to use cache
auth: switch get_attribute to use cache
auth: cache: add heterogeneous map lookups
auth: switch query_all to use cache
auth: switch query_all_directly_granted to use cache
auth: cache: add ability to go over all roles
raft: service: reload auth cache before service levels
service: raft: move update_service_levels_effective_cache check
`storage_group_of()` sits on the replica-side token lookup hot path, yet it called `tablet_map::get_tablet_id_and_range_side()`, which always computes both the tablet id and the post-split range side — even though most callers only need the storage group id.
The range-side computation is only relevant when a storage group is in tablet splitting mode, but we were paying for it unconditionally on every lookup.
This series fixes that by:
1. Adding `tablet_map::get_tablet_range_side()` so the range side can be computed independently when needed.
2. Adding lazy `select_compaction_group()` overloads that defer the range-side computation until splitting mode is actually active.
3. Switching `storage_group_of()` to use the cheaper `get_tablet_id()` path, only computing the range side on demand.
Improvements. No backport is required.
Closesscylladb/scylladb#28963
* github.com:scylladb/scylladb:
replica/table: avoid computing token range side in storage_group_of() on hot path
replica/compaction_group: add lazy select_compaction_group() overloads
locator/tablets: add tablet_map::get_tablet_range_side()
Factor out ks_opts() to build keyspace options with tablets handling
and use it across all existing replication strategy guardrail tests.
No behavioral changes.
This facilitates further modification of the tests later in this
patch series.
Refs: SCYLLADB-257
After repair, the test does a major to compact all sstables into a
single one, so the results can be simply checked by a select from
mutation_fragments() query. Sometimes off-strategy happens parallel to
this major, so after the major there are still 2 sstables, resulting in
the test failing when checking that the query returns just a single row.
To fix, just use tablets for the test table, tablets don't use
off-strategy anymore.
Fixes: SCYLLADB-940
Closesscylladb/scylladb#29071
Previously the test test_interrupt_view_build_shard_registration stopped
the node ungracefully and used commitlog periodic mode to persist the
view build progress in a not very reliable way.
It can happen that due to timing issues, the view build progress is not
persisted, or some of it is persisted in a different ordering than
expected.
To make the test more reliable we change it to stop the node gracefully,
so the commitlog is persisted in a graceful and consistent way, without
using the periodic mode delay. We need to also change the injection for
the shutdown to not get stuck.
Fixes SCYLLADB-1005
Closesscylladb/scylladb#29008
As reported in SCYLLADB-1013, the directory lister must be closed also when an exception is thrown.
For example, see backtrace below:
```
seastar::on_internal_error(seastar::logger&, std::basic_string_view<char, std::char_traits<char>>) at ./build/release/seastar/./seastar/src/core/on_internal_error.cc:57
directory_lister::~directory_lister() at ./utils/lister.cc:77
replica::table::get_snapshot_details(std::filesystem::__cxx11::path, std::filesystem::__cxx11::path) (.resume) at ./replica/table.cc:4081
std::__n4861::coroutine_handle<seastar::internal::coroutine_traits_base<db::snapshot_ctl::table_snapshot_details>::promise_type>::resume() const at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/coroutine:247
(inlined by) seastar::internal::coroutine_traits_base<db::snapshot_ctl::table_snapshot_details>::promise_type::run_and_dispose() at ././seastar/include/seastar/core/coroutine.hh:129
seastar::reactor::task_queue::run_tasks() at ./build/release/seastar/./seastar/src/core/reactor.cc:2695
(inlined by) seastar::reactor::task_queue_group::run_tasks() at ./build/release/seastar/./seastar/src/core/reactor.cc:3201
seastar::reactor::task_queue_group::run_some_tasks() at ./build/release/seastar/./seastar/src/core/reactor.cc:3185
(inlined by) seastar::reactor::do_run() at ./build/release/seastar/./seastar/src/core/reactor.cc:3353
seastar::reactor::run() at ./build/release/seastar/./seastar/src/core/reactor.cc:3245
seastar::app_template::run_deprecated(int, char**, std::function<void ()>&&) at ./build/release/seastar/./seastar/src/core/app-template.cc:266
seastar::app_template::run(int, char**, std::function<seastar::future<int> ()>&&) at ./build/release/seastar/./seastar/src/core/app-template.cc:160
scylla_main(int, char**) at ./main.cc:756
```
Fixes: [SCYLLADB-1013](https://scylladb.atlassian.net/browse/SCYLLADB-1013)
* Requires backport to 2026.1 since the leak exists since 004c08f525
[SCYLLADB-1013]: https://scylladb.atlassian.net/browse/SCYLLADB-1013?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQClosesscylladb/scylladb#29084
* github.com:scylladb/scylladb:
test/boost/database_test: add test_snapshot_ctl_details_exception_handling
table: get_snapshot_details: fix indentation inside try block
table: per-snapshot get_snapshot_details: fix typo in comment
table: per-snapshot get_snapshot_details: always close lister using try/catch
table: get_snapshot_details: always close lister using deferred_close
Run keyspace compaction asynchronously in
`test_tombstone_gc_correctness_during_tablet_split` and only await it
after `split_sstable_rewrite` is disabled.
The problem is that `keyspace_compaction()` starts with a flush, and that
flush can take around five seconds. During that window the split
compaction is stopped before major compaction is retried. The stop aborts
the in-flight major compaction attempt, then the split proceeds far enough
to enter the `split_sstable_rewrite` injection point.
At that point the test used to wait synchronously for major compaction to
finish, but major compaction cannot finish yet: when it retries, it needs
the same semaphore that is still effectively tied up behind the blocked
split rewrite. So the test waits for major compaction, while the split
waits for the injection to be released, and the code that would release
that injection never runs.
Starting major compaction as a task breaks that cycle. The test can first
disable `split_sstable_rewrite`, let the split get out of the way, and
only then wait for major compaction to complete.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-827.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Closesscylladb/scylladb#29066
During decommission, we first mark a topology request as done, then shut
down a node and in the following steps we remove node from the topology.
Thus, finished request does not imply that a node is removed from
the topology.
Due to that, in node_ops_virtual_task::wait, while gathering children
from the whole cluster, we may hit the connection exception - because
a node is still in topology, even though it is down.
Modify the get_children method to ignore the exception and warn
about the failure instead.
Keep token_metadata_ptr in get_children to prevent topology from changing.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-867
Needs backports to all versions
Closesscylladb/scylladb#29035
* github.com:scylladb/scylladb:
tasks: fix indentation
tasks: do not fail the wait request if rpc fails
tasks: pass token_metadata_ptr to task_manager::virtual_task::impl::get_children
Document that `SELECT ... WHERE` clause currently accepts only conjunctions
of relations joined by `AND` (`OR` is not supported), and that
parentheses cannot be used to group boolean subexpressions.
Add an unsupported query example and point readers to equivalent `IN`
rewrites when applicable.
This problem has been raised by one of our users in
https://forum.scylladb.com/t/error-parsing-query-or-unsupported-statement/5299,
and while one could infer answer to user's question by looking at the
syntax of the `SELECT ... WHERE`, it's not immediately obvious to
non-advanced users, so clarifying these concepts is justified.
Fixes: SCYLLADB-1116
Closesscylladb/scylladb#29100
A compaction group has a separator buffer that holds the mixed segments
alive until the separator buffer is flushed. A mixed segment can be
freed only after all separator buffers that hold writes from the segment
are flushed.
Typically a separator buffer is flushed when it becomes full. However
it's possible for example that one compaction groups is filled slower
than others and holds many segments.
To fix this we trigger a separator flush periodically for separator
buffers that hold old segments. We track the active segment sequence
number and for each separator buffer the oldest sequence number it
holds.
Fix the logstor recovery to work with compaction groups. When recovering
a segment find its token range and add it to the appropriate compaction
groups. if it doesn't fit in a single compaction group then write each
record to its compaction group's separator buffer.
Change the primary index to be a btree that is ordered by token,
similarly to a memtable, and create a index per-table instead of a
single global index.
Add a segment_set member to replica::compaction_group that manages the
logstor segments that belong to the compaction group, similarly to how
it manages sstables. Add also a separator buffer in each compaction
group.
When writing a mutation to a compaction group, the mutation is written
to the active segment and to the separator buffer of the compaction
group, and when the separator buffer is flushed the segment is added to
the compaction_group's segment set.
when logstor is enabled, update the db dirty memory limits dynamically.
previously the threshold is set to 0.5 of the available memory, so 0.5
goes to memtables and 0.5 to others (cache).
when logstor is enabled, we calculate the available memory excluding
logstor, and divide it evenly between memtables and cache.
add a write gate to write_buffer. when writing a record to the write
buffer, the gate is held and passed back to the caller, and the caller
holds the gate until the write operation is complete, including
follow-up operations such as updating the index after the write.
in particular, when writing a mutation in logstor::write, the write
buffer is held open until the write is completed and updated in the
index.
when writing the write buffer to the active segment, we write the buffer
and then wait for the write buffer gate to close, i.e. we wait for all
index updates to complete before proceeding. the segment is held open
until all the write operations and index updates are complete.
this property is useful for correctness: when a segment is closed we
know that all the writes to it are updated in the index. this is needed
in compaction for example, where we take closed segments and check
which records in them are alive by looking them up in the index. if the
index is not updated yet then it will be wrong.
implement freeing all segments of a table for table truncate.
first do barrier to flush all active and mixed segments and put all the
table's data in compaction groups, then stop compaction for the table,
then free the table's segments and remove the live entries from the
index.
add barrier operation that forces switch of the active segment and
separator, and waits for all existing segments to close and all
separators to flush.
add tracking of the total separator debt - writes that were written to a
separator and waiting to be flushed, and add flow control to keep the
debt in control by delaying normal writes.
on recovery we may find mixed segments. recover them by adding them to a
separator, reading all their records and writing them to the separator,
and flush the separator.
we free a segment from compaction after updating all live records in the
segment to point to new locations in the index. we need to ensure they
are no running operations that use the old locations before we free the
segment.
initial implementation of the separator. it replaces "mixed" segments -
segments that have records from different groups, to segments by group.
every write is written to the active segment and to a buffer in the
active separator. the active separator has in-memory buffers by group.
at some threshold number of segments we switch the active segment and
separator atomically, and start flushing the separator.
the separator is flushed by writing the buffers into new non-mixed
segments, adding them to a compaction group, and frees the mixed
segments.
initial and basic recovery implementation.
* find all files, read their segments and populate the index with the
newest record for each key.
* find which segments are used and build the usage histogram
add segment generation number that is incremented when the segment is
reused, and it's written to every buffer that is written to the segment.
this is useful for recovery.
reserve segments for compaction so it always has enough segments to run
and doesn't get stuck.
do the compaction writes into full new segments instead of the active
segment.
add group_id value to each log record that is passed with the mutation
when writing it.
the group_id will be used to group log records in segments, such that a
segment will contain records only from a single group.
this will be useful for tablet migration. we want for each tablet to
have their own segments with all their records, so we can migrate them
efficiently by copying these segments.
the group_id value is set to a value equivalent to the tablet id.
basic utility for generation numbers that will be useful next. a
generation number is an unsigned integer that can be incremented and
compared even if it wraparounds, assuming the values we compare were
written around the same time.
initial implementation of the logstor storage engine for key-value
tables that supports writes, reads and basic compaction.
main components:
* logstor: this is the main interface to users that supports writing and
reading back mutations, and manages the internal components.
* index: the primary index in-memory that maps a key to a location on
disk.
* write buffer: writes go initially to a write buffer. it accumulates
multiple records in a buffer and writes them to the segment manager in
4k sized blocks.
* segment manager: manages the storage - files, segments, compaction. it
manages file and segment allocation, and writes 4k aligned buffers to
the active segment sequentially. it tracks the used space in each
segment. the compaction finds segment with low space usage and writes
them to new segments, and frees the old segments.
GRANT/REVOKE fails on the maintenance socket connections, because maintenance_auth_service uses allow_all_authorizer. allow_all_authorizer allows all operations, but not GRANT/REVOKE, because they make no sense in its context.
This has been observed during PGO run failure in operations from ./pgo/conf/auth.cql file.
This patch introduces maintenance_socket_authorizer that supports the capabilities of default_authorizer ('CassandraAuthorizer') without needing authorization.
Refs SCYLLADB-1070
This is an improvement, no need for backport.
Closesscylladb/scylladb#29080
* github.com:scylladb/scylladb:
test: use NetworkTopologyStrategy in maintenance socket tests
test: use cleanup fixture in maintenance socket auth tests
auth: add maintenance_socket_authorizer
Make the pattern static const so it is compiled once at first call rather
than on every Content-Range header parse.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29054
It handles 0, and could generate better code for that. On Broadwell
architecture, it translates to a single instruction (LZCNT). We're
still on Westmere, so it translates to BSR with a conditional move.
Also, drop unnecessary casts and bit arithmetic, which saves a few
instructions.
Move to header so that it's inlined in parsers.
This change reduces the cost of partition index page construction and
LSA migration. This is achieved by several things working together:
- index entries don't store keys as separate small objects (managed_bytes)
They are written into one managed_bytes fragmented storage, entries
hold offset into it.
Before, we paid 16 bytes for managed_bytes plus LSA descriptor for
the storage (1 byte) plus back-reference in the storage (8 bytes),
so 25 bytes. Now we only pay 4 bytes for the size offset. If keys are 16
bytes, that's a reduction from 31 bytes to 20 bytes per key.
- index entries and key storage are now trivially moveable, so LSA
migration can use memcpy() which amortizes the cost per key.
memcpy().
LSA eviction is now trivial and constant time for the whole page
regardless of the number of entries. Page eviction dropped from
14 us to 1 us.
This improves throughput in a CPU-bound miss-heavy read workload where
the partition index doesn't fit in memory.
scylla perf-simple-query -c1 -m200M --partitions=1000000
Before:
15328.25 tps (150.0 allocs/op, 14.1 logallocs/op, 45.4 tasks/op, 286769 insns/op, 218134 cycles/op, 0 errors)
15279.01 tps (149.9 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 287696 insns/op, 218637 cycles/op, 0 errors)
15347.78 tps (149.7 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 285851 insns/op, 217795 cycles/op, 0 errors)
15403.68 tps (149.6 allocs/op, 14.1 logallocs/op, 45.2 tasks/op, 285111 insns/op, 216984 cycles/op, 0 errors)
15189.47 tps (150.0 allocs/op, 14.1 logallocs/op, 45.5 tasks/op, 289509 insns/op, 219602 cycles/op, 0 errors)
15295.04 tps (149.8 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 288021 insns/op, 218545 cycles/op, 0 errors)
15162.01 tps (149.8 allocs/op, 14.1 logallocs/op, 45.4 tasks/op, 291265 insns/op, 220451 cycles/op, 0 errors)
After:
21620.18 tps (148.4 allocs/op, 13.4 logallocs/op, 43.7 tasks/op, 176817 insns/op, 153183 cycles/op, 0 errors)
20644.03 tps (149.8 allocs/op, 13.5 logallocs/op, 44.3 tasks/op, 187941 insns/op, 160409 cycles/op, 0 errors)
20588.06 tps (150.1 allocs/op, 13.5 logallocs/op, 44.5 tasks/op, 188090 insns/op, 160818 cycles/op, 0 errors)
20789.29 tps (149.5 allocs/op, 13.5 logallocs/op, 44.2 tasks/op, 186495 insns/op, 159382 cycles/op, 0 errors)
20977.89 tps (149.5 allocs/op, 13.4 logallocs/op, 44.2 tasks/op, 183969 insns/op, 158140 cycles/op, 0 errors)
21125.34 tps (149.1 allocs/op, 13.4 logallocs/op, 44.1 tasks/op, 183204 insns/op, 156925 cycles/op, 0 errors)
21244.42 tps (148.6 allocs/op, 13.4 logallocs/op, 43.8 tasks/op, 181276 insns/op, 155973 cycles/op, 0 errors)
Mostly because the index now fits in memory.
When it doesn't, the benefits are still visible due to lower LSA overhead.
It's shorter, and is supposed to be optimized for trivially-moveable
types.
Important for managed_vector<index_entry>, which can have lots of
elements.
Densely populated pages have no promoted index (small partitions), so
we can save space in such workloads by keeping promoted index in a
separate vector.
For workloads which do have a promoted index, pages have only one
partition. There aren't many such pages and they are long-lived, so
the extra allocation of the vector is amortized.
promoted_index class is removed, and replaced with equivalent
parsed_promoted_index_entry for simplicity. Because it's removed,
make_cursor() is moved into the index_reader class.
Reducing the size of index_entry is important for performence if pages
are densly populated. It helps to reduce LSA allocator pressure and
compaction/eviction speed.
This change, combined with the earlier change "Shave-off 16 bytes from
index_entry by using raw_token", gives significant improvement in
throughput in perf_simple_query run where the index doesn't fit in
memory:
scylla perf-simple-query -c1 -m200M --partitions=1000000
Before:
9714.78 tps (170.9 allocs/op, 16.9 logallocs/op, 55.3 tasks/op, 494788 insns/op, 343920 cycles/op, 0 errors)
9603.13 tps (171.6 allocs/op, 17.0 logallocs/op, 55.6 tasks/op, 502358 insns/op, 348344 cycles/op, 0 errors)
9621.43 tps (171.9 allocs/op, 17.0 logallocs/op, 55.8 tasks/op, 500612 insns/op, 347508 cycles/op, 0 errors)
9597.75 tps (171.6 allocs/op, 17.0 logallocs/op, 55.6 tasks/op, 501428 insns/op, 348604 cycles/op, 0 errors)
9615.54 tps (171.6 allocs/op, 16.9 logallocs/op, 55.6 tasks/op, 501313 insns/op, 347935 cycles/op, 0 errors)
9577.03 tps (171.8 allocs/op, 17.0 logallocs/op, 55.7 tasks/op, 503283 insns/op, 349251 cycles/op, 0 errors)
After:
15328.25 tps (150.0 allocs/op, 14.1 logallocs/op, 45.4 tasks/op, 286769 insns/op, 218134 cycles/op, 0 errors)
15279.01 tps (149.9 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 287696 insns/op, 218637 cycles/op, 0 errors)
15347.78 tps (149.7 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 285851 insns/op, 217795 cycles/op, 0 errors)
15403.68 tps (149.6 allocs/op, 14.1 logallocs/op, 45.2 tasks/op, 285111 insns/op, 216984 cycles/op, 0 errors)
15189.47 tps (150.0 allocs/op, 14.1 logallocs/op, 45.5 tasks/op, 289509 insns/op, 219602 cycles/op, 0 errors)
15295.04 tps (149.8 allocs/op, 14.1 logallocs/op, 45.3 tasks/op, 288021 insns/op, 218545 cycles/op, 0 errors)
15162.01 tps (149.8 allocs/op, 14.1 logallocs/op, 45.4 tasks/op, 291265 insns/op, 220451 cycles/op, 0 errors)
The std::optional<> adds 8 bytes.
And dht::token adds 8 bytes due to _kind, which in this case is always
kind::key.
The size changd from 56 to 48 bytes.
During decommission, we first mark a topology request as done, then shut
down a node and in the following steps we remove node from the topology.
Thus, finished request does not imply that a node is removed from
the topology.
Due to that, in node_ops_virtual_task::wait, while gathering children
from the whole cluster, we may hit the connection exception - because
a node is still in topology, even though it is down.
Modify the get_children method to ignore the exception and warn
about the failure instead.
In get_children we get the vector of alive nodes with get_nodes.
Yet, between this and sending rpc to those nodes there might be
a preemption. Currently, the liveness of a node is checked once
again before the rpcs (only with gossiper not in topology - unlike
get_nodes).
Modify get_children, so that it keeps a token_metadata_ptr,
preventing topology from changing between get_nodes and rpcs.
Remove test_get_children as it checked if the get_children method
won't fail if a node is down after get_nodes - which cannot happen
currently.
Fixes: SCYLLADB-942
Adds an injection signal _from_ table::seal_active_memtable to allow us to
reliably wait for flushing. And does so.
Closesscylladb/scylladb#29070
Since commit 509f2af8db, gate_closed_exception can be triggered for ongoing split during shutdown. The commit is correct, but it causes split failure on shutdown to log an error, which causes CI instability. Previously, aborted_exception would be triggered instead which is logged as warning. Let's do the same.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-951.
Fixes https://github.com/scylladb/scylladb/issues/24850.
Only 2026.1 is affected.
Closesscylladb/scylladb#29032
* github.com:scylladb/scylladb:
replica: Demote log level on split failure during shutdown
service: Demote log level on split failure during shutdown
The limiter scans ranges to decide whether or not to rate-limit the
query. However, when considering each range only the front one's token
is accounted. This looks like a misprint.
The limiter was introduced in cc9a2ad41f
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closesscylladb/scylladb#29050
A followup of the merge of two test cases that happened in the previous
patch. Both used `foo = N if domain == bar else M` to evaluate the
parameters for topology. Using if-else block makes it immediately obvious
which topology and scope apply for each domain value without having to
evaluate multiple inline conditionals.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The two tests differ only in the way they set up the topology for the
cluster and the post-restore checks against the resulting streams.
The merge happens with the help of a "scope_is_same" boolean parameter
and corresponding updates in the topology setup and post-checks.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The one in "different domain" test is simpler because the test performs
less checks. Next patch will merge both tests and making regexp-s look
identical makes the merge even smother.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Update test running instructions to reflect unified pytest-based runner.
The test.py now requires full test paths with file extensions for both
C++ and Python tests.
No backport: The change is only relevant for recent test.py changes in
master.
Closesscylladb/scylladb#29062
The auth::cache::includes_table function also covers role_members and
role_attributes. The existing check was removed because it blocked these
tables from triggering necessary cache updates.
While previously non-critical (due to unused attributes and table coupling),
maintaining a correct cache is essential for upcoming changes.
Verify that the directory listers opened by get_snapshot_details
are properly closed when handling an (injected) exception.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
The comment says the snapshot directory may contain a `schema.sql` file,
but the code treats `schema.cql` as the special-case schema file.
Reported-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Since this is a coroutine, we cannot just use deferred_close,
but rather we need to catch an error, close the lister, and then
return the error, is applicable.
Fixes: SCYLLADB-1013
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
NetworkTopologyStrategy is the preferred choice. We should not
use SimpleStrategy anymore. This patch changes the topology strategy
for all the maintenance socket tests.
Refs SCYLLADB-1070
Add a cql_clusters pytest fixture that tracks CQL driver Cluster
objects and shuts them down automatically after test completion.
This replaces manual shutdown() calls at the end of each test.
Also consolidate shutdown() calls in retry helpers into finally
blocks for consistent cleanup.
Refs SCYLLADB-1070
GRANT/REVOKE fails on the maintenance socket connections,
because maintenance_auth_service uses allow_all_authorizer.
allow_all_authorizer allows all operations, but not GRANT/REVOKE,
because they make no sense in its context.
This has been observed during PGO run failure in operations from
./pgo/conf/auth.cql file.
This patch introduces maintenance_socket_authorizer that supports
the capabilities of default_authorizer ('CassandraAuthorizer')
without needing authorization.
Refs SCYLLADB-1070
This patch is mostly for the purpose of running pgo CI job.
We may receive connection error if asyncio.sleep(5) in
pgo.py is not sufficient waiting time.
In pgo.py we do wait for port but only for cql,
anyway it's better to have high level check than
trying to wait for alternator port there.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-1071
Backport: 2026.1 - it failed on CI for that build
Closesscylladb/scylladb#29063
* github.com:scylladb/scylladb:
perf: add abort_source support to wait-for-port loops
perf-alternator: wait for alternator port before running workload
- fix s3::range max value for object size which is 50TiB and not 5.
- refactor constants to make it accessible for all interested parties, also reuse these constants in tests
No need to backport, doubt we will encounter an object larger than 5TiB
Closesscylladb/scylladb#28601
* github.com:scylladb/scylladb:
s3_client: reorganize tests in part_size_calculation_test
s3_client: switch using s3 limits constants in tests
s3_client: fix the s3::range max object size
s3_client: remove "aws" prefix from object limits constants
s3_client: make s3 object limits accessible
This commit replaces the Open Soruce example from the Installation section for CentOS.
We updated the example for Ubuntu, but not for CentOS.
We don't want to have any Open Source information in the docs.
Fixes https://github.com/scylladb/scylladb/issues/29087
This test was observed to fail in CI recently but there is not enough information in the logs to figure out what went wrong. This PR makes a few improvements to make the next investigation easier, should it be needed:
* storage-service: add table name to mutation write failure error messages.
* database: the `database_apply` error injection used to cause trouble, catching writes to bystander tables, making tests flaky. To eliminate this, it gained a filter to apply only to non-system keyspaces. Unfortunately, this still allows it to catch writes to the trace tables. While this should not fail the test, it reduces observability, as some traces disappear. Improve this error injection to only apply to selected table. Also merge it with the `database_apply_wait` error injection, to streamline the code a bit.
* test/test_data_resurrection_in_memtable.py: dump data from the datable, before the checks for expected data, so if checks fail, the data in the table is known.
Refs: SCYLLADB-812
Refs: SCYLLADB-870
Fixes: SCYLLADB-1050 (by restricting `database_apply` error injection, so it doesn't affect writes to system traces)
Backport: test related improvement, no backport
Closesscylladb/scylladb#28899
* github.com:scylladb/scylladb:
test/cluster/test_data_resurrection_in_memtable.py: dump rows before check
replica/database: consolidate the two database_apply error injections
service/storage_proxy: add name of table to error message for write errors
Previously, all stream-table fixtures in test_streams.py used scope="function",
forcing a fresh table to be created for every test, slowing down the test a bit
(though not much), and discouraging writing small new tests.
This was a workaround for a DynamoDB quirk (that Alternator doesn't have):
LATEST shard iterators have a time slack and may point slightly before the true
stream head, causing leftover events from a previous test to appear in the next
test's reads.
The first two tests in this series fix small problems that turn up once we start
sharing test tables in test_streams.py. The final patch fixes the "LATEST" problem
and enables sharing the test table by using "module" scope fixtures instead of
"function".
After this series, test_streams.py run time went down a bit, from 20.2 seconds to 17.7 seconds.
Closesscylladb/scylladb#28972
* github.com:scylladb/scylladb:
test/alternator: speed up test_streams.py by using module-scope fixtures
test/alternator: test_streams.py don't use fixtures in 4 tests
test/alternator: fix do_test() in test_streams.py
This test relies on the cache entry being evicted after 200ms past the
TTL. This may not happen on a busy CI machine. Make the test less
reliant on timing by using eventually_true().
Simplify the test by dropping the second entry, it doesn't add anything
to the test.
Fixes: SCYLLADB-811
Closesscylladb/scylladb#28958
This fixtures starts the mock server and immediately connects to it to
setup the expected requests. The connection attempt might be too early,
so there is a retry loop with a timeout. The loop currently checks for
requests.exception.ConnectionError. We've seen a case where the
connection is successful but the request fails with 404. The mock
started the server but didn't setup the routes yet. Add a retry for http
404 to handle this.
Fixes: SCYLLADB-966
Closesscylladb/scylladb#29003
When handling `repair_stream_cmd::end_of_current_rows`, passing the
foreign list directly to `put_row_diff_handler` triggered a massive
synchronous deep copy on the destination shard. Additionally, destroying
the list triggered a synchronous deallocation on the source shard. This
blocked the reactor and triggered the CPU stall detector.
This commit fixes the issue by introducing `clone_gently()` to copy the
list elements one by one, and leveraging the existing
`utils::clear_gently()` to destroy them. Both utilize
`seastar::coroutine::maybe_yield()` to allow the reactor to breathe
during large cross-shard transfers and cleanups.
Fixes SCYLLADB-403
Closesscylladb/scylladb#28979
vector
The lister loop in get() pre-fetches records in batches and keeps them
in a _info vector, iterating over it with the help of _pos cursor. When
the vector is re-read, the cursor must be reset too.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The lister loop in get() method looks weird. It uses do-while(false)
loop and calls continue; inside when filter asks to skip a entry.
Skipping, thus, aborts the whole thing and EOF-s, which is not what's
supposed to happen.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
This series fixes a metrics visibility gap in Alternator and adds regression coverage.
Until now, BatchGetItem and BatchWriteItem updated global latency histograms but did not consistently update per-table latency histograms. As a result, table-level latency dashboards could miss batch traffic.
It updates the batch read/write paths to compute request duration once and record it in both global and per-table latency metrics.
Add the missing tests, including a metric-agnostic helper and a dedicated per-table latency test that verifies latency counters increase for item and batch operations.
This change is metrics-only (no API/behavior change for requests) and improves observability consistency between global and per-table views.
Fixes#28721
**We assume the alternator per-table metrics exist, but the batch ones are not updated**
Closesscylladb/scylladb#28732
* github.com:scylladb/scylladb:
test(alternator): add per-table latency coverage for item and batch ops
alternator: track per-table latency for batch get/write operations
Remove outdated references to filtering on columns provided in the
index definition, and remove the note about equal relations (= and IN)
being the only supported operations. Vector search filtering currently
supports WHERE clauses on primary key columns only.
Closesscylladb/scylladb#28949
Permits in the `waiting_for_memory` state represent already-executing reads that are blocked on memory allocation. Preemptively aborting them is wasteful -- these reads have already consumed resources and made progress, so they should be allowed to complete.
Restrict the preemptive abort check in maybe_admit_waiters() to only apply to permits in the `waiting_for_admission` state, and tighten the state validation in `on_preemptive_aborted()` accordingly.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-1016
Backport not needed. The commit introducing replica load shedding is not part of 2026.1
Closesscylladb/scylladb#29025
* github.com:scylladb/scylladb:
reader_concurrency_semaphore: skip preemptive abort for permits waiting for memory
reader_concurrency_semaphore_test: detect memory leak on preemptive abort of waiting_for_memory permit
Check abort_source on each retry iteration in
wait_for_alternator and wait_for_cql so the
wait can be interrupted on shutdown.
Didn't use sleep_abortable as the sleep is very short
anyway.
Fixes: SCYLLADB-244
Disables snapshot control such that any active ops finish/fail
before proceeding with decommission.
Note: snapshot control provided as argument, not member ref
due to storage_service being used from both main and cql_test_env.
(The latter has no snapshot_ctl to provide).
Could do the snapshot lockout on API level, but want to do
pre-checks before this.
Note: this just disables backup/snapshot fully. Could re-enable
after decommission, but this seems somewhat pointless.
v2:
* Add log message to snapshot shutdown
* Make test use log waiting instead of timeouts
Closesscylladb/scylladb#28980
This patch is mostly for the purpose of running pgo CI job.
We may receive connection error if asyncio.sleep(5) in
pgo.py is not sufficient waiting time.
In pgo.py we do wait for port but only for cql,
anyway it's better to have high level check than
trying to wait for alternator port there.
Some tests, when create a cluster, configure nodes with the rf-rack-valid option, because sometimes they want to have it OFF. For that the option is explicitly carried around, but the cluster creating helper can guess this option itself -- out of the provided topology and replication factor.
Removing this option simplifies the code and (which a nicer outcome) the test "signature" that's used e.g. in command-line to run a specific test.
Improving tests, not backporting
Closesscylladb/scylladb#28860
* github.com:scylladb/scylladb:
test: Relax topology_rf_validity parameter for some tests
test: Auto detect rf-rack-valid option in create_cluster()
Dtest failed with:
table - Failed to load SSTable .../me-3gyn_0qwi_313gw2n2y90v2j4fcv-big-Data.db
of origin memtable due to std::runtime_error (Cannot split
.../me-3gyn_0qwi_313gw2n2y90v2j4fcv-big-Data.db because manager has compaction
disabled, reason might be out of space prevention), it will be unlinked...
The reason is that the error above is being triggered when the cause is
shutdown, not out of space prevention. Let's distinguish between the two
cases and log the error with warning level on shutdown.
Fixes https://github.com/scylladb/scylladb/issues/24850.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
`test_raft_no_quorum.py::test_cannot_add_new_node` is currently flaky in dev
mode. The bootstrap of the first node can fail due to `add_entry()` timing
out (with the 1s timeout set by the test case).
Other test cases in this test file could fail in the same way as well, so we
need a general fix. We don't want to increase the timeout in dev mode, as it
would slow down the test. The solution is to keep the timeout unchanged, but
set it only after quorum is lost. This prevents unexpected timeouts of group0
operations with almost no impact on the test running time.
A note about the new `update_group0_raft_op_timeout` function: waiting for
the log seems to be necessary only for
`test_quorum_lost_during_node_join_response_handler`, but let's do it
for all test cases just in case (including `test_can_restart` that shouldn't
be flaky currently).
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-913Closesscylladb/scylladb#28998
Since commit 509f2af8db, gate_closed_exception can be triggered
for ongoing split during shutdown. The commit is correct, but it
causes split failure on shutdown to log an error, which causes
CI instability. Previously, aborted_exception would be triggered
instead which is logged as warning. Let's do the same.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-951.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Replaced multiple per-action workflow jobs with a single consolidated
call to main_pr_events_jira_sync.yml. Added 'edited' event trigger.
This makes CI actions in PRs more readable and workflow execution faster.
Fixes:PM-253
Closesscylladb/scylladb#29042
changes in this commit:
1)rename class from 'TestContext' to 'Context' so pytest will not consider this class as a test
2)extend pytest filterwarnings list to ignore warnings from external libs
3) use datetime.datetime.now(datetime.UTC) unstead datetime.datetime.utcnow()
4) use ResultSet.one() instead ResultSet[0]
Fixes SCYLLADB-904
Fixes SCYLLADB-908
Related SCYLLADB-902
Closesscylladb/scylladb#28956
The test assumes that the sleep duration will be at least the value of
the sleep parameter. However, the actual sleep time can be slightly less
than requested (e.g., a 100ms sleep request might result in a 99ms
sleep).
This commit adjusts the test's time comparison to be more lenient,
preventing test flakiness.
When a `with_connect` operation timed out, the underlying connection
attempt continued to run in the reactor. This could lead to a crash
if the connection was established/rejected after the client object had
already been destroyed. This issue was observed during the teardown
phase of a upcoming high-availability test case.
This commit fixes the race condition by ensuring the connection attempt
is properly canceled on timeout.
Additionally, the explicit TLS handshake previously forced during the
connection is now deferred to the first I/O operation, which is the
default and preferred behavior.
Fixes: SCYLLADB-832
In this series we add support for forwarding strongly consistent CQL requests to suitable replicas, so that clients can issue reads/writes to any node and have the request executed on an appropriate tablet replica (and, for writes, on the Raft leader). We return the same CQL response as what the user would get while sending the request to the correct replica and we perform the same logging/stats updates on the request coordinator as if the coordinator was the appropriate replica.
The core mechanism of forwarding a strongly consistent request is sending an RPC containing the user's cql request frame to the appropriate replica and returning back a ready, serialized `cql_transport::response`. We do this in the CQL server - it is most prepared for handling these types and forwarding a request containing a CQL frame allows us to reuse near-top-level methods for CQL request handling in the new RPC handler (such as the general `process`)
For sending the RPC, the CQL server needs to obtain the information about who should it forward the request to. This requires knowledge about the tablet raft group members and leader. We obtain this information during the execution of a `cql3/strong_consistency` statement, and we return this information back to the CQL server using the generalized `bounce_to_shard` `response_message`, where we now store the information about either a shard, or a specific replica to which we should forward to. Similarly to `bounce_to_shard`, we need to handle this `result_message` in a loop - a replica may move during statement execution, or the Raft leader can change. We also use it for forwarding strongly consistent writes when we're not a member of the affected tablet raft group - in that case we need to forward the statement twice - once to any replica of the affected tablet, then that replica can find the leader and return this information to the coordinator, which allows the second request to be directed to the leader.
This feature also allows passing through exception messages which happened on the target replica while executing the statement. For that, many methods of the `cql_transport::cql_server::connection` for creating error responses needed to be moved to `cql_transport::cql_server`. And for final exception handling on the coordinator, we added additional error info to the RPC response, so that the handling can be performed without having the `result_message::exception` or `exception_ptr` itself.
Fixes [SCYLLADB-71](https://scylladb.atlassian.net/browse/SCYLLADB-71)
[SCYLLADB-71]: https://scylladb.atlassian.net/browse/SCYLLADB-71?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQClosesscylladb/scylladb#27517
* github.com:scylladb/scylladb:
test: add tests for CQL forwarding
transport: enable CQL forwarding for strong consistency statements
transport: add remote statement preparation for CQL forwarding
transport: handle redirect responses in CQL forwarding
transport: add exception handling for forwarded CQL requests
transport: add basic CQL request forwarding
idl: add a representation of client_state for forwarding
cql_server: handle query, execute, batch in one case
transport: inline process_on_shard in cql_server::process
transport: extract process() to cql_server
transport: add messaging_service to cql_server
transport: add response reconstruction helpers for forwarding
transport: generalize the bounce result message for bouncing to other nodes
strong consistency: redirect requests to live replicas from the same rack
transport: pass foreign_ptr into sleep_until_timeout_passes and move it to cql_server
transport: extract the error handling from process_request_one
transport: move error response helpers from connection to cql_server
Update warn and fail messages for the write_consistency_levels_warned
and write_consistency_levels_disallowed guardrails to include the
configuration option name and actionable guidance. The main motivation
is to make the messages follow the conventions of other guardrails.
Refs: SCYLLADB-257
When computing table sizes via load_stats to determine if a split/merge is needed, we are filtering tablets which are being migrated, in order to avoid counting them twice (both on leaving and pending replica) in the total table size. The tablets are filtered so that they are counted on the leaving replica until the streaming stage, and on the pending replica after the streaming stage.
Currently, the procedure for collecting tablet sizes for load balancing also uses this same filter. This should be changed, because the load balancer needs to have as much information about tablet sizes as possible, and could ignore a node due to missing tablet sizes for tablets in the `write_both_read_new` and `use_new` stages.
For tablet size collection, we should include all the tablets which are currently taking up disk space. This means:
- on leaving replica, include all tablets until the `cleanup` stage
- on pending replica, include all tablets starting with the `write_both_read_new` and later stages
While this is an improvement, it causes problems with some of the tests, and therefore needs to be backported to 2026.1
Fixes: SCYLLADB-829
Closesscylladb/scylladb#28587
* github.com:scylladb/scylladb:
load_stats: add filtering for tablet sizes
load_stats: move tablet filtering for table size computation
load_stats: bring the comment and code in sync
Tests that call create_cluster() helper no longer need to carry the
rf-validity parameter. This simplifies the code and test signature.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The helper accepts its as boolean argument, but it can easily estimate
one from the provided topology.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Before b59b3d4 the migration code checked that service level controller
is on v2 version before migration and the check also implicitly checked
that _sl_data_accessor field is already initialized, but now that the
check is gone the migration can start before service level controller is
fully initialized. Re add the check, but to a different place.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-1049Closesscylladb/scylladb#29021
Permits in the `waiting_for_memory` state represent already-executing
reads that are blocked on memory allocation. Preemptively aborting
them is wasteful -- these reads have already consumed resources and
made progress, so they should be allowed to complete.
Restrict the preemptive abort check in maybe_admit_waiters() to only
apply to permits in the `waiting_for_admission` state, and tighten
the state validation in `on_preemptive_aborted()` accordingly.
Adjust the following tests:
+ test_reader_concurrency_semaphore_abort_preemptively_aborted_permit
no longer relies on requesting memory
+ test_reader_concurrency_semaphore_preemptive_abort_requested_memory_leak
adjusted to the fix
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-1016
- Added VECTOR to the comma-separated list of Jira project keys in `call_sync_milestone_to_jira.yml`.
- The `jira_project_keys` value changed from `SCYLLADB,CUSTOMER,SMI,RELENG` to `SCYLLADB,CUSTOMER,SMI,RELENG,VECTOR`.
- The VECTOR project needs to sync with scylladb.git milestones, so that when a GitHub milestone is created or closed in scylladb/scylladb, the corresponding Jira release is also created or released in the VECTOR project.
- Previously only SCYLLADB, CUSTOMER, SMI, and RELENG projects were synced.
Fixes:PM-220
Closesscylladb/scylladb#29014
This PR adds integrity verification for SSTable component files during loading. When component digests are present in Scylla metadata, the loader now validates each component's CRC32 digest against the stored expected value, catching silent corruption of component files. Index, Rows and Partitions components digests are also validated duriung scrub in validate mode
Added corruption tests that write an SSTable, flip a bit in a specific component file, then verify that reloading the SSTable detects the corruption and throws the expected exception.
Depends on https://github.com/scylladb/scylladb/pull/28338
Backport is not required, this is new feature
Fixes https://github.com/scylladb/scylladb/issues/20103Closesscylladb/scylladb#28761
* github.com:scylladb/scylladb:
test/cqlpy: test --ignore-component-digest-mismatch flag in scylla sstable upgrade
docs: document --ignore-component-digest-mismatch flag for scylla sstable upgrade
sstables: propagate ignore_component_digest_mismatch config to all load sites
sstables: add option to ignore component digest mismatches
sstable_compaction_test: Add scrub validate test for corrupted index
sstables: add tests for component digest validation on corrupted SSTables
sstables: validate index components digests during SSTable scrub in validate mode
sstables: verify component digests on SSTable load
sstables: add digest_file_random_access_reader for CRC32 digest computation
Fix several test cases that did not await async tasks:
- test_restart_leaving_replica_during_cleanup
- test_restart_in_cleanup_stage_after_cleanup
- test_tablet_back_and_forth_migration
- test_staging_backlog_is_preserved_with_file_based_streaming
Fixes SCYLLADB-910
* Minor fixes, no backport needed
Closesscylladb/scylladb#28908
* github.com:scylladb/scylladb:
test_tablets_migration: test_staging_backlog_is_preserved_with_file_based_streaming: convert for loop to asyncio.gather
test_tablets_migration: test_tablet_back_and_forth_migration: await move_tablet
test_tablets_migration: test_restart_in_cleanup_stage_after_cleanup: await move_task
test_tablets_migration: test_restart_leaving_replica_during_cleanup: await move_task
test_tablets_migration: drop unused imports from cassandra.query
Fixes#25084
Add slirp4netns and use for nested containers. This will allow nested container port aliasing, helping CI stability.
Note: this contains and updated Dockerfile for dbuild image, but since chicken and eggs, right now will force install slirp4netns before anything in dbuild script.
Updates the mock server handling to use ephemeral ports and query from container, ensuring we don't get port collisions. (boost as well as pytest).
Includes a timeout up, and a tweak to our scylla_cluster handling, ensuring we don't deadlock when pipe size is less than requires for our sys notify messages.
Closesscylladb/scylladb#28727
* github.com:scylladb/scylladb:
gcs_fixture: Change to use docker helper
aws_kms_fixture: Modify to use docker helper
test/lib/proc_util: Add docker helper
pytest: use ephemeral port publish for docker mock servers
dbuild: Use container network in dbuild nested containers
scylla_cluster: Read notify sock in background to prevent deadlock
A few days ago, in commit 7b30a39 we added to pytest.ini the option
xfail_strict. This option causes every time a test XPASSes, i.e., an xfail
test actually passes - to be considered an error and fail the test.
But some tests demonstrate a timing-related bug and do not reproduce the
bug every single time. An example we noticed in one CI run is:
test/cluster/test_alternator.py::test_alternator_concurrent_rmw_same_partition_different_server
This test reproduces a timing-related bug (if you do an LWT write to
one partition on to two different coordinators "at the same time", you
can get a failure), but only most of the time, not 100% of the time.
The solution is to add "strict=False" for the xfail marker on this specific
test. This undoes the xfail_strict for this specific test, accepting that
this specific test can either pass or fail. Note that this does NOT make
this test worthless - we still see this test failing most of the time, and
when a developer finally fixes this issue, the test will begin to pass all
the time.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-941
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#29016
When it deadlocks, groups stop merging and compaction group merge
backlog will run-away.
Also, graceful shutdown will be blocked on it.
Found by flaky unit test
test_merge_chooses_best_replica_with_odd_count, which timed-out in 1
in 100 runs.
Reason for deadlock:
When storage groups are merged, the main compaction group of the new
storage group takes a compaction lock, which is appended to
_compaction_reenablers_for_merging, and released when the merge
completion fiber is done with the whole batch.
If we accumulate more than 1 merge cycle for the fiber, deadlock
occurs. Lock order will be this
Initial state:
cg0: main
cg1: main
cg2: main
cg3: main
After 1st merge:
cg0': main [locked], merging_groups=[cg0.main, cg1.main]
cg1': main [locked], merging_groups=[cg2.main, cg3.main]
After 2nd merge:
cg0'': main [locked], merging_groups=[cg0'.main [locked], cg0.main, cg1.main, cg1'.main [locked], cg2.main, cg3.main]
merge completion fiber will try to stop cg0'.main, which will be
blocked on compaction lock. which is held by the reenabler in
_compaction_reenablers_for_merging, hence deadlock.
The fix is to wait for background merge to finish before we start the
next merge. It's achieved by holding old erm in the background merge,
and doing a topology barrier from the merge finalizing transition.
Background merge is supposed to be a relatively quick operation, it's
stopping compaction groups. So may wait for active requests. It
shouldn't prolong the barrier indefinitely.
Tablet boost unit tests which trigger merge need to be adjusted to
call the barrier, otherwise they will be vulnerable to the deadlock.
Two cluster tests were removed because they assumed that merge happens
in the backgournd. Now that it happens as part of merge finalization,
and blocks topology state machine, those tests deadlock because they
are unable to make topology changes (node bootstrap) while background
merge is blocked.
The test "test_tablets_merge_waits_for_lwt" needed to be adjusted. It
assumed that merge finalization doesn't wait for the erm held by the
LWT operation, and triggered tablet movement afterwards, and assumed
that this migration will issue a barrier which will block on the LWT
operation. After this commit, it's the barrier in merge finalization
which is blocked. The test was adjusted to use an earlier log mark
when waiting for "Got raft_topology_cmd::barrier_and_drain", which
will catch the barrier in merge finalization.
Fixes SCYLLADB-928
Needs to be ordered before split finalization, because storage_group
must be in split mode already at finalization time. There must be
split-ready compaction groups, otherwise finalization fails with this
error:
Found 0 split ready compaction groups, but expected 2 instead.
Exposed by increased split activity in tests.
Will be called in tests. It does the local part of the global topology
barrier.
The comment:
// We capture the topology version right after the checks
// above, before any yields. This is crucial since _topology_state_machine._topology
// might be altered concurrently while this method is running,
// which can cause the fence command to apply an invalid fence version.
was dropped, because it's no longer true after
fad6c41cee, and it doesn't make sense in
the context of local_topology_barrier(). We'd have to propagate the
version to local_topology_barrier(), but it's pointless. The fence
version is decided before calling the local barrier, and it will be
valid even if local version moves ahead.
This is short cleanup after recent removal of creating default cassandra superuser and auth-v1 code removal.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-1036
Backport: no, just code cleanup
Closesscylladb/scylladb#29004
* github.com:scylladb/scylladb:
auth: remove DEFAULT_SUPERUSER_NAME constant and dead DEFAULT_USER_PASSWORD
auth: use configurable default_superuser in describe_roles
auth: move default_superuser to common, remove _superuser member
auth: use LOCAL_ONE for all auth queries
auth: remove get_auth_ks_name indirection
can_use_effective_service_level_cache() always returns true now, so the function can be dropped entirely and all the code that assumes it may return false can be dropped as well. Also drop async versions of find_effective_service_level and get_user_scheduling_group since they are unused.
No need to backport, code removal,
Closesscylladb/scylladb#29002
* github.com:scylladb/scylladb:
service level: make maybe_update_per_service_level_params synchronous
service level: remove unused get_user_scheduling_group function
service level: drop async find_effective_service_level
service level: remove remnants of version 1 service level
Introduced by 54bddeb3b5, the yield was
added to write_cell(), to also help the general case where there is no
collection. Arguably this was unnecessary and this patch moves the yield
to write_collection(), to the cell write loop instead, so regular cells
don't have to poll the preempt flag.
Closesscylladb/scylladb#29013
This PR shortens two sleeps from 1s to 100ms to speed up bootstrap in tests.
The changed sleeps are:
- the pause duration in group0 discovery,
- the retry period in `wait_for_cql`.
Refs: https://scylladb.atlassian.net/browse/SCYLLADB-918
No backport: performance improvements mostly relevant to tests.
Closesscylladb/scylladb#29020
* github.com:scylladb/scylladb:
test: pylib: util: wait for CQL being ready with a shorter period
group0: discovery: shorten the pause duration
Add basic cluster tests for CQL forwarding.
The test cases include:
- basic reads and writes
- prepared statements with binds
- forwarding from a non-replica
- exception passthrough during forwarding (using an injection)
- re-preparing a statement on the target node, even if the user
query is also an EXECUTE request on a prepared statement
- verification metric updates
The existing test_basic_write_read was modified so that a few extra
cases could be validated on the same cluster.
We enable CQL forwarding by starting to return the bounce_to_node
result message in redirect_statement() instead of throwing. The
forwarding code introduced in the preceding patches reacts to these
messages, allowing the requests to be forwarded.
With the update, some tests assuming that requests can't be forwarded
need to be adjusted, so we do that as well.
During forwarding of CQL EXECUTE requests, the target node may
not have the prepared statement in its cache. If we do have this
statement as a coordinator, instead of returning PREPARED NOT FOUND
to the client, we want to prepare the statement ourselves on target
node.
For that, we add a new FORWARD_CQL_PREPARE RPC. We use the new RPC
after gettting the prepared_not_found status during forwarding. When
we try to forward a request, we always have the query string (we
decide whether to forward based on this query), so we can always use
the new RPC when getting the prepared_not_found status.
After receiving the response, we try forwarding the EXECUTE request
again.
During CQL forwarding, when the target node can't handle the request,
it will find another node which can execute the request or which knows
where the request can be executed. We return this information in
responses to CQL forwarding, and in this patch, we add handling of
this kind of a response.
After getting a redirect response, we retry forwarding to the returned
host/shard until success or timeout. This can happen many times during
a single request, when we first forward to a replica and later to the
coordinator, or when a replica/coordinator migrated while we were
performing the forwarding
When a forwarded request fails on the remote node, we can't use the
exception handling that happens in process_request_one because we
don't go through this code path. Instead, we use the previously
extracted cql_server::handle_exception handler, which performs
all accounting on the forwarded-to node, and which prepares the
response. For the read_failure_exception_with_timeout exception,
we need to perform the sleep on the source node, so we return the
timeout in the forwarding response and use it on the source node
to know how long to sleep without any extra calculations.
The handle_forward_execute() method is extracted from the inline handler
lambda to make the error catching wrapper cleaner.
Add the infrastructure for forwarding CQL requests to other nodes.
When a process() call results in a node bounce (as opposed to a shard
bounce), the coordinator serializes the request and sends it via the
FORWARD_CQL_EXECUTE RPC verb to the target node.
In this patch we omit several features that allow handling more
scenarios that can happen when trying to forward a CQL request,
but the RPC request and response are already prepared for them.
They will be handled in the following commits.
Use rolling_max_tracker to record gross bytes allocated during each
CQL parse. The rolling maximum is then added to the memory estimate
for incoming QUERY and PREPARE requests so that the admission control
in the CQL transport layer accounts for parsing overhead.
The measured memory footprint serves as upper bound rather than
exact number but it's purpose is to prevent OOMs under unprepared
statements heavy load.
In benchmark 1G memory node shows decrease of non-LSA memory usage
from peak 320MB (our coordinator budget is 10% of 1G) to 96MB. While
tps drops from 1.2 kops to 0.8 kops. Drop in tps is expected as
memory admission kicks in trying to prevent OOM.
This is phase 1 of OOM prevention, potential next steps:
- add second admission in query_processor::get_statement trying to prevent potential thundering herd problem
- decrease cql_server memory pool size
- count reads in the memory pool
- add per service level memory pool and a shared one
Related https://scylladb.atlassian.net/browse/SCYLLADB-740
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-938
Backport: no, new feature, but we may reconsider if some customer needs it
Closesscylladb/scylladb#28919
* github.com:scylladb/scylladb:
cql3: track CQL parsing memory cost and use it for admission control
utils: add rolling max tracker
In the following patches, when we start allowing to forward CQL
requests to other nodes, we'll need to use the same client state
for executing the request on the destination node as we had on the
source. client_state contains many fields and we need to create
a new instance of it when we start handling the forwarded request,
so to prepare for the forwarding RPC, we add a serializable format
of the client_state as an IDL struct. The new class is missing some
fields that are not used while executing requests, and some whose
value is determined by the fact that the client state is used for
a forwarded request.
These include:
- driver name, driver version, client options - not used for executing
requests. Instead, we use these as data sources for the virtual
"clients" system table.
- auth_state - must be READY - we reached a bounce message, so we were
able to try executing the request locally
- _control_connection - used for altering a cql_server::connection, which
we don't have on the target node
- _default_timeout_config - used when updating service levels, also only
per-connection
- workload_type - used for deciding whether to allow shedding at the
start of processing the request, and for getting per-connection service
level params (for an API)
Currently we perform the same steps when handling query, execute
and batch CQL requests. So instead of creating multiple functions
performing these steps, we can handle them all in one fallthrough
case in cql_server::connection::process_request_one.
The process_on_shard method is relatively short, it's only used
in the process() method and the Process concept that is uses
is as long as the function itself. This area will be made more
complex by the following patches for cql forwarding, so we simplify
it by inlining process_on_shard in cql_server::process.
Move process() and process_on_shard() from cql_server::connection to
cql_server. The process() method is no longer a template - instead, it
takes an opcode parameter and uses get_process_fn_for_opcode() to select
the appropriate internal processing function.
The process_query, process_execute, and process_batch wrappers on
connection now delegate to _server.process() with the appropriate opcode.
This refactoring is preparation for CQL request forwarding, where
process() will need to be called from a context other than connection
- the forwarding RPC handler).
The messaging service will be used by cql_server to register RPC
handlers for forwarding CQL requests between nodes.
We pass it through the controller to cql_server.
Expose response::flags() and response::extract_body(), and a new constructor.
It will be needed for creating a cql_transport::response from the response body returned
during CQL forwarding.
In the following patches, we'll start allowing forwarding requests to strongly
consistent tables so that they'll get executed on the suitable tablet Raft group
members. For that we'll reuse the approach that we already have for bouncing
requests to other shards - we'll try to execute a request locally, and the
result of that will be a bounce message with another replica as the target.
In this patch we generalize the former bounce_to_shard result message so that
it will be able to specify the target of the bounce as another shard or specific
replica.
We also rename it to result_message::bounce so that it stops implying that only
another shard may be its target.
Aside from the host_id and the shard, the new message also includes the timeout,
because in the service handling the forwarding we won't have the access to it,
and it's needed for specifying how long we should wait for the forwarded
requests. It also includes an information whether this is a write request
to return correct timeout response in case the deadline is exceeded.
We will return other hosts in the new bounce message when executing requests to
strongly consistent tables when we can't handle the request because we aren't
a suitable replica. We can't handle this message yet, so we don't return it
anywhere and we still assume that every bounce message is a bounce to the same
host.
Forwarding CQL requests is not implemented yet, but we're already
prepared to return the target to forward to when trying to execute
strongly consistent requests. Currently, if we're not a replica
of the affected tablet, we redirect the request to the first replica
in the list.
This is not optimal, because this replica may be down or it may be
in another rack, making us perform cross-rack requests during forwarding.
Instead, we should forward the request to the replica from the same
rack and handle the case where the replica is down.
In this patch we change the replica selection for forwarding strongly
consistent requests, so that when the coordinator isn't a replica, it
redirects the request to the replica from the same rack.
If the replica from the same rack is down, or there is no replica in
our rack, we choose the next closest replica (preferring same-DC replicas
over other DCs). If no replica is alive, the query fails - the driver
should retry when some replica comes back up.
A permit in `waiting_for_memory` state can be preemptively aborted by
maybe_admit_waiters(). This is wrong: such permits have already been
admitted and are actively processing a read — they are merely blocked
waiting for memory under serialize-limit pressure.
When `on_preemptive_aborted()` fires on a `waiting_for_memory` permit,
it does not clear `_requested_memory`. A subsequent `request_memory()`
call accumulatesa on top of the stale value, causing `on_granted_memory()`
to consume more than resource_units tracks.
This commit adds a test that confirms that scenario by counting
internal_errors.
The test_raft_voters_multidc_kill_dc scenario had become weaker after group0 voter count was made always odd.
In particular, the old num_nodes == 1 case (dc1=2, dc2=1, dc3=1) could pass even without the intended balancing logic, because with 3 voters total we naturally get one voter per DC.
This change restores coverage of the original intent:
- Replace num_nodes parametrization with explicit DC triples.
- Use (3, 1, 1) to force a meaningful asymmetric topology where voter placement logic is required.
- Keep a larger topology case (6, 3, 3) for broader coverage.
- Mark (6, 3, 3) as skip_mode(debug) with reason:
larger topology case is too slow in debug on minipcs.
Also updated comments/docstring to match the new setup.
Fixes: SCYLLADB-794
backport: None, it is done to deflake minipcs that will start working only on master
Closesscylladb/scylladb#29000
Change sleep_until_timeout_passes() to accept a foreign_ptr<std::unique_ptr<response>>.
We can easily create the foreign_ptr for the responses created in the CQL server,
but we'll need this when we get responses when forwarding CQL statements - the responses
may come from other shards.
We also move it from cql_server::connection to cql_server, because for forwarded CQL
requests, we'll need to handle it at the cql_server level.
The method also loses its const qualifier - the abort_source that we pass into
sleep_abortable needs to be non-const. Apparently, we could still use it in a const
method of cql_server::connection because we passed it as _server._abort_source which
caused the const qualifier to be lost.
After stop() moved _reaper, in-flight with_connection() callbacks could
still call reap(), which accessed the moved-from future causing a
SIGSEGV in future_base::detach_promise(). Add a seastar::gate so
stop() waits for all in-flight operations before moving _reaper.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-1043Closesscylladb/scylladb#29015
`wait_for_cql` is used in hundreds, if not thousands, of places in tests.
We shouldn't waste up to 1s for every call.
Also, the 1s period is clearly too long compared to the bootstrap time,
which is usually 0-3s in dev mode.
The following test speeds up from 50s to 42s with the change:
```
for _ in range(10):
servers = await manager.servers_add(3)
await manager.get_ready_cql(servers)
```
Nodes currently pause group0 discovery for 1s. This case is always hit while
adding multiple nodes in parallel to an empty cluster by all nodes except the
one that becomes the group0 leader.
This is fine in production, but in tests, the slowdown is quite significant.
Every `manager.servers_add(n)` call for n > 1 becomes 1s slower when the
cluster is empty. Many cluster tests are affected.
In this commit, we decrease the sleep duration from 1s to 100ms to speed up
tests. The consequence of this change is that nodes might perform more steps
in group0 discovery, but the increase in CPU usage and network traffic should
be negligible.
Currently the test iterates on all servers and calls manager.api.disable_injection
but it doesn't await those calls.
Use asyncio.gather to await all calls in parallel.
Co-authored-by: Copilot CLI
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
serialize_collection_mutation() copies the serialized collection into
the returned collection_mutation object. Change to move to avoid the
copy.
Fixes: SCYLLADB-1041
Closesscylladb/scylladb#29010
can_use_effective_service_level_cache() always returns true now, so the
function can be dropped entirely and all the code that assumes it may
return false can be dropped as well.
The Alternator test test_compressed_request.py::test_gzip_request_oversized
checks that a very large request that compresses to a small size is still
rejected. This test passed on Alternator, but used to fail on DynamoDB
because DynamoDB didn't reject this case. This was a bug in DynamoDB
(a "decompression bomb" vulnerability), and after I reported it, it
was fixed.
So now this test does pass on DynamoDB (after a small modification to
allow for different error codes). So remove its scylla_only marker,
and make the comment true to the current state.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#28820
Use rolling_max_tracker to record gross bytes allocated during each
CQL parse. The rolling maximum is then added to the memory estimate
for incoming QUERY and PREPARE requests so that the admission control
in the CQL transport layer accounts for parsing overhead.
The measured memory footprint serves as upper bound rather than
exact number but it's purpose is to prevent OOMs under unprepared
statements heavy load.
In benchmark 1G memory node shows decrease of non-LSA memory usage
from peak 320MB (our coordinator budget is 10% of 1G) to 96MB. While
tps drops from 1.2 kops to 0.8 kops. Drop in tps is expected as
memory admission kicks in trying to prevent OOM.
In this patch we replace every single use of SCYLLA_ASSERT(), abort() and assert() in the cql3/ directory by throwing_assert().
The problem with SCYLLA_ASSERT()/abort()/assert() is that when it fails, it crashes Scylla. This is almost always a bad idea (see #7871 discussing why), but it's even riskier in front-end code like cql3/: In front-end code, there is a risk that due to a bug in our code, a specific user request can cause Scylla to crash. A malicious user can send this query to all nodes and crash the entire cluster. When the user is not malicious, it causes a small problem (a failing request) to become a much worse crash - and worse, the user has no idea which request is causing this crash and the crash will repeat if the same request is tried again.
All of this is solved by using the new throwing_assert(), which is the same as SCYLLA_ASSERT() but throws an exception (using on_internal_error()) instead of crashing. The exception will prevent the code path with the invalid assumption from continuing, but will result in only the current user request being aborted, with a clear error message reporting the internal server error due to an assertion failure.
I reviewed all the changes that I did in these patches to check that (to the best of my understanding) none of the assertions in cql3/ involve the sort of serious corruption that might require crashing the Scylla node entirely.
throwing_assert() also improves logging of assertion failures compared to the original SCYLLA_ASSERT()/abort() - SCYLLA_ASSERT() printed a message to stderr which in many installations is lost, and abort() often prints no message at all. But throwing_assert() uses Scylla's standard logger, and also includes a backtrace in the log message.
Fixes#13970 (Exorcise assertions from CQL code paths)
Refs #7871 (Exorcise assertions from Scylla)
Closesscylladb/scylladb#28847
* github.com:scylladb/scylladb:
cql3: remove unnecessary assert()
cql3: replace abort() by throwing_assert()
cql3: Replace SCYLLA_ASSERT by throwing_assert
The `test/cqlpy/cassandra_tests/validation/entities/json_test.py::testJsonOrdering` was failing because of differences between Cassandra and Scylla in printing
JSON floating point values - e.g. Cassandra prints 30.0, where Scylla prints 30.
Both are valid, so in this patch, instead of comparing strings, we compare parsed JSON using `EquivalentJson`.
Fixes#28467Closesscylladb/scylladb#28924
Replace the hardcoded meta::DEFAULT_SUPERUSER_NAME comparison with
default_superuser(_qp) which reads from the auth_superuser_name
config option. This makes the IF NOT EXISTS clause in DESCRIBE
output correct for clusters with a non-default superuser name.
_"A journey of a thousand miles begins with a single step" Lao Tzu_
ScyllaDB uses estimated_histogram in many places.
We already have a more efficient alternative: approx_exponential_histogram. It is both CPU and
memory-efficient and can be exported as Prometheus native histograms.
Its main limitation (which has its benefits) is that the bucket layout is fixed at compile time, so
histograms with different configurations cannot be mixed.
The end goal is to replace all uses of estimated_histogram in the codebase.
That migration needs a few small API adjustments, so I am splitting the work
into steps for easier review.
This series is the first step. It introduces a base template for fixed-size
estimated histograms, and switches the Alternator's estimated_histogram with the template.
This change is self-contained and valuable on its own, while keeping the scope limited.
Minor adjustments were made to the code and tests so that the tests would pass.
Follow-up PRs will apply the same pattern to the rest of the code.
**New feature no need to backport**
Closesscylladb/scylladb#28987
* github.com:scylladb/scylladb:
alternator: migrate to operation_size_kb histograms
test/alternator/test_metrics.py: Update the bucket in the histogram search
alternator: Use batch_histogram for batch size histograms
estimated_histogram.hh: adds estimated_histogram_with_max
When we forward CQL statements, we'll need to handle the errors
on the destination node. Only for read_failure_exception_with_timeout
exception, we'll still need to wait until timeout passes on the
source node.
For that we extract the exception handling to a separate method.
Additionally, we separate the waiting and all other handling,
so that all handling aside from waiting will be reusable after
forwarding, and we'll also be able to sleep on the source node
if necessary.
These methods are used only in the error handler in the cql server,
and outside of 3 cases, they don't need any information from the
cql_server::connection. We move them from cql_server::connection
to cql_server, so that they can be used in the following patches
for methods for CQL request forwarding where we'll have no instance
of cql_server::connection on the node forwarded to.
After the change the methods require no access to the server's
or connection's fields, so we also make them static methods.
Switch Alternator operation-size metrics from the legacy estimated
histogram implementation to estimated_histogram_with_max<512> and export
them through the native approx-exponential histogram path.
Add a dedicated operation-size histogram type alias based on
estimated_histogram_with_max<512>.
Replace all per-operation size histograms (GetItem/PutItem/DeleteItem/
UpdateItem/BatchGetItem/BatchWriteItem) with the new type.
Remove the custom legacy histogram-to-metrics adapter and use
to_metrics_histogram() for operation size metrics, aligning export
behavior with other approx-exponential histograms.
Update Alternator metrics tests to compute expected le bucket boundaries using
approx-exponential bucket math (including deduplication of equal
bounds), so assertions match the new exported histogram schema.
Update bucket helper signatures to use (max, precision) parameters and keep
+Inf handling unchanged.
Replace byte-to-KB ceiling conversion with plain integer division (bytes
/ 1024): histogram export already reports each bucket by its upper bound
(le), so rounding input values up before bucketing is unnecessary and
would over-shift borderline samples into higher buckets.
Move default_superuser() to auth::meta in common.{hh,cc} and remove the
cached _superuser member from both standard_role_manager and
password_authenticator. The superuser name comes from config which is
immutable at runtime, so caching it is unnecessary.
Removes auth-v1 hack for cassandra superuser as auth-v1
code no longer exists.
Also CL is not really used when quering raft replicated
tables (like auth ones), but LOCAL_ONE is the least confusing
one.
Replace get_auth_ks_name(qp) with db::system_keyspace::NAME directly.
The function always returned the constant "system" and its qp
parameter was unused.
nodetool cluster repair without additional params repairs all tablet
keyspaces in a cluster. Currently, if a table is dropped while
the command is running, all tables are repaired but the command finishes
with a failure.
Modify nodetool cluster repair. If a table wasn't specified
(i.e. all tables are repaired), the command finishes successfully
even if a table was dropped.
If a table was specified and it does not exist (e.g. because it was
dropped before the repair was requested), then the behavior remains
unchanged.
Fixes: SCYLLADB-568.
Closesscylladb/scylladb#28739
What changed
* Added closed to milestone event types in
call_sync_milestone_to_jira.yml (types: [created] -> types: [created, closed])
* Added VECTOR to the list of Jira project keys being synced
(jira_project_keys: SCYLLADB,CUSTOMER,SMI,RELENG -> jira_project_keys: SCYLLADB,CUSTOMER,SMI,RELENG,VECTOR)
Why (Requirements Summary)
* The call_sync_milestone_to_jira.yml workflow only triggered on
milestone creation. When a GitHub milestone is closed, the
corresponding Jira versions (in SCYLLADB, CUSTOMER, SMI, RELENG
projects) should be marked as released. Adding the closed trigger
enables the called workflow (main_sync_milestone_to_jira_release.yml
in github-automation) to handle both creating and releasing Jira
versions from GitHub milestone events.
* Added the VECTOR project so its Jira versions are also created/released
when milestones are created or closed in scylladb.git.
* This is consistent with the same change already applied to the staging
and scylla-machine-image repos.
Fixes:PM-216
Update call_sync_milestone_to_jira.yml in scylladb.git - add close trigger and VECTOR project sync
Closesscylladb/scylladb#28981
This patch adds estimated_histogram_with_max template that will be a
based for specific estimated_histograms, eventually replacing the current
struct implementation.
Introduce estimated_histogram_with_max<Max> as a reusable wrapper around
approx_exponential_histogram<1, Max, 4>, providing merge support and the
same add helpers used by existing estimated_histogra type.
Add estimated_histogram_with_max_merge()
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Fix an invalid condition, when searching for a parent shard, when table
is based on vnodes. Shards have associated with them `last token` -
token, than marks the end of the range of tokens they consume (inclusive).
An additional assumptions are whole token space is used and
(for vnodes) token space wraps around.
Previously code looked like this:
auto pid = std::upper_bound(..., [](const dht::token& t, const cdc::stream_id& id) {
return t < id.token();
});
if (pid != pids.begin()) {
pid = std::prev(pid);
}
An `upper_bound` call with `t < id.token()` means it is looking for
an iterator, for which value `t < id.token()` changed to true,
which effectively means a position, where iterator is bigger
then searched value. Then we move iterator backward once if possible.
Assuming token space <-2, 2> and parents [0, 2], when we search for:
- -1 -> we will get 0, it's first, so we can't move backward, so 0 (ok)
- 0 -> we will get 2, it's not first, so we go back and we return 0 (ok)
- 1 -> we will get 2, it's not first, so we go back and we return 0
(not ok - should be 2)
The fix is to replace it with `std::lower_bound` and remove conditional
backward motion. Since we've a guarantees that whole token space is used
if `std::lower_bound` ends with `end()` value, then we have a wrap
around case and we need to pick `begin()` as result.
Fixes#28354
Fixes: SCYLLADB-537
Closesscylladb/scylladb#28382
Changes dockerized_service to use ephermal port publish, and
query the published port from podman/docker.
Modifies client code to use slightly changed usage syntax.
query_processor::prepare() could race with prepared statement invalidation: after loading from the prepared cache, we converted the cached object to a checked weak pointer and then continued asynchronous work (including error-injection waitpoints). If invalidation happened in that window, the weak handle could no longer be promoted and the prepare path could fail nondeterministically.
This change keeps a strong cache entry reference alive across the whole critical section in prepare() by using a pinned cache accessor (get_pinned()), and only deriving the weak handle while the entry is pinned. This removes the lifetime gap without adding retry loops.
Test coverage was extended in test/cluster/test_prepare_race.py:
- reproduces the invalidation-during-prepare window with injection,
- verifies prepare completes successfully,
- then invalidates again and executes the same stale client prepared object,
- confirms the driver transparently re-requests/re-prepares and execution succeeds.
This change introduces:
- no behavior change for normal prepare flow besides stronger lifetime guarantees,
- no new protocol semantics,
- preserves existing cache invalidation logic,
- adds explicit cluster-level regression coverage for both the race and driver reprepare path.
- pushes the re prepare operation twards the driver, the server will return unprepared error for the first time and the driver will have to re prepare during execution stage
Fixes: https://github.com/scylladb/scylladb/issues/27657
Backport to active branches recommended: No node crash, but user-visible PREPARE failures under rare schema-invalidation race; low-risk timeout-bounded retry improves robustness.
Closesscylladb/scylladb#28952
* github.com:scylladb/scylladb:
transport/messages: hold pinned prepared entry in PREPARE result
cql3: pin prepared cache entry in prepare() to avoid invalid weak handle race
Remove the host network setting, ensuring we use private networks
(slirp4netns). This will allow nested container port aliasing,
helping CI stability (can use ephemeral ports and container
introspection).
This also makes the nested podman setup non-conditional,
since we only run podman containers inside dbuild, and need
the setup regardless if host container is docker or not.
Starts a thread to process scylla notify messages (NOTIFY_SOCKET)
instead of just processing inline, non-blocking. This because it
is possible for the pipe created to be to small to hold enough
messages for us to reach the point where we otherwise even read
from said pipe, allowing other end (scylla) to proceed.
This series adds a global read barrier to raft_group0_client, ensuring that Raft group0 mutations are applied on all live nodes before returning to the caller.
Currently, after a group0_batch::commit, the mutations are only guaranteed to be applied on the leader. Other nodes may still be catching up, leading to stale reads. This patch introduces a broadcast read barrier mechanism. Calling send_group0_read_barrier_to_live_members after committing will cause the coordinator to send a read barrier RPC to all live nodes (discovered via gossiper) and waits for them to complete. This is best effort attempt to get cluster-wide visibility of the committed state before the response is returned to the user.
Auth and service levels write paths are switched to use this new mechanism.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-650
Backport: no, new feature
Closesscylladb/scylladb#28731
* https://github.com/scylladb/scylladb:
test: add tests for global group0_batch barrier feature
qos: switch service levels write paths to use global group0_batch barrier
auth: switch write paths to use global group0_batch barrier
raft: add function to broadcast read barrier request
raft: add gossiper dependency to raft_group0_client
raft: add read barrier RPC
Remove the rest of the code that assumes that either group0 does not exist yet or a cluster is till not upgraded to raft topology. Both of those are not supported any more.
No need to backport since we remove functionality here.
Closesscylladb/scylladb#28841
* github.com:scylladb/scylladb:
service level: remove version 1 service level code
features: move GROUP0_SCHEMA_VERSIONING to deprecated features list
migration_manager: remove unused forward definitions
test: remove unused code
auth: drop auth_migration_listener since it does nothing now
schema: drop schema_registry_entry::maybe_sync() function
schema: drop make_table_deleting_mutations since it should not be needed with raft
schema: remove calculate_schema_digest function
schema: drop recalculate_schema_version function and its uses
migration_manager: drop check for group0_schema_versioning feature
cdc: drop usage of cdc_local table and v1 generation definition
storage_service: no need to add yourself to the topology during reboot since raft state loading already did it
storage_service: remove unused functions
group0: drop with_raft() function from group0_guard since it always returns true now
gossiper: do not gossip TOKENS and CDC_GENERATION_ID any more
gossiper: drop tokens from loaded_endpoint_state
gossiper: remove unused functions
storage_service: do not pass loaded_peer_features to join_topology()
storage_service: remove unused fields from replacement_info
gossiper: drop is_safe_for_restart() function and its use
storage_service: remove unused variables from join_topology
gossiper: remove the code that was only used in gossiper topology
storage_service: drop the check for raft mode from recovery code
cdc: remove legacy code
test: remove unused injection points
auth: remove legacy auth mode and upgrade code
treewide: remove schema pull code since we never pull schema any more
raft topology: drop upgrade_state and its type from the topology state machine since it is not used any longer
group0: hoist the checks for an illegal upgrade into main.cc
api: drop get_topology_upgrade_state and always report upgrade status as done
service_level_controller: drop service level upgrade code
test: drop run_with_raft_recovery parameter to cql_test_env
group0: get rid of group0_upgrade_state
storage_service: drop topology_change_kind as it is no longer needed
storage_service: drop check_ability_to_perform_topology_operation since no upgrades can happen any more
service_storage: remove unused functions
storage_service: remove non raft rebuild code
storage_service: set topology change kind only once
group0: drop in_recovery function and its uses
group0: rename use_raft to maintenance_mode and make it sync
The `service_error` struct: 6dc2c42f8b/service/vector_store_client.hh (L64)
currently stores just the error status code. For this reason whenever the HTTP error occurs, only the error code can be forwarded to the client. For example see here: 6dc2c42f8b/service/vector_store_client.cc (L580)
For this reason in the output of the drivers full description of the error is missing which forces user to take a look into Scylla server logs.
The objective of this PR is to extend the support for HTTP errors in Vector Store client to handle messages as well.
Moreover, it removes the quadratic reallocation in response_content_to_sstring() helper function that is used for getting the response in case of error.
Fixes: VECTOR-189
Closesscylladb/scylladb#26139
* github.com:scylladb/scylladb:
vector_search: Avoid quadratic reallocation in response_content_to_sstring
vector_store_client: Return HTTP error description, not just code
Refs: SCYLLADB-557
We should use full replication in KS/CF creation and population,
for at least two reasons:
1.) Ensure we wait fully for and write to all nodes
2.) Make test more "real", behaving like a proper cluster
Closesscylladb/scylladb#28959
In cql3/, there was one call to assert() (not SCYLLA_ASSERT or
throwing_assert), and it was:
const auto shard_num = smp::count;
assert(shard_num > 0)
Rather than converting this assert() to throwing_assert() as I did in
previous patches, I decided to outright remove it: Seastar guarantees
that smp::count is not zero. Many other places in the code use
smp::count assuming that it is correct, no other place bothers to assert
it isn't zero.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
After the previous patch replaced all SCYLLA_ASSERT() calls by
throwing_assert(), this patch also replaces all calls to abort().
All these abort() calls are supposedly cases that can never happen,
but if they ever do happen because of a bug, in none of these places
we absolutely need to crash - and exception that aborts the current
operation should be enough.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
In this patch we replace every single use of SCYLLA_ASSERT() in the cql3/
directory by throwing_assert().
The problem with SCYLLA_ASSERT() is that when it fails, it crashes Scylla.
This is almost always a bad idea (see #7871 discussing why), but it's even
riskier in front-end code like cql3/: In front-end code, there is a risk
that due to a bug in our code, a specific user request can cause Scylla
to crash. A malicious user can send this query to all nodes and crash
the entire cluster. When the user is not malicious, it causes a small
problem (a failing request) to become a much worse crash - and worse,
the user has no idea which request is causing this crash and the crash
will repeat if the same request is tried again.
All of this is solved by using the new throwing_assert(), which is the
same as SCYLLA_ASSERT() but throws an exception (using on_internal_error())
instead of crashing. The exception will prevent the code path with the
invalid assumption from continuing, but will result in only the current
user request being aborted, with a clear error message reporting the
internal server error due to an assertion failure.
I reviewed all the changes that I did in this patch to check that (to the
best of my understanding) none of the assertions in cql3/ involve the
sort of serious corruption that might require crashing the Scylla node
entirely.
throwing_assert() also improves logging of assertion failures compared
to the original SCYLLA_ASSERT() - SCYLLA_ASSERT() printed a message to
stderr which in many installations is lost, whereas throwing_assert()
uses Scylla's standard logger, and also includes a backtrace in the
log message.
Fixes#13970 (Exorcise assertions from CQL code paths)
Refs #7871 (Exorcise assertions from Scylla)
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
To restore how streaming scopes work there are two tests that greatly duplicate each other -- test_restore_with_streaming_scopes from cluster/object_store suite and test_refresh_with_streaming_scopes from cluster suite.
This patch generalizes both into a do_test_streaming_scopes() non-test function
Closesscylladb/scylladb#28874
* github.com:scylladb/scylladb:
test: Re-sort comments around do_test_streaming_scopes()
test: Split do_load_sstables()
test: Drop load_fn argument from do_load_sstables()
test: Re-use do_test_streaming_scopes() in refresh test
test: Introduce SSTablesOnLocalStorage
test: Introduce SSTablesOnObjectStorage
test: Move test_restore_with_streaming_scopes() into do_test_streaming_scopes()
Document how to migrate a ScyllaDB cluster to different instance
types using the add-and-replace node cycling approach.
Closes: QAINFRA-42
Closesscylladb/scylladb#28458
Recently, in commit 7b30a39, we added to pytest.ini the option xfail_strict.
This option causes every time a test XPASSes, i.e., an XFAIL test actually
passes, to be considered an error and fail the test.
While this has some benefits, it's a big problem when running tests
against a reference implementation like DynamoDB or Cassandra: We
typically mark a test "xfail" if the test shows a known bug - i.e., if
the test fails on Scylla but passes on the reference system (DynamoDB
or Cassandra). This means that when running "test/cqlpy/run-cassandra"
or "test/alternator/run --aws", we expect to see many tests XPASS,
and now this will cause these runs to "fail".
So in this patch we add the xfail_strict=false to cqlpy/run-cassandra
and alternator/run --aws. This option is not added to cqlpy/run or
to alternator/run without --aws, and also doesn't affect test.py or
Jenkins.
P.S. This is another nail in the coffin of doing "cd test/alternator;
pytest --aws". You should get used to running Alternator tests through
test/alternator/run, even if you don't need to run Scylla (the "--aws"
option doesn't run Scylla).
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#28973
During tests I noticed that if the number of tablets is very small,
say 2, and the number of nodes is 3 (2 shards per node), using the
number of storage groups on each shard, a shard may end up holding 0 groups,
whilst the other holds 1 group. And in some nodes even both shards have
0 groups.
Taking the minimum among shards here was showing in manifests a tablet
count of 0 for all 3 nodes, which is incorrect.
Signed-off-by: Robert Bindar <robert.bindar@scylladb.com>
Closesscylladb/scylladb#28978
Currently, for repair tasks tablet_virtual_task::wait gathers the
ids of tablets that are to be repaired. The gathered set is later
used to check if the repair is still ongoing.
However, if the tablets are resized (split or merged), the gathered
set becomes irrelevant. Those, we may end up with invalid tablet id
error being thrown.
Wait until repair is done for all tablets in the table.
Fixes: https://github.com/scylladb/scylladb/issues/28202
Backport to 2026.1 needed as it contains the change introducing the issue d51b1fea94Closesscylladb/scylladb#28323
* github.com:scylladb/scylladb:
service: fix indentation
test: add test_tablet_repair_wait
service: remove status_helper::tablets
service: tasks: scan all tablets in tablet_virtual_task::wait
A bug or some bad operator intervention can lead to a sstable existing in a
node after the tablet replica was moved to a different node.
This will result sstable loading during boot failing, requiring operator
intervention.
The log today just dumps the name of the "orphaned" sstable, but one
investigating it might want to know which process (repair, memtable, whatever)
generated that sstable, if the sstable was created locally or remotely,
and the current replica set of the underlying tablet.
From the original identifier, we can know the exact time the sstable was
created on its original node. From the current id, we know the time it
was created on the current node.
All this info can help the investigator to correlate with events in other nodes
(includes actions from the coordinator) to get closer to the root cause.
The new log will look like this:
"Unable to load SSTable .../me-3gyg_1fsw_2u0u826b00b71vc46o-big-Data.db
(originated from compaction with id 913f41c0-18c2-11f1-8f08-cb8521b3f330
on host e483238c-2287-4022-8bc4-b4f1c4cb2b0d)
of tablet 6 (replica set: [e483238c-2287-4022-8bc4-b4f1c4cb2b0d:0])"
Refs https://scylladb.atlassian.net/browse/SCYLLADB-788.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Closesscylladb/scylladb#28921
Vector deserialization is an operation which performance is critical for
vector similarity search feature because it is frequently executed during
rescoring operation. Some of the identified performance bottlenecks
for it include:
1. Per-element virtual dispatch in deserialize(): each of the N elements
went through visit() which switches on ~28 type variants. For a
1024-dimension float vector, that's 1024 redundant type switches when
the element type is the same for all of them.
2. Redundant work in split_fragmented(): value_length_if_fixed() was
called inside the loop (N virtual calls), and no reserve() was done
on the output vector causing repeated reallocations.
This series fixes both:
- Introduce deserialize_vector_visitor that dispatches on the element
type once for the entire vector, then loops inside the resolved
handler. Simple numeric types (float, int, etc.) call
deserialize_value() directly with no virtual dispatch per element.
String types (ascii, utf8) get a dedicated handler that skips
make_empty() (sstring has no empty_t constructor). Complex types
(list, map, tuple, etc.) fall back to per-element dispatch.
- In split_fragmented(), reserve the output vector to _dimension and
cache value_length_if_fixed() before the loop.
Benchmark results (1024-dim float vector, release build, -O3 -flto):
deserialize: 15.73 us -> 11.70 us (1.34x, 26% faster)
split_fragmented: 10.34 us -> 7.45 us (1.39x, 28% faster)
References: SCYLLADB-471
Backport: none, unless we observe some critical performance improvement for quantization.
Closesscylladb/scylladb#28618
* github.com:scylladb/scylladb:
types: optimize reading vector fragments
types: optimize vector deserialization for high-dimensional vectors
Similarly to LWTs, we reject queries with user-provided timestamps
when they target strongly consistent tables.
Such statements could force us to rewrite history, and that contradicts
the philosophy of linearizability we aim for.
Fixes SCYLLADB-879
Closesscylladb/scylladb#28867
* seastar d2953d2a...4d268e0e (32):
> Merge 'prometheus: support multiple __name__ filters and prefixed names' from Travis Downs
doc: update prometheus.md with __name__ filter enhancements
prometheus: support prefixed names in __name__ filter
prometheus: add benchmarks for name filter performance
prometheus: support multiple __name__ query parameters
prometheus: move write_body_args to header
> fair_queue: Subtract from _queued_capacity on pop_front()
> memory: expose cumulative allocated bytes statistic
> Merge 'Add ability to configure IO bandwidth limit for supergroup' from Pavel Emelyanov
test: IO bandiwdth throttler unit tests
code: Add ability to configure IO bandwidth limit for supergroup
io_queue: Have more than one throttler par class
io_queue: Introduce bandwidth_throttler helper class
io_queue: Nest io_group::priotiy_class_data-s
io_queue: Update class bandwidth on group's class data
io_queue: Make io_group::priority_class_data::tokens() static
fair_queue: Introduce group (un)plugging
> Fix _shard_to_numa_node_mapping double population
> Use exception parameter in log_timer_callback_exception()
> Fix wakeup_granularity() fallback debug-fs reading
> test_fixture: Fix SEASTAR_FIXTURE_THREAD_TEST_CASE thread not propagated
> build: support tuning -ffile-prefix-map
> test: Remove unused C::dup() method of testing class
> src/core/reactor: introduce reactor::get_backend_name()
> util/process: add pid() accessor
> Merge 'Add source location to task and tasktrace object' from Radosław Cybulski
coroutine.hh: disable source_location for GCC to avoid ICE
reactor: improve do_dump_task_queue reporting
Use source_location in `do_dump_task_queue`
Update backtrace with source locations of resume points
Add calls to update resume_point
Add a std::source_location (resume_point) to task object.
> Merge 'Refine posix file .dup() implementation' from Pavel Emelyanov
file: Templatize posix_file_handle_impl
file: Don't dup() non-read-only files
file: Split ..._impl::dup() implementations
test: Add a simple test for dup()
> Merge 'Deprecate reactor::make_pollable_fd(socket_address, int)' from Pavel Emelyanov
reactor: Deprecate make_pollable_fd()
net/posix: Create file_desc for sockets in-place
reactor,net: Keep sock_need_nonblock boolean on posix_network_stack
net/posix: Re-format constructor initializer lists
> Merge 'test: add fuzz testing infrastructure and sstring fuzzer' from Travis Downs
test: add fuzz tests to CI workflow
test: add sstring differential fuzzer
test: add fuzz testing infrastructure
> Introduce "integrated queue length" metrics and use it for IO classes (#3210)
> reactor: Remove get_sg_data(unsigned) overload
> memcached: Stop using scattered_message
> reactor: Mark uptime() method const
> alien: Remove deprecated run_on and submit_to calls
> file: make open_flags and access_flags constexpr
> scheduling: Unfriend some methods from scheduling_group
> reactor: Move _dying bit to epoll backend
> file: coroutinize the with_file templates
> configure: validate --cook ingredient names
> fix trailing whitespace
> Merge 'Estimate timing overhead, allow failing if it is too high' from Travis Downs
perf_tests: document overhead column and threshold options
perf_tests: add measurement overhead tracking and warnings
perf_tests: remove inline/hot attributes from time_measurement methods
perf_tests: move time_measurement class to implementation file
perf_tests: move perf counters into time_measurement singleton
> rpc: log handler type
> Merge 'Add pre-commit with trailing whitespace hook' from Travis Downs
Add GitHub Actions workflow for pre-commit enforcement
Add pre-commit setup documentation to HACKING.md
Add pre-commit configuration with trailing-whitespace hook
Remove trailing whitespace from source files
> posix-stack: Make internal::posix_connect() resolve exceptions into futures
> sstring: fix npos to be size_t for consistency with std::string
Closesscylladb/scylladb#28954
There was a redundant work in split_fragmented(): value_length_if_fixed() was
called inside the loop (N virtual calls), and no reserve() was done
on the output vector causing repeated reallocations.
This patch reserves the output vector to _dimension and
caches value_length_if_fixed() before the loop.
Additionally, split read_vector_element() into two specialized functions:
read_vector_element_fixed() and read_vector_element_variable(), and hoist
the branch on fixed_len outside the loop in split_fragmented() and
deserialize_loop(). This avoids a conditional branch per element in the
hot path.
Benchmark results (1024-dim float vector, release build, -O3 -flto):
10.34 us -> 7.45 us (1.39x, 28% faster)
Verify that scylla sstable upgrade fails when an sstable has a
corrupted Statistics component digest, and succeeds when the
--ignore-component-digest-mismatch flag is provided.
Add ignore_component_digest_mismatch option to db::config (default false).
When set, sstable loading logs a warning instead of throwing on component
digest mismatches, allowing a node to start up despite corrupted non-vital
components or bugs in digest calculation.
Propagate the config to all production sstable load paths:
- distributed_loader (node startup, upload dir processing)
- storage_service (tablet storage cloning)
- sstables_loader (load-and-stream, download tasks, attach)
- stream_blob (tablet streaming)
Add `ignore_component_digest_mismatch` option to `sstable_open_config`
that logs a warning instead of throwing `malformed_sstable_exception`
on component digest mismatch. This is useful for recovering sstables
with corrupted non-vital components or working around bugs in digest
calculation.
Expose the option in scylla-sstable via the
`--ignore-component-digest-mismatch` flag for the upgrade operation.
Generalize corrupt_sstable() and scrub_validate_corrupted_file() to
accept a component_type parameter, defaulting to Data, so they can be
reused for corrupting other components.
Add tests that verify SSTable component digest validation detects
corruption on load. Each test writes an SSTable, corrupts a specific
component file by flipping a bit, then asserts that reloading the
SSTable throws malformed_sstable_exception with the expected digest
mismatch message.
Add integrity verification for SSTable component files by validating
their CRC32 digests against the expected values stored in Scylla
metadata during SSTable loading.
The following components are validated on load: TOC, Scylla metadata,
CompressionInfo, Statistics, Summary, and Filter.
Add a new reader that wraps file_random_access_reader and computes a
running CRC32 digest over the data as it is read. The digest accumulates
across sequential read_exactly() calls and is reset on seek(), since a
non-sequential read invalidates the running checksum.
One of the performance bottlenecks while deserializing vectors was
per-element virtual dispatch in deserialize(): each of the N elements
went through visit() which switches on ~28 type variants. For a
1024-dimension float vector, that's 1024 redundant type switches when
the element type is the same for all of them.
This patch introduces deserialize_vector_visitor that dispatches on the element
type once for the entire vector, then loops inside the resolved
handler. Simple numeric types (float, int, etc.) call
deserialize_value() directly with no virtual dispatch per element.
String types (ascii, utf8) get a dedicated handler that skips
make_empty() (sstring has no empty_t constructor). Complex types
(list, map, tuple, etc.) fall back to per-element dispatch.
Benchmark results (1024-dim float vector, release build, -O3 -flto):
15.73 us -> 11.70 us (1.34x, 26% faster)
`e4da0afb8d5491bf995cbd1d7a7efb966c79ac34` introduces a protection
against resources that are "made up" of thin air to
`reader_concurrency_semaphore`. If there are more `_resources` than
the `_initial_resources`, it means there is a negative leak, and
`on_internal_error_noexcept` is called. In addition to it,
`_resources` is set to `std::max(_resources, _initial_resources)`.
However, the commit message of `e4da0afb8d5491bf995cbd1d7a7efb966c79ac34`
states the opposite: "The detection also clamps the
_resources to _initial_resources, to prevent any damage".
Before this commit, the protection mechanism doesn't clamp
`_resources` to `_initial_resources` but instead keeps `_resources` high,
possibly even indefinitely growing. This commit changes `std::max` to
`std::min` to make the code behave as intended.
Refs: SCYLLADB-163
Closesscylladb/scylladb#28982
Pre-compute the total size and allocate a single uninitialized sstring
before copying the buffers, following the pattern from Seastar's
read_entire_stream_contiguous(). This avoids iterative reallocation
which is O(n^2) for large responses.
This simple patch adds support for storing the HTTP error description
that Vector Store client receives from vector store. Until now it was
just printed to the log but it was not returned. For this reason it
was not forwarded to the drivers which forced users to access ScyllaDB
server logs to understand what is wrong with Vector Store.
This patch also updates formatter to print the message next to the
error code.
Fixes: VECTOR-189
Previously, all stream-table fixtures in this test file used
scope="function", forcing a fresh table to be created for every test,
slowing down the test a bit (though not much), and discouraging writing
small new tests.
This was a workaround for a DynamoDB quirk (that Alternator doesn't have):
LATEST shard iterators have a time slack and may point slightly before
the true stream head, causing leftover events from a previous test to
appear in the next test's reads.
We fix this by draining the stream inside latest_iterators() and
shards_and_latest_iterators() after obtaining the LATEST iterators:
fetch records in a loop until two consecutive polling rounds both return
empty, guaranteeing the iterators are positioned past all pre-existing
events before the caller writes anything. With this guarantee in place,
all stream-table fixtures can safely use scope="module".
After this patch, test_streams.py continues to pass on DynamoDB.
On Alternator, the test file's run time went down a bit, from
20.2 seconds to 17.7 seconds.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
In the next patch, we plan to make the fixtures in test_streams.py
shared between tests. Most tests work well with shared tables, but two
(test_streams_trim_horizon and test_streams_starting_sequence_number)
were written to expect a new table with an empty history, and two
other (test_streams_closed_read and test_streams_disabled_stream) want
to disable streaming and would break a shared table.
So this patch we modify these four tests to create their own new table
instead of using a fixture.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
This pull request adds support for calculation and storing CRC32 digests for all SSTable components.
This change replaces plain file_writer with crc32_digest_file_writer for all SSTable components that should be checksummed. The resulting component digests are stored in the sstable structure
and later persisted to disk as part of the Scylla metadata component during writer::consume_end_of_stream.
Several test cases where introduced to verify expected behaviour.
Additionally, this PR adds new rewrite component mechanism for safe sstable component rewriting.
Previously, rewriting an sstable component (e.g., via rewrite_statistics) created a temporary file that was renamed to the final name after sealing. This allowed crash recovery by simply removing the temporary file on startup.
However, with component digests stored in scylla_metadata (#20100),
replacing a component like Statistics requires atomically updating both the component
and scylla_metadata with the new digest - impossible with POSIX rename.
The new mechanism creates a clone sstable with a fresh generation:
- Hard-links all components from the source except the component being rewritten and scylla_metadata
- Copies original sstable components pointer and recognized components from the source
- Invokes a modifier callback to adjust the new sstable before rewriting
- Writes the modified component along with updated scylla_metadata containing the new digest
- Seals the new sstable with a temporary TOC
- Replaces the old sstable atomically, the same way as it is done in compaction
This is built on the rewrite_sstables compaction framework to support batch operations (e.g., following incremental repair).
In case of any failure durning the whole process, sstable will be automatically deleted on the node startup due to
temporary toc persistence.
Backport is not required, it is a new feature
Fixes https://github.com/scylladb/scylladb/issues/20100, https://github.com/scylladb/scylladb/issues/27453Closesscylladb/scylladb#28338
* github.com:scylladb/scylladb:
docs: document components_digests subcomponent and trailing digest in Scylla.db
sstable_compaction_test: Add tests for perform_component_rewrite
sstable_test: add verification testcases of SSTable components digests persistance
sstables: store digest of all sstable components in scylla metadata
sstables: replace rewrite_statistics with new rewrite component mechanism
sstables: add new rewrite component mechanism for safe sstable component rewriting
compaction: add compaction_group_view method to specify sstable version
sstables: add null_data_sink and serialized_checksum for checksum-only calculation
sstables: extract default write open flags into a constant
sstables: Add write_simple_with_digest for component checksumming
sstables: Extract file writer closing logic into separate methods
sstables: Implement CRC32 digest-only writer
Currently, status_helper::tablets, which keeps a vector of processed
tablet ids, is used only in tablet_virtual_task::get_status_helper,
so there is no point in returning it. Also, in get_status_helper,
it is used only to determine if any tablets are processed.
Remove status_helper::tablets. Use a flag instead of the vector
in get_status_helper.
Currently, for repair tasks tablet_virtual_task::wait gathers the
ids of tablets that are to be repaired. The gathered set is later
used to check if the repair is still ongoing.
However, if the tablets are resized (split or merged), the gathered
set becomes irrelevant. Those, we may end up with invalid tablet id
error being thrown.
Wait until repair is done for all tablets in the table.
There is a race condition in driver that raises the RuntimeException.
This pollutes the output, so this PR is just silencing this exception.
Fixes: SCYLLADB-900
Closesscylladb/scylladb#28957
result_message::prepared now owns a strong pinned prepared-cache entry instead of relying only on a weak pointer view. This closes the remaining lifetime gap after query_processor::prepare() returns, so users of the returned PREPARE message cannot observe an invalidated weak handle during subsequent
processing.
- update result_message::prepared::cql constructor to accept pinned entry
- construct weak view from owned pinned entry inside the message
- pass pinned cache entry from query_processor::prepare() into the message constructor
query_processor::prepare() could race with prepared statement invalidation: after loading from the prepared cache, we converted the cached object to a checked weak pointer and then continued asynchronous work (including error-injection waitpoints). If invalidation happened in that window, the weak
handle could no longer be promoted and the prepare path could fail nondeterministically.
This change keeps a strong cache entry reference alive across the whole critical section in prepare() by using a pinned cache accessor (get_pinned()), and only deriving the weak handle while the entry is pinned. This removes the lifetime gap without adding retry loops.
Test coverage was extended in test/cluster/test_prepare_race.py:
- reproduces the invalidation-during-prepare window with injection,
- verifies prepare completes successfully,
- then invalidates again and executes the same stale client prepared object,
- confirms the driver transparently re-requests/re-prepares and execution succeeds.
This change introduces:
- no behavior change for normal prepare flow besides stronger lifetime guarantees,
- no new protocol semantics,
- preserves existing cache invalidation logic,
- adds explicit cluster-level regression coverage for both the race and driver reprepare path.
- pushes the re prepare operation twards the driver, the server will return unprepared error for the first time and the driver will have to re prepare during execution stage
Fixed several places where ScyllaLogFile.grep() was called without await, resulting in checking coroutine objects for truthiness instead of actual log matches.
Fixes: SCYLLADB-903
No backport, framework fix and one test fix.
Closesscylladb/scylladb#28909
* github.com:scylladb/scylladb:
test.py: fix unawaited ScyllaLogFile.grep() coroutines
tests: fix test_group0_recovers_after_partial_command_application
When dropping a table, make_drop_table_or_view_mutations() creates
a point tombstone in system_schema.columns for every column in the table.
The clustering key of system_schema.columns is (table_name, column_name).
A clustering key with only the table_name component acts as a prefix
tombstone. That tombstone covers all columns belonging to that table.
This approach is already used by make_table_deleting_mutations() during
CREATE TABLE.
Apply the same prefix tombstone approach to DROP TABLE for the columns,
view_virtual_columns, computed_columns, and dropped_columns schema tables.
This reduces tombstone accumulation in schema table sstables.
In test_max_cells test case, which repeatedly creates and drops a table
with 32768 columns, overall test time improved from ~180s to ~157s, which
is ~12.7% improvement.
Refs SCYLLADB-815
Closesscylladb/scylladb#28976
The function checks that the node's state is not left or removed in
gossiper during restart, but with raft topology a removed node will
not be able to contact the cluster to get this information since it will
be banned.
Also remove test_auth_raft_command_split test which is irrelevant since 5ba7d1b116
because it does not use the function that injects max sized command after the
commit.
Schema pull was used by legacy schema code which is not supported for a
long time now and during legacy recovery which is no longer supported as
well. It can be dropped now.
Simplify code by getting rid of group0_upgrade_state since upgrade is no
longer supported, so no need to track its state. The none upgraded node
will simply not boot and to detect that the patch checks the state
directly from the system table.
The only support mode is topology_change_kind::raft, so always set it in
storage_service::join_cluster during join or regular boot. Drop the check
for legacy mode from raft_group0::setup_group0_if_exist since the mode
will not be set at this point any longer. The wrong upgrade will still
be detected in storage_service::join_cluster where topology.upgrade_state
is checked directly.
group0_upgrade_state::recovery is now used only in maintenance mode
so rename the function to indicate it. Also there is no preemption point
in the function any more and it can be a regular function, not a
co-routine.
The test description of refreshing test is very elaborated and it's
worth having it as the description of the streaming scopes test itself.
Callers of the helper can go with smaller descriptions.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
This helper does two things -- sorts sstables per server according to
scope in use and calls sstables_storage.restore(). The code looks better
if the sorting of sstables stays in a helper and the call for .restore()
is moved to the caller.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Now it's possible to replace the whole body of the
test_refresh_with_streaming_scopes() test by calling the corresponding
helper function from backup/restore test module. This helper does
exactly the same, and the SSTablesOnLocalStorage class provides the
necessary save/restore implementations.
One more thing to mention -- the refreshing test for some reason only
wants to run with restored min-tablet-count equal to the original one.
The do_test_streaming_scopes() needs to account for that, as it runs the
tests for more options.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
This class implements some of the sstables manipulations performed by
test_refresh_with_streaming_scopes(). It's here to facilitate next patch
that will use it to call do_test_streaming_scopes() helper.
This patch moves two blocks of code out of the test into this new class.
The shutil.rmtree(tmpbackup) is seemingly lost, but it really isn't --
the tmpbackup variable holds a name of a _subdir_ inside servers'
workdirs. This path doesn't really exist on disk on its own, so removing
it is a no-op.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The class in question performs two operations for
do_test_streaming_scopes(): saves sstables and restores them. Current
caller of the helper is the test_restore_with_streaming_scopes() test
that need to backup sstables on object storage and restore them from
there with the restoration API. The SSTablesOnObjectStorage class does
exactly that.
The change in do_load_sstables() that checks for sstables_storage to be
non None is needed to keep test_refresh_with_streaming_scopes() work --
that test doesn't provide sstables_storage (yet) and the function in
question will call the load_fn callback. Next patch will eliminate it.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The body of this test is duplicated by
test_refresh_with_streaming_scopes() test from other module. Keeping it
in a non-test top-level function will help generalizing these two tests.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
A few days ago, in commit 7b30a3981b we added to pytest.ini the option
xfail_strict. This option causes every time a test XPASSes, i.e., an
xfail test actually passes - to be considered an error and fail the test.
But some tests demonstrate a timing-related bug and do not reproduce the
bug every single time. An example we noticed in one CI run is:
test/cqlpy/test_secondary_index.py::test_unbuilt_index_not_used
This test reproduces a timing-related bug (if you read from a secondary
index "too quickly" you can get wrong results), but only about 90% of the
time, not 100% of the time.
The solution is to add "strict=False" for the xfail marker on this
specific test. This undoes the xfail_strict for this specific test,
accepting that this specific test can either pass or fail. Note that
this does NOT make this test worthless - we still see this test failing
most of the time, and when a developer finally fixes this issue, the
test will begin to pass all the time.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-956
(we'll probably need to follow up this fix with the same fix for other
xfail tests that can sometime pass).
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#28942
When we generate view updates, we check whether we can skip the
entire view update if all columns selected by the view are unmodified.
However, for collection columns, we only check if they were unset
before and after the update.
In this patch we add a check for the actual collection contents.
We perform this check for both virtual and non-virtual selections.
When the column is only a virtual column in the view, it would be
enough to check the liveness of each collection cell, however for
that we'd need to deserialize the entire collection anyway, which
should be effectively as expensive as comparing all of its bytes.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-808Closesscylladb/scylladb#28839
* github.com:scylladb/scylladb:
mv: allow skipping view updates when a collection is unmodified
mv: allow skipping view updates if an empty collection remains unset
Instead of dht::partition_ranges_vector, which is an std::vector<> and
have been seen to cause large allocations when calculating ranges to be
invalidated after compaction:
seastar_memory - oversized allocation: 147456 bytes. This is non-fatal, but could lead to latency and/or fragmentation issues. Please report: at
[Backtrace #0]
void seastar::backtrace<seastar::current_backtrace_tasklocal()::$_0>(seastar::current_backtrace_tasklocal()::$_0&&, bool) at ./build/release/seastar/./seastar/include/seastar/util/backtrace.hh:89
(inlined by) seastar::current_backtrace_tasklocal() at ./build/release/seastar/./seastar/src/util/backtrace.cc:99
seastar::current_tasktrace() at ./build/release/seastar/./seastar/src/util/backtrace.cc:136
seastar::current_backtrace() at ./build/release/seastar/./seastar/src/util/backtrace.cc:169
seastar::memory::cpu_pages::warn_large_allocation(unsigned long) at ./build/release/seastar/./seastar/src/core/memory.cc:840
seastar::memory::cpu_pages::check_large_allocation(unsigned long) at ./build/release/seastar/./seastar/src/core/memory.cc:903
(inlined by) seastar::memory::cpu_pages::allocate_large(unsigned int, bool) at ./build/release/seastar/./seastar/src/core/memory.cc:910
(inlined by) seastar::memory::allocate_large(unsigned long, bool) at ./build/release/seastar/./seastar/src/core/memory.cc:1533
(inlined by) seastar::memory::allocate_slowpath(unsigned long) at ./build/release/seastar/./seastar/src/core/memory.cc:1679
seastar::memory::allocate(unsigned long) at ././seastar/src/core/memory.cc:1698
(inlined by) operator new(unsigned long) at ././seastar/src/core/memory.cc:2440
(inlined by) std::__new_allocator<interval<dht::ring_position>>::allocate(unsigned long, void const*) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/new_allocator.h:151
(inlined by) std::allocator<interval<dht::ring_position>>::allocate(unsigned long) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/allocator.h:203
(inlined by) std::allocator_traits<std::allocator<interval<dht::ring_position>>>::allocate(std::allocator<interval<dht::ring_position>>&, unsigned long) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/alloc_traits.h:614
(inlined by) std::_Vector_base<interval<dht::ring_position>, std::allocator<interval<dht::ring_position>>>::_M_allocate(unsigned long) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/stl_vector.h:387
(inlined by) std::vector<interval<dht::ring_position>, std::allocator<interval<dht::ring_position>>>::reserve(unsigned long) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/vector.tcc:79
dht::to_partition_ranges(utils::chunked_vector<interval<dht::token>, 131072ul> const&, seastar::bool_class<utils::can_yield_tag>) at ./dht/i_partitioner.cc:347
compaction::compaction::get_ranges_for_invalidation(std::vector<seastar::lw_shared_ptr<sstables::sstable>, std::allocator<seastar::lw_shared_ptr<sstables::sstable>>> const&) at ./compaction/compaction.cc:619
(inlined by) compaction::compaction::get_compaction_completion_desc(std::vector<seastar::lw_shared_ptr<sstables::sstable>, std::allocator<seastar::lw_shared_ptr<sstables::sstable>>>, std::vector<seastar::lw_shared_ptr<sstables::sstable>, std::allocator<seastar::lw_shared_ptr<sstables::sstable>>>) at ./compaction/compaction.cc:719
(inlined by) compaction::regular_compaction::replace_remaining_exhausted_sstables() at ./compaction/compaction.cc:1362
compaction::compaction::finish(std::chrono::time_point<db_clock, std::chrono::duration<long, std::ratio<1l, 1000l>>>, std::chrono::time_point<db_clock, std::chrono::duration<long, std::ratio<1l, 1000l>>>) at ./compaction/compaction.cc:1021
compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0::operator()() at ./compaction/compaction.cc:1960
(inlined by) compaction::compaction_result std::__invoke_impl<compaction::compaction_result, compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>(std::__invoke_other, compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/invoke.h:63
(inlined by) std::__invoke_result<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>::type std::__invoke<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>(compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/bits/invoke.h:98
(inlined by) decltype(auto) std::__apply_impl<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0, std::tuple<>>(compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&, std::tuple<>&&, std::integer_sequence<unsigned long, ...>) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/tuple:2920
(inlined by) decltype(auto) std::apply<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0, std::tuple<>>(compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&, std::tuple<>&&) at /usr/lib/gcc/x86_64-redhat-linux/15/../../../../include/c++/15/tuple:2935
(inlined by) seastar::future<compaction::compaction_result> seastar::futurize<compaction::compaction_result>::apply<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>(compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&, std::tuple<>&&) at ././seastar/include/seastar/core/future.hh:1930
(inlined by) seastar::futurize<std::invoke_result<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>::type>::type seastar::async<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>(seastar::thread_attributes, compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&)::'lambda'()::operator()() const at ././seastar/include/seastar/core/thread.hh:267
(inlined by) seastar::noncopyable_function<void ()>::direct_vtable_for<seastar::futurize<std::invoke_result<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>::type>::type seastar::async<compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0>(seastar::thread_attributes, compaction::compaction::run(std::unique_ptr<compaction::compaction, std::default_delete<compaction::compaction>>)::$_0&&)::'lambda'()>::call(seastar::noncopyable_function<void ()> const*) at ././seastar/include/seastar/util/noncopyable_function.hh:138
seastar::noncopyable_function<void ()>::operator()() const at ./build/release/seastar/./seastar/include/seastar/util/noncopyable_function.hh:224
(inlined by) seastar::thread_context::main() at ./build/release/seastar/./seastar/src/core/thread.cc:318
dht::partition_ranges_vector is used on the hot path, so just convert
the problematic user -- cache invalidation -- to use
utils::chunked_vector<dht::partition_range> instead.
Fixes: SCYLLADB-121
Closesscylladb/scylladb#28855
Fixed several places where ScyllaLogFile.grep() was called without
await, resulting in checking coroutine objects for truthiness instead
of actual log matches.
Fixes: SCYLLADB-903
Many tests in test/alternator/test_streams.py use a do_test() function
which performs a user-defined function that runs some write requests,
and then verifies that the expected output appears on the stream.
Because DynamoDB drops do-nothing changes from the stream - such as
writing to an item a value that it already has - these tests need to
write to a different item each time, so do_test() invents a random key
and passes it to the user-defined function to use. But... we had a bug,
the random number generation was done only once, instead of every time.
The fix is to do the random number generation on every call.
We never noticed this bug when each test used a brand new table. But the
next patch will make the tests share the test table, and tests start
to fail. It's especially visible if you run the same test twice against
DynamoDB, e.g.,
test/alternator/run --count 2 --aws \
test_streams.py::test_streams_putitem_keys_only
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
`storage_group_of()` is on the replica-side token lookup hot path but
used `tablet_map::get_tablet_id_and_range_side()`, which computes both
tablet id and post-split range side.
Most callers only need the storage group id. Switch `storage_group_of()`
to use `get_tablet_id()` via `tablet_id_for_token()`, and select the
compaction group via new overloads that compute the range side only
when splitting mode is active.
Change `storage_group::select_compaction_group()` to accept a token
(and tablet_map) and compute the tablet range side only when
splitting_mode() is active.
Add an overload for selecting the compaction group for an sstable
spanning a token range.
Add `tablet_map::get_tablet_range_side(token)` to compute the
post-split range side without computing the tablet id.
Pure addition, no behavior change.
Trie-based sstable indexes are supposed to be (hopefully)
a better default than the old BIG indexes.
Make them the new default.
If we change our mind, this change can be reverted later.
Trie-based indexes and older indexes have a difference in metrics,
and the test uses the metrics to check for bypass cache.
To choose the right metrics, it uses highest_supported_sstable_format,
which is inappropriate, because the sstable format chosen for writes
by Scylla might be different than highest_supported_sstable_format.
Use chosen_sstable_format instead.
Returns the sstable version currently chosen for use in for new sstables.
We are adding it because some tests want to know what format they are
writing (tests using upgradesstable, tests which check stats that only
apply to one of the index types, etc).
(Currently they are using `highest_supported_sstable_format` for this
purpose, which is inappropriate, and will become invalid if a non-latest
format is the default).
cluster.dtest_alternator_tests.test_slow_query_logging performs
a bootstrap with 768 token ranges.
It works with `me` sstables, which have 2 open file descriptors
per open sstable, but with `ms` sstables, which have 3 open
file descriptors per open sstable, it fails with EMFILE.
To avoid this problem, let's just decrease the number of vnodes
for in the test suite. It's appropriate anyway, because it avoids some
unneeded work without weakening the tests.
(Note: pylib-based have been setting `num_tokens` to 16 for a long time too).
This breaks `bypass_cache_test`, which is written in a way that expects
a certain number of token ranges. We adjust the relevant parameter
accordingly.
This function ensures that all alive nodes executed
read barrier.
It will be usefull for the following commits which would
eventually delay returning response to the user until
mutations are applied on other nodes so that the user
may perceive better data consistency accross nodes.
When a user tries to use ALTER TABLE on a materialized view,
the resulting error message is `Cannot use ALTER TABLE on Materialized View`.
The intention behind this error is that ALTER MATERIALIZED VIEW should
be used instead.
But we observed that some users interpret this error message as a general
"You cannot do any ALTER on this thing".
This patch enhances the error message (and others similar to it)
to prevent the confusion.
Closesscylladb/scylladb#28831
tablet_virtual_task::wait throws if a table on which a tablet operation
was working is dropped.
Treat the tablet operation as successful if a table is dropped.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-494
Needs backport to all live releases
Closesscylladb/scylladb#28933
* github.com:scylladb/scylladb:
test: add test_tablet_repair_wait_with_table_drop
service: tasks: return successful status if a table was dropped
SNI works only with DNS hostnames. Adding an IP address causes warnings
on the server side.
This change adds SNI only if it is not an IP address.
This change has no unit tests, as this behavior is not critical,
since it causes a warning on the server side.
The critical part, that the server name is verified, is already covered.
This PR also adds warning logs to improve future troubleshooting of connections to the vector-store nodes.
Fixes: VECTOR-528
Backports to 2025.04 and 2026.01 are required, as these branches are also affected.
Closesscylladb/scylladb#28637
* github.com:scylladb/scylladb:
vector_search: fix TLS server name with IP
vector_search: add warn log for failed ann requests
The first node in the cluster is not guaranteed to be the coordinator
node. Hardcoding node 0 as the coordinator causes test flakiness. This
patch dynamically finds the actual coordinator node and targets it for
error injection, log checking, and restarts.
Additionally, inject `tablet_force_tablet_count_decrease_once` across
all servers to force the tablet merge process to trigger once.
Fixes SCYLLADB-865
Closesscylladb/scylladb#28945
Mention that role and permission changes are durable but may
not be immediately visible on other nodes due to asynchronous
replication.
Fixes: SCYLLADB-651
Closesscylladb/scylladb#28900
Currently, the view_update_generator::mutate_MV function acquires a
reference to the keyspace relevant to the operation, then it calls
max_concurrent_for_each and uses that reference inside the lambda passed
to that function. max_concurrent_for_each can preempt and there is no
mechanism that makes sure that the keyspace is alive until the view
updates are generated, so it is possible that the keyspace is freed by
the time the reference is used.
Fix the issue by precomputing the necessary information based on the
keyspace reference right away, and then passing that information by
value to the other parts of the code. It turns out that we only need to
know whether the keyspace uses tablets and whether it uses a network
topology strategy.
Fixes: scylladb/scylladb#28925Closesscylladb/scylladb#28928
The tests in test_out_of_space_prevention.py are flaky. Three issues contribute:
1. After creating/removing the blob file that simulates disk pressure,
the tests immediately checked derived state (e.g., "compaction_manager
- Drained") without first confirming the disk space monitor had detected
the utilization change. Fix: explicitly wait for "Reached/Dropped below
critical disk utilization level" right after creating/removing the blob
file, before checking downstream effects.
2. Several tests called `manager.driver_connect()` or omitted reconnection
entirely after `server_restart()` / `server_start()`. The pre-existing
driver session can silently reconnect multiple times, causing subsequent
CQL queries to fail. Fix: call `reconnect_driver()` after every node restart.
Additionally, call `wait_for_cql_and_get_hosts()` where CQL is used afterward,
to ensure all connection pools are established.
3. Some log assertions used marks captured before a restart, so they could
match pre-restart messages or miss messages emitted in the correct post-restart
window. Fix: refresh marks at the right points.
Apart from that, the patch fixes a typo: autotoogle -> autotoggle.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-655Closesscylladb/scylladb#28626
Fixes: SCYLLADB-915
Test was quite broken; Not waiting for coro:s, as well as a bunch
of checks no longer even close to valid (this is a ported dtest, and
not a very good one).
Closesscylladb/scylladb#28887
vector_search: small improvements
This PR addresses several minor code quality issues and style inconsistencies within the vector_search module.
No backport is needed as these improvements are not visible to the end user.
Closesscylladb/scylladb#28718
* github.com:scylladb/scylladb:
vector_search: fix names of private members
vector_search: remove unused global variable
Tests use `start_writes` as a simple write workload to test that writes
succeed when they should (e.g., there is no availability loss), but not to
test performance. There is no reason to overload the CPU, which can lead to
test failures.
I suspect this function to be the cause of SCYLLADB-929, where the failures
of `test_raft_recovery_user_data` (that creates multiple write workloads
with `start_writes`) indicated that the machine was overloaded.
The relevant observations:
- two runs failed at the same time in debug mode,
- there were many reactor stalls and RPC timeouts in the logs (leading to
unexpected events like servers marking each other down and group0
leader changes).
I didn't prove that `start_writes` really caused this, but adding this sleep
should be a good change, even if I'm wrong.
The number of writes performed by the test decreases 30-50 times with the
sleep.
Note that some other util functions like `start_writes_to_cdc_table` have
such a sleep.
This PR also contains some minor updates to `test_raft_recovery_user_data`.
Fixes SCYLLADB-929
No backport:
- the failures were observed only in master CI,
- no proof that the change fixes the issue, so backports could be a waste
of time.
Closesscylladb/scylladb#28917
* github.com:scylladb/scylladb:
test: test_raft_recovery_user_data: replace asyncio.gather with gather_safely
test: test_raft_recovery_user_data: use the exclude_node API
test: test_raft_recovery_user_data: drop tablet_load_stats_cfg
test: cluster: util: sleep for 0.01s between writes in do_writes
This patch adds a test file, test/alternator/test_encoding.py, testing
how Alternator stores its data in the underlying CQL database. We test
how tables are named, and how attributes of different types are encoded
into CQL.
The test, which begins with a long comment, also doubles as developer-
oriented *documention* on how Alternator encodes its data in the CQL
database. This documentation is not intended for end-users - we do
not want to officially support reading or writing Alternator tables
through CQL. But once in a while, this information can come in handy
for developers.
More importantly, this test will also serve as a regression test,
verifying that Alternator's encoding doesn't change unintentionally.
If we make an unintentional change to the way that Alternator stores
its data, this can break upgrades: The new code might not be able to
read or write the old table with its old encoding. So it's important
to make sure we never make such unintentional changes to the encoding
of Alternator's data. If we ever do make *intentional* changes to
Alternator's data encoding, we will need to fix the relevant test;
But also not forget to make sure that the new code is able to read
the old encoding as well.
The new tests use both "dynamodb" (Alternator) and "cql" fixtures,
to test how CQL sees the Alternator tables. So naturally are these
tests are marked "scylla_only" and skipped on DynamoDB.
Fixes#19770.
Closesscylladb/scylladb#28866
The motivations for this patch are as follows:
- Guardrails should follow similar conventions, e.g. for config names,
metrics names, testing. Keeping guardrails together makes it easier
to find and compare existing guardrails when new guardrails are
implemented.
- The configuration is used to auto-generate the documentation
(particularly, the `configuration-parameters` page). Currently,
the order of parameters in the documentation is inconsistent (e.g.
`minimum_replication_factor_fail_threshold` before
`minimum_replication_factor_warn_threshold` but
`maximum_replication_factor_fail_threshold` after
`maximum_replication_factor_warn_threshold`), which can be confusing
to customers.
Fixes: SCYLLADB-256
Closesscylladb/scylladb#28932
This series improves timeout handling consistency across the test framework and makes build-mode effects explicit in LWT tests. (starting with LWT test that got flaky)
1. Centralize timeout scaling
Introduce scale_timeout(timeout) fixture in runner.py to provide a single, consistent mechanism for scaling test timeouts based on build mode.
Previously, timeout adjustments were done in an ad-hoc manner across different helpers and tests. Centralizing the logic:
Ensures consistent behavior across the test suite
Simplifies maintenance and reasoning about timeout behavior
Reduces duplication and per-test scaling logic
This becomes increasingly important as tests run on heterogeneous hardware configurations, where different build modes (especially debug) can significantly impact execution time.
2. Make scale_timeout explicit in LWT helpers
Propagate scale_timeout explicitly through BaseLWTTester and Worker, validating it at construction time instead of relying on implicit pytest fixture injection inside helper classes.
Additionally:
Update wait_for_phase_ops() and wait_for_tablet_count() to use scale_timeout_by_mode() for consistent polling behavior across modes
Update all LWT test call sites to pass build_mode explicitly
Increase default timeout values, as the previous defaults were too short and prone to flakiness, particularly under slower configurations such as debug builds
Overall, this series improves determinism, reduces flakiness, and makes the interaction between build mode and test timing explicit and maintainable.
backport: not required just an enhansment for test.py infra
Closesscylladb/scylladb#28840
* https://github.com/scylladb/scylladb:
test/auth_cluster: align service-level timeout expectations with scaled config
test/lwt: propagate scale_timeout through LWT helpers; scale resize waits Pass scale_timeout explicitly through BaseLWTTester and Worker, validating it at construction time instead of relying on implicit pytest fixture injection inside helper classes. Update wait_for_phase_ops() and wait_for_tablet_count() to use scale_timeout_by_mode() so polling behavior remains consistent across build modes. Adjust LWT test call sites to pass scale_timeout explicitly. Increase default timeout values, as the previous defaults were too short and prone to flakiness under slower configurations (notably debug/dev builds).
test/pylib: introduce scale_timeout fixture helper
This patch fixes 2 issues within strong consistency state machine:
- it might happen that apply is called before the schema is delivered to the node
- on the other hand, the apply may be called after the schema was changed and purged from the schema registry
The first problem is fixed by doing `group0.read_barrier()` before applying the mutations.
The second one is solved by upgrading the mutations using column mappings in case the version of the mutations' schema is older.
Fixes SCYLLADB-428
Strong consistency is in experimental phase, no need to backport.
Closesscylladb/scylladb#28546
* https://github.com/scylladb/scylladb:
test/cluster/test_strong_consistency: add reproducer for old schema during apply
test/cluster/test_strong_consistency: add reproducer for missing schema during apply
test/cluster/test_strong_consistency: extract common function
raft_group_registry: allow to drop append entries requests for specific raft group
strong_consistency/state_machine: find and hold schemas of applying mutations
strong_consistency/state_machine: pull necessary dependencies
db/schema_tables: add `get_column_mapping_if_exists()`
`test_proxy_protocol_port_preserved_in_system_clients` failed because it
didn't see the just created connection in system.clients immediately. The
last lines of the stacktrace are:
```
# Complete CQL handshake
await do_cql_handshake(reader, writer)
# Now query system.clients using the driver to see our connection
cql = manager.get_cql()
rows = list(cql.execute(
f"SELECT address, port FROM system.clients WHERE address = '{fake_src_addr}' ALLOW FILTERING"
))
# We should find our connection with the fake source address and port
> assert len(rows) > 0, f"Expected to find connection from {fake_src_addr} in system.clients"
E AssertionError: Expected to find connection from 203.0.113.200 in system.clients
E assert 0 > 0
E + where 0 = len([])
```
Explanation: we first await for the hand-made connection to be completed,
then, via another connection, we're querying system.clients, and we don't
get this hand-made connection in the resultset.
The solution is to replace the bare cql.execute() calls with await wait_for_results(), a helper
that polls via cql.run_async() until the expected row count is reached
(30 s timeout, 100 ms period).
Fixes: SCYLLADB-819
The flaky test is present on master and in previous release, so backporting only there.
Closesscylladb/scylladb#28849
* github.com:scylladb/scylladb:
test_proxy_protocol: introduce extra logging to aid debugging
test_proxy_protocol: fix flaky system.clients visibility checks
Move all ${{ }} expression interpolations into env: blocks so they are
passed as environment variables instead of being expanded directly into
shell scripts. This prevents an attacker from escaping the heredoc in
the Validate Comment Trigger step and executing arbitrary commands on
the runner.
The Verify Org Membership step is hardened in the same way for
defense-in-depth.
Refs: GHSA-9pmq-v59g-8fxp
Fixes: SCYLLADB-954
Closesscylladb/scylladb#28935
When we generate view updates, we check whether we can skip the
entire view update if all columns selected by the view are unmodified.
However, for collection columns, we only check if they were unset
before and after the update.
In this patch we add a check for the actual collection contents.
We perform this check for both virtual and non-virtual selections.
When the column is only a virtual column in the view, it would be
enough to check the liveness of each collection cell, however for
that we'd need to deserialize the entire collection anyway, which
should be effectively as expensive as comparing all of its bytes.
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-808
Currently, when we generate view updates, we skip the view update if
all columns selected by the view are unchanged in the base table update.
However, this does not apply for collection columns - if the base table
has a collection regular column, we never allow skipping generating
view updates and the reason for that is missing implementation.
We can easily relax this for the case where the collection was missing
before and after the update - in this commit we move the check for
collections after the check for missing cells.
The ini-level strict_config was removed/never existed as a config key in pytest 8 — it's only a command-line flag(and back in pytest 9)
In pytest 8.3.5, the equivalent is the --strict-config CLI flag, not an ini option
Fixes SCYLLADB-955
Closesscylladb/scylladb#28939
Document the new `components_digests` subcomponent (tag 12) added to the
Scylla.db metadata component, which stores CRC32 digests of all checksummed
SSTable component files. Also document the trailing CRC32 digest that
stores digest of the scylla metadata itself.
Add two test cases to verify the correctness of the perform_component_rewrite
functionality:
- test_perform_component_rewrite_single_sstable: Tests rewriting the Statistics
component of a single sstable
- test_perform_component_rewrite_multiple_sstables: Tests rewriting 5 out of 10
sstables
Adds a generic test helper that writes a random SSTable, reloads it, and
verifies that the persisted CRC32 digest for each component matches the
digest computed from disk. Those covers all checksummed components test cases.
This change replaces plain file_writer with crc32_digest_file_writer
for all SSTable components that should be checksummed. The resulting component
digests are stored in scylla metadata component.
This also extends new rewrite component mechanism,
to rewrite metadata with updated digest together with the component.
When the local entry with `read_idx` belongs to the current term, it's
safe to update the local `commit_idx` to `read_idx`.
The motivation for this change is to speed up read barriers. `wait_for_apply`
executed at the end of `read_barrier` is delayed until the follower learns
that the entry with `read_idx` is committed. It usually happens quickly in
the `read_quorum` message. However, non-voters don't receive this message,
so they have to wait for `append_entries`. If no new entries are being
added, `append_entries` can come only from `fsm::tick_leader()`. For group0,
this happens once every 100ms.
The issue above significantly slows down cluster setups in tests. Nodes
join group0 as non-voters, and then they are met with several read barriers
just after a write to group0. One example is `global_token_metadata_barrier`
in `write_both_read_new` performed just after `update_topology_state` in
`write_both_read_old`.
I tested the performance impact of this change with the following test:
```python
for _ in range(10):
await manager.servers_add(3)
```
It consistently takes 44-45s with the change and 50-51s without the change
in dev mode.
No backport:
- non-critical performance improvement mostly relevant in tests,
- the change requires some soak time in master.
Closesscylladb/scylladb#28891
* github.com:scylladb/scylladb:
raft: server: fix the repeating typo
raft: clarify the comment about read_barrier_reply
raft: read_barrier: update local commit_idx to read_idx when it's safe
raft: log: clarify the specification of term_for
In case of an error, we want to see the contents of the system.clients
table to have a better understanding of what happened - whether the
row(s) are really missing or maybe they are there, but 1 digit doesn't
match or the row is half-written.
We'll therefore query for the whole table on the CQL side, and then
filter out the rows we want to later proceed with on the python side.
This way we can dump the contents of the whole system.clients table if
something goes south.
`test_proxy_protocol_port_preserved_in_system_clients` failed because it
didn't see the just created connection in system.clients immediately. The
last lines of the stacktrace are:
```
# Complete CQL handshake
await do_cql_handshake(reader, writer)
# Now query system.clients using the driver to see our connection
cql = manager.get_cql()
rows = list(cql.execute(
f"SELECT address, port FROM system.clients WHERE address = '{fake_src_addr}' ALLOW FILTERING"
))
# We should find our connection with the fake source address and port
> assert len(rows) > 0, f"Expected to find connection from {fake_src_addr} in system.clients"
E AssertionError: Expected to find connection from 203.0.113.200 in system.clients
E assert 0 > 0
E + where 0 = len([])
```
Explanation: we first await for the hand-made connection to be completed,
then, via another connection, we're querying system.clients, and we don't
get this hand-made connection in the resultset.
The solution is to replace the bare cql.execute() calls with await wait_for_results(), a helper
that polls via cql.run_async() until the expected row count is reached
(30 s timeout, 100 ms period).
Fixes: SCYLLADB-819
tablet_virtual_task::wait throws if a table on which a tablet operation
was working is dropped.
Treat the tablet operation as successful if a table is dropped.
The one accepts long list of arguments, some of those is not really needed. Also some callers can be relaxed not to provide default values for arguments with such.
Improving tests, not backporting
Closesscylladb/scylladb#28861
* github.com:scylladb/scylladb:
test: Remove passing default "expected_replicas" to check_mutation_replicas()
test: Remove scope and primary-replica-only arguments from check_mutation_replicas() helper
Previous code performed endian conversion by bulk-copying raw bytes
into a std::vector<float> and then iterating over it via a
reinterpret_cast<uint32_t*> pointer. Accessing float storage through a
uint32_t* violates C++ strict aliasing rules, giving the compiler
freedom to reorder or elide the stores, causing undefined behavior.
Replace the two-pass approach with a single-pass loop using
seastar::consume_be<uint32_t>() and std::bit_cast<float>(), which is
both well-defined and auto-vectorizable.
Follow-up #28754Closesscylladb/scylladb#28912
Add support for non_gating, the opposite of gating in dtest terminology, tests in test.py
codebase
This test will/should not be run by any current gating job (ci/next/nightly)
Closesscylladb/scylladb#28902
HostRegistry initialized in several places in the framework, this can
lead to the overlapping IP, even though the possibility is low it's not
zero. This PR makes host registry initialized once for the master
thread and pytest. To avoid communication between with workers, each
worker will get its own subnet that it can use solely for its own goals.
This simplifies the solution while providing the way to avoid overlapping IP's.
Closesscylladb/scylladb#28520
This PR disables running FXAIL tests on ci run to speed it up.
tests will continue run on "nightly" job and FAIL on unexpected pass
and will continue run on "NEXT" job and NOT FAIL on unexpected pass
Closesscylladb/scylladb#28886
The purpose of `add_column_for_post_processing` is to add columns that are required for processing of a query,
but are not part of SELECT clause and shouldn't be returned. They are added to the final result set, but later are not serialized.
Mainly it is used for filtering and grouping columns, with a special case of `WHERE primary_key IN ... ORDER BY ...` when the whole result set needs additional final sorting,
and ordering columns must be added as well.
There was a bug that manifested in #9435, #8100 and was actually identified in #22061.
In case of selection with processing (e.g functions involved), result set row is formed in two stages.
Initially it is a list of columns fetched from replicas - on which filtering and grouping is performed.
After that the actual selection is resolved and the final number of columns can change.
Ordering is performed on this final shape, but the ordering column index returned by `add_column_for_post_processing` refereed to initial shape.
If selection refereed to the same column twice (e.g. `v, TTL(v)` as in #9435) final row was longer than initial and ordering refereed to incorrect column.
If a function in selection refereed to multiple columns (e.g. as_json(.., ..) which #8100 effectively uses) the final row was shorter
and ordering tried to use a non-existing column.
This patch fixes the problem by making sure that column index of the final result set is used for ordering.
The previously crashing test `cassandra_tests/validation/entities/json_test.py::testJsonOrdering` doesn't have to be skipped, but now it is failing on issue #28467.
Fixes#9435Fixes#8100Fixes#22061Closesscylladb/scylladb#28472
Tests use `start_writes` as a simple write workload to test that writes
succeed when they should (e.g., there is no availability loss), but not to
test performance. There is no reason to overload the CPU, which can lead to
test failures.
I suspect this function to be the cause of SCYLLADB-929, where the failures
of `test_raft_recovery_user_data` (that creates multiple write workloads
with `start_writes`) indicated that the machine was overloaded.
The relevant observations:
- two runs failed at the same time in debug mode,
- there were many reactor stalls and RPC timeouts in the logs (leading to
unexpected events like servers marking each other down and group0
leader changes).
I didn't prove that `start_writes` really caused this, but adding this sleep
should be a good change, even if I'm wrong.
The number of writes performed by the test decreases 30-50 times with the
sleep.
Note that some other util functions like `start_writes_to_cdc_table` have
such a sleep.
Fixes SCYLLADB-929
Similar to `raft_drop_incoming_append_entries`, the new error injection
`raft_drop_incoming_append_entries_for_specified_group` skips handler
for `raft_append_entries` RPC but it allows to specify id of raft group
for which the requests should be dropped.
The id of a raft group should be passed in error injection parameters
under `value` key.
It might happen that a strong consistency command will arrive to a node:
- before it knows about the schema
- after the schema was changes and the old version was removed from the
memory
To fix the first case, it's enough to perform a read barrier on group0.
In case of the second one, we can use column mapping the upgrade the
mutation to newer schema.
Also, we should hold pointers to schemas until we finish `_db.apply()`,
so the schema is valid for the whole time.
And potentially we should hold multiple pointers because commands passed
to `state_machine::apply()` may contain mutations to different schema
versions.
This commit relies on a fact that the tablet raft group and its state
machine is created only after the table is created locally on the node.
Fixes SCYLLADB-428
Set enable_schema_commitlog for each group0 tables.
Assert that group0 tables use schema commitlog in ensure_group0_schema
(per each command).
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-914.
Needs backport to all live releases as all are vulnerable
Closesscylladb/scylladb#28876
* github.com:scylladb/scylladb:
test: add test_group0_tables_use_schema_commitlog
db: service: remove group0 tables from schema commitlog schema initializer
service: ensure that tables updated via group0 use schema commitlog
db: schema: remove set_is_group0_table param
Consider this:
- repair takes the lock holder
- tablet merge filber destories the compaction group and the compaction state
- repair fails
- repair destroy the lock holder
This is observed in the test:
```
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036] Repair 1 out of 1 tablets: table=sec_index.users range=(432345564227567615,504403158265495551] replicas=[0e9d51a5-9c99-4d6e-b9db-ad36a148b0ea:15, 498e354c-1254-4d8d-a565-2f5c6523845a:9, 5208598c-84f0-4526-bb7f-573728592172:28]
...
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036]: Started to repair 1 out of 1 tables in keyspace=sec_index, table=users, table_id=ea2072d0-ccd9-11f0-8dba-c5ab01bffb77, repair_reason=repair
repair - Enable incremental repair for table=sec_index.users range=(432345564227567615,504403158265495551]
table - Disabled compaction for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Got unrepaired compaction and repair lock for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Disabled compaction for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Got unrepaired compaction and repair lock for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036]: get_sync_boundary: got error from node=0e9d51a5-9c99-4d6e-b9db-ad36a148b0ea, keyspace=sec_index, table=users, range=(432345564227567615,504403158265495551], error=seastar::rpc::remote_verb_error (Compaction state for table [0x60f008fa34c0] not found)
compaction_manager - Stopping 1 tasks for 1 ongoing compactions for table sec_index.users compaction_group=238 due to tablet merge
compaction_manager - Stopping 1 tasks for 1 ongoing compactions for table sec_index.users compaction_group=238 due to tablet merge
....
scylla[10793] Segmentation fault on shard 28, in scheduling group streaming
```
The rwlock in compaction_state could be destroyed before the lock holder
of the rwlock is destroyed. This causes user after free when the lock
the holder is destroyed.
To fix it, users of repair lock will now be waited when a compaction
group is being stopped.
That way, compaction group - which controls the lifetime of rwlock -
cannot be destroyed while the lock is held.
Additionally, the merge completion fiber - that might remove groups -
is properly serialized with incremental repair.
The issue can be reproduced using sanitize build consistently and can not
be reproduced after the fix.
Fixes#27365Closesscylladb/scylladb#28823
* github.com:scylladb/scylladb:
repair: Fix rwlock in compaction_state and lock holder lifecycle
repair: Prevent repair lock holder leakage after table drop
The comment could be misleading. It could suggest that the returned index is
already safe to read. That's not necessarily true. The entry with the
returned index could, for example, be dropped by the leader if the leader's
entry with this index had a different term.
When the local entry with `read_idx` belongs to the current term, it's
safe to update the local `commit_idx` to `read_idx`. The argument for
safety is in the new comment above `maybe_update_commit_idx_for_read`.
The motivation for this change is to speed up read barriers. `wait_for_apply`
executed at the end of `read_barrier` is delayed until the follower learns
that the entry with `read_idx` is committed. It usually happens quickly in
the `read_quorum` message. However, non-voters don't receive this message,
so they have to wait for `append_entries`. If no new entries are being
added, `append_entries` can come only from `fsm::tick_leader()`. For group0,
this happens once every 100ms.
The issue above significantly slows down cluster setups in tests. Nodes
join group0 as non-voters, and then they are met with several read barriers
just after a write to group0. One example is `global_token_metadata_barrier`
in `write_both_read_new` performed just after `update_topology_state` in
`write_both_read_old`.
Writing a test for this change would be difficult, so we trust the nemesis
tests to do the job. They have already found consistency issues in read
barriers. See #10578.
Both migration manager and system keyspace will be used in next commit.
The first one is needed to execute group0 read barrier and we need
system keyspace to get column mappings.
Use scale_timeout_by_mode() in make_scylla_conf() to derive
request_timeout_in_ms in test/pylib/scylla_cluster.py.
Update test_connections_parameters_auto_update in
test/cluster/auth_cluster/test_raft_service_levels.py to expect the
mode-specific timeout string returned by the REST endpoint after this
scaling change.
Pass scale_timeout explicitly through BaseLWTTester and Worker, validating it at construction time instead of relying on implicit pytest fixture injection inside helper classes.
Update wait_for_phase_ops() and wait_for_tablet_count() to use scale_timeout_by_mode() so polling behavior remains consistent across build modes.
Adjust LWT test call sites to pass scale_timeout explicitly.
Increase default timeout values, as the previous defaults were too short and prone to flakiness under slower configurations (notably debug/dev builds).
Introduce scale_timeout(mode) to centralize test timeout scaling logic based on build mode, the function will return a callable that will handle the timeout by mode.
This ensures consistent timeout behavior across test helpers and eliminates ad-hoc per-test scaling adjustments.
Centralizing the logic improves maintainability and makes timeout behavior easier to reason about.
This becomes increasingly important as we run tests on heterogeneous hardware configurations.
Different build modes (especially debug) can significantly affect execution time, and having a single scaling mechanism helps keep test stability predictable across environments.
No functional change beyond unifying existing timeout scaling behavior.
This commit updates the documentation for the unified installer.
- The Open Source example is replaced with version 2025.1 (Source Available, currently supported, LTS).
- The info about CentOS 7 is removed (no longer supported).
- Java 8 is removed.
- The example for cassandra-stress is removed (as it was already removed on other installation pages).
Fixes https://github.com/scylladb/scylladb/issues/28150Closesscylladb/scylladb#28152
In scenarios where we want to firsty check if a column mapping exists
and if we don't want do flow control with exception, it is very wasteful
to do
```
if (column_mapping_exists()) {
get_column_mapping();
}
```
especially in a hot path like `state_machine::apply()` becase this will
execute 2 internal queries.
This commit introduces `get_column_mapping_if_exists()` function,
which simply wrapps result of `get_column_mapping()` in optional and
doesn't throw an exception if the mapping doesn't exist.
this commit enables 3 strict pytest options:
strict_config - if any warnings encountered while parsing the pytest section of the configuration file will raise errors.
xfail_strict - if markers not registered in the markers section of the configuration file will raise errors.
strict-markers - if tests marked with @pytest.mark.xfail that actually succeed will by default fail the test suite
and fix errors that occur after enabling these options
Closesscylladb/scylladb#28859
Currently, repair-mode tombstone-gc cannot be used on tables with RF=1. We want to make repair-mode the default for all tablet tables (and more, see https://github.com/scylladb/scylladb/issues/22814), but currently a keyspace created with RF=1 and later altered to RF>1 will end up using timeout-mode tombstone gc. This is because the repair-mode tombstone-gc code relies on repair history to determine the gc-before time for keys/ranges. RF=1 tables cannot run repairs so they will have empty repair history and consequently won't be able to purge tombstones.
This PR solves this by keeping a registry of RF=1 tables and consulting this registry when creating `tombstone_gc_state` objects. If the table is RF=1, tombstone-gc will work as if the table used immediate-mode tombstone-gc. The registry is updated on each replication update. As soon as the table is not RF=1 anymore, the tombstone-gc reverts to the natural repair-mode behaviour.
After this PR, tombstone-gc defaults to repair-mode for all tables, regardless of RF and tablets/vnodes.
Fixes: SCYLLADB-106.
New feature, no backport required.
Closesscylladb/scylladb#22945
* github.com:scylladb/scylladb:
test/{boost,cluster}: add test for tombstone gc mode=repair with RF=1
tombstone_gc: allow use of repair-mode for RF=1 tables
replica/table: update rf=1 table registry in shared tombstone-gc state
tombstone_gc: tombstone_gc_before_getter: consider RF when getting gc before time
tombstone_gc: unpack per_table_history_maps
tombstone_gc: extract _group0_gc_time from per_table_history_map
tombstone_gc: drop tombstone_gc_state(nullptr) ctor and operator bool()
test/lib/random_schema: use timeout-mode tombstone_gc
tombstone_gc_options: add C++ friendly constructor
test: move away from tombstone_gc_state(nullptr) ctor
treewide: move away from tombstone_gc_state(nullptr) ctor
sstable: move away from tombstone_gc_mode::operator bool()
replica/table: add get_tombstone_gc_state()
compaction: use tombstone_gc_state with value semantics
db/row_cache: use tombstone_gc_state with value semantics
tombstone_gc: introduce tombstone_gc_state::for_tests()
The test test_mv_merge_allowed asserts in two places that the tablet
count is 2. It does so by calling an async function but, mistakenly, the
returned coroutine was not awaited. The coroutine is, apparently, truthy
so the assertions always passed.
Fix the test to properly await the coroutines in the assertions.
Fixes: SCYLLADB-905
Closesscylladb/scylladb#28875
This PR solves a series of similar problems related to executing
methods on an already aborted `raft::server`. They materialize
in various ways:
* For `add_entry` and `modify_config`, a `raft::not_a_leader` with
a null ID will be returned IF forwarding is disabled. This wasn't
a big problem because forwarding has always been enabled for group0,
but it's something that's nice to fix. It's might be relevant for
strong consistency that will heavily rely on this code.
* For `wait_for_leader` and `wait_for_state_change`, the calls may
hang and never resolve. A more detailed scenario is provided in a
commit message.
For the last two methods, we also extend their descriptions to indicate
the new possible exception type, `raft::stopped_error`. This change is
correct since either we enter the functions and throw the exception
immediately (if the server has already been aborted), or it will be
thrown upon the call to `raft::server::abort`.
We fix both issues. A few reproducer tests have been included to verify
that the calls finish and throw the appropriate errors.
Fixes SCYLLADB-841
Backport: Although the hanging problems haven't been spotted so far
(at least to the best of my knowledge), it's best to avoid
running into a problem like that, so let's backport the
changes to all supported versions. They're small enough.
Closesscylladb/scylladb#28822
* https://github.com/scylladb/scylladb:
raft: Make methods throw stopped_error if server aborted
raft: Throw stopped_error if server aborted
test: raft: Introduce get_default_cluster
Into a single database_apply one. Add three parameters:
* ks_name and cf_name to filter the tables to be affected
* what - what to do: throw or wait
This leads to smaller footprint in the code and improved filtering for
table names at the cost of some extra error injection params in the
tests.
This patch series implements `write_consistency_levels_warned` and `write_consistency_levels_disallowed` guardrails, allowing the configuration of which consistency levels are unwanted for writes. The motivation for these guardrails is to forbid writing with consistency levels that don't provide high durability guarantees (like CL=ANY, ONE, or LOCAL_ONE).
Neither guardrail is enabled by default, so as not to disrupt clusters that are currently using any of the CLs for writes. The warning guardrail may seem harmless, as it only adds a warning to the CQL response; however, enabling it can significantly increase network traffic (as a warning message is added to each response) and also decrease throughput due to additional allocations required to prepare the warning. Therefore, both guardrails should be enabled with care. The newly added `writes_per_consistency_level` metric, which is incremented unconditionally, can help decide whether a guardrail can be safely enabled in an existing cluster.
This commit adds additional `if` instructions on the critical path. However, based on the `perf_simple_query` benchmark for writes, the difference is marginal (~40 additional instructions, which is a relative difference smaller than 0.001).
BEFORE:
```
291443.35 tps ( 53.3 allocs/op, 16.0 logallocs/op, 14.2 tasks/op, 48067 insns/op, 18885 cycles/op, 0 errors)
throughput:
mean= 289743.07 standard-deviation=6075.60
median= 291424.69 median-absolute-deviation=1702.56
maximum=292498.27 minimum=261920.06
instructions_per_op:
mean= 48072.30 standard-deviation=21.15
median= 48074.49 median-absolute-deviation=12.07
maximum=48119.87 minimum=48019.89
cpu_cycles_per_op:
mean= 18884.09 standard-deviation=56.43
median= 18877.33 median-absolute-deviation=14.71
maximum=19155.48 minimum=18821.57
```
AFTER:
```
290108.83 tps ( 53.3 allocs/op, 16.0 logallocs/op, 14.2 tasks/op, 48121 insns/op, 18988 cycles/op, 0 errors)
throughput:
mean= 289105.08 standard-deviation=3626.58
median= 290018.90 median-absolute-deviation=1072.25
maximum=291110.44 minimum=274669.98
instructions_per_op:
mean= 48117.57 standard-deviation=18.58
median= 48114.51 median-absolute-deviation=12.08
maximum=48162.18 minimum=48087.18
cpu_cycles_per_op:
mean= 18953.43 standard-deviation=28.76
median= 18945.82 median-absolute-deviation=20.84
maximum=19023.93 minimum=18916.46
```
Fixes: SCYLLADB-259
Refs: SCYLLADB-739
No backport, it's a new feature
Closesscylladb/scylladb#28570
* github.com:scylladb/scylladb:
scylla.yaml: add write CL guardrails to scylla.yaml
scylla.yaml: reorganize guardrails config to be in one place
test: add cluster tests for write CL guardrails
test: implement test_guardrail_write_consistency_level
cql3: start using write CL guardrails
cql3/query_processor: implement metrics to track CL of writes
db: cql3/query_processor: add write_consistency_levels enum_sets
config: add write_consistency_levels_* guardrails configuration
The test/alternator/run script currently fails, Scylla fails to boot
complaining that "--alternator-ttl-period-in-seconds" is specified
twice (which is, unfortunately, not allowed). The problem is that
recently we started to set this option in test/cqlpy/run.py, for
CQL's new row-level TTL, so now it is no longer needed in
test/alternator/run - and in fact not allowed and we must remove it.
This patch only affects the script test/alternator/run, and has no
affect on running tests through test.py or Jenkins.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#28868
This test forgot to await its check() calls, which is the pass-condition
of the test. Once the await was added, the test started failing. Turns
out, the test was broken, but this was never discovered, because due to
the missing await, the errors were not propagated.
This patch adds the missing await and fixes the discovered problems:
* Use cql.run_async() instead of cql.execute()
* Fix json path for timestamp
* Missing flush/compact
Fixes: SCYLLADB-911
Closesscylladb/scylladb#28883
Recently we started to rely on the options "--auth-superuser-name"
and "--auth-superuser-salted-password" to ensure that a
cassandra/cassandra user exists for tests - without those options
a default superuser no longer exists.
This broke "test/cqlpy/run --release" for old releases, earlier
than 5.4 (in the enterprise stream, 2024.1 or earlier), because
those old release didn't have this option.
So in this patch we fix the "--release" logic that removes these
options from the command line when running these old versions.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#28894
The helper in question duplicates the functionality of `take_snapshot()` one from the same file. The only difference is that it additionally creates keyspace:table with yet another helper, but that helper is also going to be removed (as continuation of #28600 and #28608)
Enhancing tests, not backporting
Closesscylladb/scylladb#28834
* github.com:scylladb/scylladb:
test_backup: Remove prepare_snapshot_for_backup()
test_backup: Patch test_simple_backup_and_restore to use take_snapshot()
test_backup: Patch backup tests to use take_snapshot()
test_backup: Add helper to take snapshot on a single server
8e9c7397c5 made `rs` a reference, which can
lead to use-after-free. The `normal_nodes` map containing the referenced
value can be destroyed before the last use of `rs` when the topology state
is reloaded after a context switch on some `co_await`. The following move
assignment in `storage_service::topology_state_load` causes this:
```
_topology_state_machine._topology = co_await _sys_ks.local().load_topology_state(tablet_hosts);
```
This issue has been discovered in next-2026.1 CI after queueing the
backport of #28558. `test_truncate_during_topology_change` failed after
ASan reported a heap-use-after-free in
```
co_await _repair.local().bootstrap_with_repair(get_token_metadata_ptr(), rs.ring.value().tokens, session);
```
This test enables `delay_bootstrap_120s`, which makes the bug much more
likely to reproduce, but it could happen elsewhere.
No backport needed, as the only backport of #28558 hasn't been merged yet.
The backport PR will cherry-pick this commit.
Closesscylladb/scylladb#28772
Set enable_schema_commitlog for each group0 tables.
Assert that group0 tables use schema commitlog in ensure_group0_schema
(per each command).
Fixes: https://scylladb.atlassian.net/browse/SCYLLADB-914.
set_is_group0_table takes an enabled flag, based on which it decides
whether it's a group0 table. The method is called only with enabled = true.
Drop the param. For not group0 tables nothing should be set.
After the previous changes in `raft::server::{add_entry, modify_config}`
(cf. SCYLLADB-841), we also go through other methods of `raft::server`
and verify that they handle the aborted state properly.
I found two methods that do not:
(A) `wait_for_leader`
(B) `wait_for_state_change`
What happened before these changes?
In case (A), the dangerous scenario occurred when `_leader_promise` was
empty on entering the function. In that case, we would construct the
promise and wait on the corresponding future. However, if the server
had been already aborted before the call, the future would never
resolve and we'd be effectively stuck.
Case (B) is fully analogous: instead of `_leader_promise`, we'd work
with `_stte_change_promise`.
There's probably a more proper solution to this problem, but since I'm
not familiar with the internal code of Raft, I fix it this way. We can
improve it further in the future.
We provide two simple validation tests. They verify that after aborting
a `raft::server`, the calls:
* do not hang (the tests would time out otherwise),
* throw raft::stopped_error.
Fixes SCYLLADB-841
Before the change, calling `add_entry` or `modify_config` on an already
aborted Raft server could result in an error `not_a_leader` containing
a null server ID. It was possible precisely when forwarding was
disabled in the server configuration.
`not_a_leader` is supposed to return the ID of the current leader,
so that was wrong. Furthermore, the description of the function
specified that if a server is aborted, then it should throw
`stopped_error`.
We fix that issue. A few small reproducer tests were provided to verify
that the functions behave correctly with and without forwarding enabled.
Refs SCYLLADB-841
Two calls in test_client_routes_upgrade were missing `await`,
so they were never actually executed. This caused Python
to emit RuntimeWarning about unawaited coroutines, and more
importantly, the test skipped important verification steps, which
could mask real bugs or cause flakiness.
Additionally, increase 10s timeouts to 60s to avoid flakiness in slow
environments. Although these tests haven't failed so far, similar
issues have already been observed in other tests with too-short
timeouts.
Fixes: [SCYLLADB-909](https://scylladb.atlassian.net/browse/SCYLLADB-909)
Backport to 2026.1, as the test is also there.
[SCYLLADB-909]: https://scylladb.atlassian.net/browse/SCYLLADB-909?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQClosesscylladb/scylladb#28877
* github.com:scylladb/scylladb:
test: increase timeouts in test_client_routes.py
test: add missing awaits in test_client_routes_upgrade
The main loops iterating over vector components were not vectorized due to:
- "cannot prove it is safe to reorder floating-point operations"
- "Cannot vectorize early exit loop with more than one early exit"
The first issue is fixed with adding `#pragma clang fp contract(fast) reassociate(on)`, which allows compiler to optimize floating point operations.
The second issue is solved by refactoring the operations in the affected loop.
Additionally using float operations instead of double increases throughput and numerical accuracy is not the main consideration in vector search scenarios.
Performance measured:
- scylla built using dbuild
- using https://github.com/zilliztech/VectorDBBench (modified to call `SELECT id, similarity_cosine({vector<float, 1536>}, {vector<float, 1536>}) ...` without ANN search):
- client concurrency 20
before: ~2250 QPS
`float` operations: ~2350 QPS
`compute_cosine_similarity` vectorization: ~2500QPS
`extract_float_vector` vectorization: ~3000QPS
Follow-up https://github.com/scylladb/scylladb/pull/28615
Ref https://scylladb.atlassian.net/browse/SCYLLADB-764Closesscylladb/scylladb#28754
Increase 10s timeouts to 60s to avoid flakiness in slow
environments. Although these tests haven't failed so far, similar
issues have already been observed in other tests with too-short
timeouts.
Test execution time is unaffected; the entire suite in `dev` takes ~30s
before and after this change.
Two calls in test_client_routes_upgrade were missing `await`,
so they were never actually executed. This caused Python
to emit RuntimeWarning about unawaited coroutines, and more
importantly, the test skipped important verification steps, which
could mask real bugs or cause flakiness.
Fixes: SCYLLADB-909
The comparator used to sort per-IP client rows was not a strict-weak-ordering (it could return true in both directions for some pairs), which makes `std::ranges::sort` behavior undefined. A concrete pair that breaks it (and is realistic in system.clients):
a = (port=9042, client_type="cql")
b = (port=10000, client_type="alternator")
With the current comparator:
cmp(a,b) = (9042 < 10000) || ("cql" < "alternator") = true || false = true
cmp(b,a) = (10000 < 9042) || ("alternator" < "cql") = false || true = true
So both directions are true, meaning there is no valid ordering that sort can achieve.
The fix is to sort lexicographically by (port, client_type) to match the table's clustering key and ensure deterministic ordering.
Closesscylladb/scylladb#28844
This series closes a gap in the approx_exponential_histogram implementation to
cover integer values starting from small Min values.
While the original implementation was focused on durations, where this limitation
was not an issue, over time, there has been a growing need for histograms that
cover smaller values, such as the number of SSTables or the number of items in a
batch.
The reason for the original limitation is inherent to the exponential histogram
math. The previous code required Min to be at least Precision to avoid negative
bit shifts in the exponential calculations.
After this series, approx_exponential_histogram allows Min to be smaller than
Precision by scaling values during indexing. The value is shifted left by
log2 Precision minus log2 Min or zero whichever is larger, and the existing
exponential math is applied. Bucket limits are then scaled back to the original
units. This keeps insertion and retrieval O(1) without runtime branching, at the
cost of repeated bucket limits for some values in the Min to Precision range.
Additional tests cover the new behavior.
Relates to #2785
** New feature, no need to backport. **
Closesscylladb/scylladb#28371
* github.com:scylladb/scylladb:
estimated_histogram_test.cc: add to_metrics_histogram test
histogram_metrics_helper.hh: Support Min < Precision
estimated_histogram_test.cc: Add tests for approx_exponential_histogram with Min<Precision
estimated_histogram.hh: support Min less than Precision histograms
`isclose` function checks if returned similarity floats are close enough to expected value, but it doesn't `assert` by itself.
Several tests missed that `assert`, effectively always passing.
With this patch similarity values checks are wrapped in helper function `assert_similarity` with predefined tolerance.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-877Closesscylladb/scylladb#28748
This patch series removes creation of default 'cassandra:cassandra' superuser on system start.
Disable creation of a superuser with default 'cassandra:cassandra' credentials to improve security. The current flow requires clients to create another superuser and then drop the default `cassandra:cassandra' role. For those who do, there is a time window where the default credentials exist. For those who do not, that role stays. We want to improve security by forcing the client to either use config to specify default values for default superuser name and password or use cqlsh over maintenance socket connection to explicitly create/alter a superuser role.
The patch series:
- Enable role modification over the maintenance socket
- Stop using default 'cassandra' value for default superuser, skipping creation instead
Design document: https://scylladb.atlassian.net/wiki/spaces/RND/pages/165773327/Drop+default+cassandra+superuserFixesscylladb/scylla-enterprise#5657
This is an improvement. It does not need a backport.
Closesscylladb/scylladb#27215
* github.com:scylladb/scylladb:
config: enable maintenance socket in workdir by default
docs: auth: do not specify password with -p option
docs: update documentation related to default superuser
test: maintenance socket role management
test: cluster: add logs to test_maintenance_socket.py
test: pylib: fix connect_driver handling when adding and starting server
auth: do not create default 'cassandra:cassandra' superuser
auth: remove redundant DEFAULT_USER_NAME from password authenticator
auth: enable role management operations via maintenance socket
client_state: add has_superuser method
client_state: add _bypass_auth_checks flag
auth: let maintenance_socket_role_manager know if node is in maintenance mode
auth: remove class registrator usage
auth: instantiate auth service with factory functors
auth: add service constructor with factory functors
auth: add transitional.hh file
service: qos: handle special scheduling group case for maintenance socket
service: qos: use _auth_integration as condition for using _auth_integration
The method is called from storage_proxy::mutate_hint() which is in turn called from hint_mutation::apply_locally(). The latter is either called from directly by hint sender, which already runs in streaming group, or via RPC HINT_MUTATION handler which uses index 1 that negotiates streaming group as well.
To be sure, add a debugging check for current group being the expected one.
Code cleanup, not backporting
Closesscylladb/scylladb#28545
* github.com:scylladb/scylladb:
hint: Don't switch group in database::apply_hint()
hint_sender: Switch to sender group on stop either
This change is a bit more careful, as the test collects files from
snapshot directory several times. Before patching it to use the helper,
it collected _all_ the files. Now the helper only provides TOC-s, but
that's fine -- the only check that relies on that may also re-collect
TOC-s and compare new set with old set.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Some of those tests need to update the hard-coded 'backup' snapshot name
to use the one provided by take_snapshot() helper. Other than that, the
patching is pretty straightforward.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The take_snapshot() helper returns a dict(server: list[string]). When
there's only one server to work with, it's more handy to just get a
single list of sstables.
Next patches will make use of that helper.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Modify the methods which calculate the default gc mode as well as that
which validates whether repair-mode can be used at all, so both accepts
use of repair-mode on RF=1 tables.
This de-facto changes the default tombstone-gc to repair-mode for all
tables. Documentation is updated accordingly.
Some tests need adjusting:
* cqlpy/test_select_from_mutation_fragments.py: disable GC for some test
cases because this patch makes tombstones they write subject to GC
when using defaults.
* test/cluster/test_mv.py::test_mv_tombstone_gc_not_inherited used
repair-mode as a non-default for the base table and expected the MV to
revert to default. Another mode has to be used as the non-default
(immediate).
* test/cqlpy/test_tools.py::test_scylla_sstable_dump_schema: don't
compare tombstone_gc schema extension when comparing dumped schema vs.
original. The tool's schema loader doesn't have access to the keyspace
definition so it will come up with different defaults for
tombstone-gc.
* test/boost/row_cache_test.cc::test_populating_cache_with_expired_and_nonexpired_tombstones
sets tombstone expiry assuming the tombstone-gc timeout-mode default.
Change the CREATE TABLE statement to set the expected mode.
Most of the functionality is tested in cqlpy tests located in
`test_guardrail_write_consistency_level.py`. Add two tests
that require the cluster framework:
- `test_invalid_write_cl_guardrail_config` checks the node startup
path when incorrect `write_consistency_levels_warned` and
`write_consistency_levels_disallowed` values are used.
- `test_write_cl_default` checks the behavior of the default
configuration using a multi-node cluster.
Tests execution time:
- Dev: 10s
- Debug: 18s
Refs: SCYLLADB-259
Implement basic tests for write consistency level guardrails,
verifying that they work for each type of write request (inserts,
updates, deletes, logged batches, unlogged batches, conditional batches,
and counter operations).
All tests are marked as Scylla-only because they currently don't
pass with Cassandra due to differences in handling superusers (see:
SCYLLADB-882).
Tests execution time:
- Dev: 3s
- Debug: 14s
Refs: SCYLLADB-259
Refs: SCYLLADB-882
In this series we introduce new system tables and use them for storing the raft metadata
for strongly consistent tables. In contrast to the previously used raft group0 tables, the
new tables can store data on any shard. The tables also allow specifying the shard where
each partition should reside, which enables the tablets of strongly consistent tables to have
their raft group metadata co-located on the same shard as the tablet replica.
The new tables have almost the same schemas as the raft group0 tables. However, they
have an additional column in their partition keys. The additional column is the shard
that specifies where the data should be located. While a tablet and its corresponding
raft group server resides on some shard, it now writes and reads all requests to the
metadata tables using its shard in addition to the group_id.
The extra partition key column is used by the new partitioner and sharder which allow
this special shard routing. The partitioner encodes the shard in the token and the
sharder decodes the shard from the token. This approach for routing avoids any
additional lookups (for the tablet mapping) during operations on the new tables
and it also doesn't require keeping any state. It also doesn't interact negatively
with resharding - as long as tablets (and their corresponding raft metadata) occupy
some shard, we do not allow starting the node with a shard count lower than the
id of this shard. When increasing the shard count, the routing does not change,
similarly to how tablet allocation doesn't change.
To use the new tables, a new implementation of `raft::persistence` is added. Currently,
it's almost an exact copy of the `raft_sys_table_storage` which just uses the new tables,
but in the future we can modify it with changes specific to metadata (or mutation)
storage for strongly consistent tables. The new storage is used in the `groups_manager`,
which combined with the removal of some `this_shard_id() == 0` checks, allows strongly
consistent tables to be used on all shards.
This approach for making sure that the reads/writes to the new tables end up on the correct shards
won in the balance of complexity/usability/performance against a few other approaches we've considered.
They include:
1. Making the Raft server read/write directly to the database, skipping the sharder, on its shard, while using
the default partitioner/sharder. This approach could let us avoid changing the schema and there should be
no problems for reads and writes performed by the Raft server. However, in this approach we would input
data in tables conflicting with the placement determined by the sharder. As a result, any read going through
the sharder could miss the rows it was supposed to read. Even when reading all shards to find a specific value,
there is a risk of polluting the cache - the rows loaded on incorrect shards may persist in the cache for an unknown
amount of time. The cache may also mistakenly remember that a row is missing, even though it's actually present,
just on an incorrect shard.
Some of the issues with this approach could be worked around using another sharder which always returns
this_shard_id() when asked about a shard. It's not clear how such a sharder would implement a method like
`token_for_next_shard`, and how much simpler it would be compared to the current "identity" sharder.
2. Using a sharder depending on the current allocation of tablets on the node. This approach relies on the
knowledge of group_id -> shard mapping at any point in time in the cluster. For this approach we'd also
need to either add a custom partitioner which encodes the group_id in the token, or we'd need to track the
token(group_id) -> shard mapping. This approach has the benefit over the one used in the series of keeping
the partition key as just group_id. However, it requires more logic, and the access to the live state of the node
in the sharder, and it's not static - the same token may be sharded differently depending on the state of the
node - it shouldn't occur in practice, but if we changed the state of the node before adjusting the table data,
we would be unable to access/fix the stale data without artificially also changing the state of the node.
3. Using metadata tables co-located to the strongly consistent tables. This approach could simplify the
metadata migrations in the future, however it would require additional schema management of all co-located
metadata tables, and it's not even obvious what could be used as the partition key in these tables - some
metadata is per-raft-group, so we couldn't reuse the partition key of the strongly consistent table for it. And
finding and remembering a partition key that is routed to a specific shard is not a simple task. Finally, splits
and merges will most likely need special handling for metadata anyway, so we wouldn't even make use of
co-located table's splits and merges.
Fixes [SCYLLADB-361](https://scylladb.atlassian.net/browse/SCYLLADB-361)
[SCYLLADB-361]: https://scylladb.atlassian.net/browse/SCYLLADB-361?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQClosesscylladb/scylladb#28509
* github.com:scylladb/scylladb:
docs: add strong consistency doc
test/cluster: add tests for strongly-consistent tables' metadata persistence
raft: enable multi-shard raft groups for strongly consistent tablets
test/raft: add unit tests for raft_groups_storage
raft: add raft_groups_storage persistence class
db: add system tables for strongly consistent tables' raft groups
dht: add fixed_shard_partitioner and fixed_shard_sharder
raft: add group_id -> shard mapping to raft_group_registry
schema: add with_sharder overload accepting static_sharder reference
The default 100ms timeout for client readiness in tests is too
aggressive. In some test environments, this is not enough time for
client creation, which involves address resolution and TLS certificate
reading, leading to flaky tests.
This commit increases the default client creation timeout to 10 seconds.
This makes the tests more robust, especially in slower execution
environments, and prevents similar flakiness in other test cases.
Fixes: VECTOR-547, SCYLLADB-802, SCYLLADB-825, SCYLLADB-826
Backport to 2025.4 and 2026.1, as the same problem occurs on these branches and can potentially make the CI flaky there as well.
Closesscylladb/scylladb#28846
* github.com:scylladb/scylladb:
vector_search: test: include ANN error in assertion
vector_search: test: fix HTTPS client test flakiness
When we create a materialized view, we consider 2 cases:
1. the view's primary key contains a column that is not
in the primary key of the base table
2. the view's primary key doesn't contain such a column
In the 2nd case, we add all columns from the base table
to the schema of the view (as virtual columns). As a result,
all of these columns are effectively "selected" in
view_updates::can_skip_view_updates. Same thing happens when
we add new columns to the base table using ALTER.
Because of this, we can never have !column_is_selected and
!has_base_non_pk_columns_in_view_pk at the same time. And
thus, the check (!column_is_selected
&& _base_info.has_base_non_pk_columns_in_view_pk) is always
the same as (!column_is_selected).
Because we immediately return after this check, the tail of
this function is also never reached - all checks after the
(column_is_selected) are affected by this. Also, the condition
(!column_is_selected && base_has_nonexpiring_marker) is always
false at the point it is called. And this in turn makes the
`base_has_nonexpiring_marker` unused, so we delete it as well.
It's worth considering, why did we even have
`base_has_nonexpiring_marker` if it's effectively unused. We
initially introduced it in bd52e05ae2 and we (incorrectly)
used it to allow skipping view updates even if the liveness of
virtual columns changed. Soon after, in 5f85a7a821, we
started categorizing virtual columns as column_is_selected == true
and we moved the liveness checks for virtual columns to the
`if (column_is_selected)` clause, before the `base_has_nonexpiring_marker`
check. We changed this because even if we have a nonexpiring marker
right now, it may be changed in the future, in which case the liveness
of the view row will depend on liveness of the virtual columns and
we'll need to have the view updates from the time the row marker was
nonexpiring.
Closesscylladb/scylladb#28838
Fixes issue #12818 with the following docs changes:
docs/dev/system_keyspace.md: Added missing system tables, added table of contents (TOC), added categories
Closesscylladb/scylladb#27789
For 2025.3 and 2025.4 this test runs order of magnitude
slower in debug mode. Potentially due to passwords::check
running in alien thread and overwhelming the CPU (this is
fixed in newer versions).
Decreasing the number of connections in test makes it fast
again, without breaking reproducibility.
As additional measure we double the timeout.
The fix is now cherry-picked to master as sometimes
test fails there too.
(cherry picked from commit 1f1fc2c2ac)
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-795
backport: 2026.1, already on other stable branches
Closesscylladb/scylladb#28848
* github.com:scylladb/scylladb:
test: add more logs to test_startup_no_auth_response
test: decrease strain in test_startup_response
Enable verification of write consistency level guardrails in
`modification_statement` and `batch_statement`.
Neither guardrail is enabled by default, so as not to disrupt clusters
that are currently using any of the CLs for writes. The warning
guardrail may seem harmless, as it only adds a warning to the CQL
response; however, enabling it can significantly increase network
traffic (as a warning message is added to each response) and also
decrease throughput due to additional allocations required to prepare
the warning. Therefore, both guardrails should be enabled with care.
The newly added `writes_per_consistency_level` metric, which is
incremented unconditionally, can help decide whether a guardrail can
be safely enabled in an existing cluster.
This commit adds additional `if` instructions on the critical path.
However, based on the `perf_simple_query` benchmark for writes,
the difference is marginal (~40 additional instructions, which is
a relative difference smaller than 0.001).
BEFORE:
```
291443.35 tps ( 53.3 allocs/op, 16.0 logallocs/op, 14.2 tasks/op, 48067 insns/op, 18885 cycles/op, 0 errors)
throughput:
mean= 289743.07 standard-deviation=6075.60
median= 291424.69 median-absolute-deviation=1702.56
maximum=292498.27 minimum=261920.06
instructions_per_op:
mean= 48072.30 standard-deviation=21.15
median= 48074.49 median-absolute-deviation=12.07
maximum=48119.87 minimum=48019.89
cpu_cycles_per_op:
mean= 18884.09 standard-deviation=56.43
median= 18877.33 median-absolute-deviation=14.71
maximum=19155.48 minimum=18821.57
```
AFTER:
```
290108.83 tps ( 53.3 allocs/op, 16.0 logallocs/op, 14.2 tasks/op, 48121 insns/op, 18988 cycles/op, 0 errors)
throughput:
mean= 289105.08 standard-deviation=3626.58
median= 290018.90 median-absolute-deviation=1072.25
maximum=291110.44 minimum=274669.98
instructions_per_op:
mean= 48117.57 standard-deviation=18.58
median= 48114.51 median-absolute-deviation=12.08
maximum=48162.18 minimum=48087.18
cpu_cycles_per_op:
mean= 18953.43 standard-deviation=28.76
median= 18945.82 median-absolute-deviation=20.84
maximum=19023.93 minimum=18916.46
```
Fixes: SCYLLADB-259
Consider this:
- repair takes the lock holder
- tablet merge filber destories the compaction group and the compaction state
- repair fails
- repair destroy the lock holder
This is observed in the test:
```
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036] Repair 1 out of 1 tablets: table=sec_index.users range=(432345564227567615,504403158265495551] replicas=[0e9d51a5-9c99-4d6e-b9db-ad36a148b0ea:15, 498e354c-1254-4d8d-a565-2f5c6523845a:9, 5208598c-84f0-4526-bb7f-573728592172:28]
...
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036]: Started to repair 1 out of 1 tables in keyspace=sec_index, table=users, table_id=ea2072d0-ccd9-11f0-8dba-c5ab01bffb77, repair_reason=repair
repair - Enable incremental repair for table=sec_index.users range=(432345564227567615,504403158265495551]
table - Disabled compaction for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Got unrepaired compaction and repair lock for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Disabled compaction for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
table - Got unrepaired compaction and repair lock for range=(432345564227567615,504403158265495551] session_id=a13a72cc-cd2d-11f0-8e9b-76d54580ab09 for incremental repair
repair - repair[5d73d094-72ee-4570-a3cc-1cd479b2a036]: get_sync_boundary: got error from node=0e9d51a5-9c99-4d6e-b9db-ad36a148b0ea, keyspace=sec_index, table=users, range=(432345564227567615,504403158265495551], error=seastar::rpc::remote_verb_error (Compaction state for table [0x60f008fa34c0] not found)
compaction_manager - Stopping 1 tasks for 1 ongoing compactions for table sec_index.users compaction_group=238 due to tablet merge
compaction_manager - Stopping 1 tasks for 1 ongoing compactions for table sec_index.users compaction_group=238 due to tablet merge
....
scylla[10793] Segmentation fault on shard 28, in scheduling group streaming
```
The rwlock in compaction_state could be destroyed before the lock holder
of the rwlock is destroyed. This causes user after free when the lock
the holder is destroyed.
To fix it, users of repair lock will now be waited when a compaction
group is being stopped.
That way, compaction group - which controls the lifetime of rwlock -
cannot be destroyed while the lock is held.
Additionally, the merge completion fiber - that might remove groups -
is properly serialized with incremental repair.
The issue can be reproduced using sanitize build consistently and can not
be reproduced after the fix.
Fixes#27365
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Prevent repair lock holder from being leaked in repair_service when table
is dropped midway.
The leakage might result in use-after-free later, since the repair lock
itself will be gone after table drop.
The RPC verb that removes the lock on success path will not be called
by coordinator after table was dropped.
Refs #27365.
Fixes https://scylladb.atlassian.net/browse/SCYLLADB-896.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
We want to enable maintenance socket by default.
This will prevent users from having to reboot a server to enable it.
Also, there is little point in having maintenance socket that is turned off,
and we want users to use it. After this patch series, they will have
to use it. Note that while config seeding exists, we do not encourage it
for production deployments.
This patch changes default maintenance_socket value from ignore to workdir.
This enables maintenance socket without specifying an explicit path.
Refs SCYLLADB-409
Specifying password with -p option is considered unsafe.
The password will be saved in bash history.
The preferred approach is to enter the password when prompted.
Any approach that passes the password via command line arguments
makes that password visible in process options (ps command), no matter
if the password is passed directly or as an environment variable.
Refs SCYLLADB-409
Update create superuser procedure:
- Remove notes about default `cassandra` superuser
- Add create superuser using existing superuser section
- Update create superuser by using `scylla.yaml` config
- Add create superuser using maintenance socket
Update password reset procedure:
- Add maintenance socket approach
- Remove the old approach with deleting all the roles
Update enabling authentication with downtime and during runtime:
- Mention creating new superuser over the maintenance socket
- Remove default superuser usage
Update enable authorization:
- Mention creating new superuser over the maintenance socket
- Remove mention of default superuser
Reasoning for deletion of the old approach:
- [old] Needs cluster downtime, removes all roles, needs recreation of roles,
needs maintenance socket anyways, if config values are not used for superuser
- [new] No cluster downtime, possibly one node restart to enable maintenance
socket, faster
Refs SCYLLADB-409
Introduce a test that cover:
- Server startup without credentials config seeding with no roles created
- Await maintenance socket role management to be enabled
- `CREATE ROLE`, `ALTER ROLE`, and `DROP ROLE` statement execution success
All the tests in the test_maintenance_socket.py module take 2-3 seconds
to execute.
Explicitly shut down Cluster objects to prevent 'RuntimeError: cannot
schedule new futures after shutdown'.
Refs SCYLLADB-409
Add logs to test_maintenance_socket.py test test_maintenance_socket.
This approach offers additional visibility in case of test failure.
Such logs will be added to new tests in a follow up patch in this
patch series.
Refs SCYLLADB-409
When connect_driver=False, the expected server up state should be
capped to HOST_ID_QUERIED. This is to avoid waiting for CQL readiness,
which requires a superuser to be present.
This logic was only in ScyllaCluster.server_start. ManagerClient.server_add
with start=True and connect_driver=False would still wait for CQL and hang
if no superuser is present. The workaround was to call
ManagerClient.server_add(start=False, connect_driver=False) followed by
ManagerClient.server_start(connect_driver=False).
This patch moves the capping from ScyllaCluster.server_start to
ManagerClient.server_add and ManagerClient.server_start, where connect_driver
is processed. ScyllaCluster only receives the already resolved
expected_server_up_state value.
Refs SCYLLADB-409
Changes the behavior of default superuser creation.
Previously, without configuration 'cassandra:cassandra' credentials
were used. Now default superuser creation is skipped if not configured.
The two ways to create default superuser are:
- Config file - auth_superuser_name and auth_superuser_salted_password fields
- Maintenance socket - connect over maintenance socket and CREATE/ALTER ROLE ...
Behavior changes:
Old behavior:
- No config - 'cassandra:cassandra' created
- auth_superuser_name only - <name>:cassandra created
- auth_superuser_salted_password only - 'cassandra:<password>' created
- Both specified - '<name>:<password>' created
New behavior:
- No config - no default superuser
- Requires maintenance socket setup
- auth_superuser_name only - '<name>:' created WITHOUT password
- Requires maintenance socket setup
- auth_superuser_salted_password only - no default superuser
- Both specified - '<name>:<password>' created
Fixes SCYLLADB-409
Introduce maintenance_socket_authenticator and rework
maintenance_socket_role_manager to support role management operations.
Maintenance auth service uses allow_all_authenticator. To allow
role modification statements over the maintenance socket connections,
we need to treat the maintenance socket connections as superusers and
give them proper access rights.
Possible approaches are:
1. Modify allow_all_authenticator with conditional logic that
password_authenticator already does
2. Modify password_authenticator with conditional logic specific
for the maintenance socket connections
3. Extend password_authenticator, overriding the methods that differ
Option 3 is chosen: maintenance_socket_authenticator extends
password_authenticator with authentication disabled.
The maintenance_socket_role_manager is reworked to lazily create a
standard_role_manager once the node joins the cluster, delegating role
operations to it. In maintenance mode role operations remain disabled.
Refs SCYLLADB-409
Encapsulate the superuser check in client_state so that it
respects _bypass_auth_checks. Connections that bypass auth
(internal callers and the maintenance socket) are always
considered superusers.
Migrate existing call sites from auth::has_superuser(service, user)
to client_state.has_superuser(). Also add _bypass_auth_checks
handling to ensure_not_anonymous().
Refs SCYLLADB-409
Authorization checks were previously skipped based on the
_is_internal flag. This couples two concerns: marking client
state as internal and bypassing authorization.
Introduce _bypass_auth_checks to handle only the authorization
bypass. Internal client state sets it to true, preserving current
behavior. External client state accepts it as a constructor
parameter, defaulting to false.
This will allow maintenance socket connections to skip
authorization without being marked as internal.
Refs SCYLLADB-409
This patch is part of preparations for dropping 'cassandra::cassandra'
default superuser. When that is implemented, maintenance_socket_role_manager
will have two modes of work:
1. in maintenance mode, where role operations are forbidden
2. in normal mode, where role operations are allowed
To execute the role operations, the node has to join a cluster.
In maintenance mode the node does not join a cluster.
This patch lets maintenance_socket_role_manager know if it works under
maintenance mode and returns appropriate error message when role
operations execution is requested.
Refs SCYLLADB-409
This patch removes class registrator usage in auth module.
It is not used after switching to factory functor initialization
of auth service.
Several role manager, authenticator, and authorizer name variables
are returned as well, and hardcoded inside qualified_java_name method,
since that is the only place they are ever used.
Refs SCYLLADB-409
Auth service is instantiated with the constructor that accepts
service_config, which then uses class registrator to instantiate
authorizer, authenticator, and role manager.
This patch switches to instantiating auth service via the constructor
that accepts factory functors. This is a step towards removing
usage of class registrator.
Refs SCYLLADB-409
Auth service can be initialized:
- [current] by passing instantiated authorizer, authenticator, role manager
- [current] by passing service_config, which then uses class registrator to instantiate authorizer, authenticator, role manager
- This approach is easy to use with sharded services
- [new] by passing factory functors which instantiate authorizer, authenticator, role manager
- This approach is also easy to use with sharded services
Refs SCYLLADB-409
In a follow-up patch in this patch series class registrator will be removed.
Adding transitional.hh file will be necessary to expose the authenticator and authorizer.
Refs SCYLLADB-409
service_level_controller has special handling for maintenance socket connections.
If the current user is not a named user, it should use the default scheduling group.
The reason is that the maintenance socket can communicate with Scylla before
auth_integration is registered.
The guard is already present, but it was omitted in get_cached_user_scheduling_group.
This also fixes flakiness in test_maintenance_socket.py tests.
Refs SCYLLADB-409
Maintenance socket connections can be established before _auth_integration is
initialized. The fix introduced with scylladb/scylladb#26856 PR check for
the value of user variable. For maintenance socket connections it will be an
anonymous user, and will fall back to using default scheduling group.
This patch changes the criteria for using default scheduling group from
the user variable to checking the _auth_integration variable itself:
- If _auth_integration is not initialized, use default scheduling group
- If _auth_integration is initialized, let it choose the scheduling group
Refs SCYLLADB-409
Add `write_consistency_levels_disallowed_violations` and
`write_consistency_levels_warned_violations` metrics to track
violations of write_consistency_levels guardrails.
Add `writes_per_consistency_level` to track what CL is used by
writes, regardless of the guardrails configuration.
Data gathered by this metric can be used to decide whether enabling
a particular write consistency level guardrail in a particular
existing cluster is safe.
Refs: SCYLLADB-259
Add enum_sets to query_processor that track the configuration
values of `write_consistency_levels_warned` and
`write_consistency_levels_disallowed`.
Refs: SCYLLADB-259
We introduce a function creating a Raft cluster with parameters usually
used by the tests. This will avoid code duplication, especially after
introducing new tests in the following commits.
Note that the test `test_aborting_wait_for_state_change` has changed:
the previous complex configuration was unnecessary for it (I wrote it).
These two are only used to print into logs on error. However, their
values can be found from previous logs and test execution context.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
When the test fails, the assertion message does not include
the error from the ANN request.
This change enhances the assertion to include the specific ANN error,
making it easier to diagnose test failures.
The default 100ms timeout for client readiness in tests is too
aggressive. In some test environments, this is not enough time for
client creation, which involves address resolution and TLS certificate
reading, leading to flaky tests.
This commit increases the default client creation timeout to 10 seconds.
This makes the tests more robust, especially in slower execution
environments, and prevents similar flakiness in other test cases.
Fixes: VECTOR-547, SCYLLADB-802
On every update of the ERM, update the state of the current table in the
registry of RF=1 tables in shared tombstone gc state.
Ensures that tombstone gc stops collection of tombstones in immediate
mode as soon as the table starts transitioning away from RF=1.
Currently we cannot use repair-mode tombstone gc on RF=1 tables, because
such tables don't need repair and so there won't be repair history to
use to produce gc_before times.
Introduce shared_tombstone_gc_state::_rf_one_tables which will keep a
registry of RF=1 tables. Keeping this up to date is left to outside code
(table.cc). Consult the registry to determine whether a table is RF=1 or
not, so the repair history check can be ellided for rf=1 tables.
Not wired in yet into the table code.
Doesn't belong there. Also, having it as a separate member of
shared_tombstone_gc_state makes updating _group0_gc_time cheaper, as the
update doesn't have to do a copy-mutate-swap of the history maps.
This is the current de-facto default for all tests using random schema
and some are apparently relying on this. Make this explicit to avoid
upsetting tests, by the impending change of this default to repair.
It is ambigous, use the appropriate no-gc or gc-all factories instead,
as appropriate.
A special note for mutation::compacted(): according to the comment above
it, it doesn't drop expired tombstones but as it is currently, it
actually does. Change the tombstone gc param for the underlying call to
compact_for_compaction() to uphold the comment. This is used in tests
mostly, so no fallout expected.
Tests are handled in the next commit, to reduce noise.
Two tests in mutation_test.cc have to be updated:
* test_compactor_range_tombstone_spanning_many_pages
has to be updated in this commit, as it uses
mutation_partition::compact_for_query() as well as
compact_for_query(). The test passes default constructed
tombstone_gc() to the latter while the former now uses no-gc
creating a mismatch in tombstone gc behaviour, resulting in test
failure. Update the test to also pass no-gc to compact_for_query().
* test_query_digest similarly uses mutation_partition::query_mutation()
and another compaction method, having to match the no-gc now used in
query_mutation().
It is ambiguous, use tombstone_gc_mode::is_gc_enabled() instead.
Note that the two has slightly different meanings, operator bool()
returned true when repair-history related functionality was enabled.
This is fine, because the only two users are logs, where the two
meanings are close enough. All other users were eliminated or migrated
already, taking the change in meaning into account.
Instead of keeping a pointer to it. Replace nullptr with
tombstone_gc_state::no_gc().
This object is now designed to be used as a value-type, after recent
refactoring.
To replace the usage of tombstone_gc_state(nullptr) usage in tests
specifically. This more verbose factory method will hopefully convey
that this is not to be used in production code. The nullptr constructor
doesn't convey this and in fact it was used in production code
here-and-there.
This commit introduces pure pytest logging into a file
Previously, test.py used pytest as a script(not a framework) and just captured pytest stdout and logged this data by itself
This commit sets up the log files format that additionaly display Python processName, threadName adn taskName because test.py test cases use them, and now it is so hard to investigate issues that are connected with parallelism inside test case themselve
In addition, commit splits the logging of different pytest workers(xdist) into different files. If pytest workers have ho failed test - log file for these workers will be deleted
There is also additional logging for failures that will contain a separate file per test failure and contain the error itself (stacktrace) and all capture logs from stdout, stderr during the test run. With --save-log-on-success it will be a separate file per test on pass as well
All this new functionality works with the new xdit scheduler (--test-py-init=True)
Fixes SCYLLADB-713
Closesscylladb/scylladb#28705
Introduced a new max_tablet_count tablet option that caps the maximum number of tablets a table can have. This feature is designed primarily for backup and restore workflows.
During backup, when load balancing is disabled for snapshot consistency, the current tablet count is recorded in the backup manifest.
During restore, max_tablet_count is set to this recorded value, ensuring the restored table's tablet count never exceeds the original snapshot's tablet distribution.
This guarantee enables efficient file-based SSTable streaming during restore, as each SSTable remains fully contained within a single tablet boundary.
Closesscylladb/scylladb#28450
For 2025.3 and 2025.4 this test runs order of magnitude
slower in debug mode. Potentially due to passwords::check
running in alien thread and overwhelming the CPU (this is
fixed in newer versions).
Decreasing the number of connections in test makes it fast
again, without breaking reproducibility.
As additional measure we double the timeout.
The fix is now cherry-picked to master as sometimes
test fails there too.
(cherry picked from commit 1f1fc2c2ac)
Previously, rewriting an sstable component (e.g., via rewrite_statistics) created a temporary file that was renamed
to the final name after sealing. This allows crash recovery by simply removing the temporary file on startup.
However, this approach won't work once component digests are stored in scylla_metadata,
as replacing a component like Statistics will require atomically updating both the component
and scylla_metadata with the new digest—impossible with POSIX rename.
The new mechanism creates a clone sstable with a fresh generation:
- Hard-links all components from the source except the component being rewritten and scylla metadata if update_sstable_id is true
- Copies original sstable components pointer and recognized components from the source
- Invokes a modifier callback to adjust the new sstable before rewriting
- Writes the modified component. If update_sstable_id is true, reads scylla metadata, generates new sstable_id and rewrites it.
- Seals the new sstable with a temporary TOC
- Replaces the old sstable atomically, the same way as it is done in compaction
This is built on the rewrite_sstables compaction framework to support batch operations (e.g., following incremental repair).
In case of any failure during the whole process, sstable will be automatically deleted on the node startup due to
temporary toc persistence.
This prepares the infrastructure for component digests. Once digests are introduced in scylla_metadata
this mechanism will be extended to also rewrite scylla metadata with the updated digest alongside the modified component, ensuring atomic updates of both.
Add make_sstable() overload that accepts sstable_version_types parameter
to compaction_group_view interface and all implementations.
This will be useful in rewrite component mechanism, as we
need to preserve sstable version when creating the new one for the replacement.
Introduce a null_data_sink and make_digest_calculator implementation that discards
all writes, enabling checksum calculation without file I/O. This allows the
existing checksummed_file_writer to be used for digest computation
without writing data to disk.
This will be used in a future commit to calculate the scylla metadata
component checksum before writing it to disk, allowing the component
to store its own checksum.
This patch adds filtering for tablet sizes collected in load_stats.
This is needed to improve the chances that the balancer will have all
the tablet sizes for the node, and that way avoid having the node
ignored during balancing.
This patch moves the table size tablet filtering code from a lambda in
storage_service::load_stats_for_tablet_based_tables() to the code
section where it will be used:
tablet_storage_group_manager::table_load_stats()
This is needed to better accomodate the next commit which will add
code for filtering tablets for tablet sizes.
Table sizes are collected in load stats, and are filtered according to
the migration stage, so as to avoid double accounting of tablet sizes.
The comment which explains the cut-off migration stage (cleanup) and
the cut-off stage in the code (streaming) are not the same.
This patch fixes that.
Add a test that exercises to_metrics_histogram when Min is smaller than
Precision.
The test verifies duplicate integer bounds are collapsed,
counts remain cumulative, and native histogram metadata is still present
with the expected schema and min id.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
to_metrics_histogram now collapses duplicate integer bucket bounds
caused by Min less than Precision scaling while always keeping native
histogram metadata.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Add three test cases to verify the hybrid linear/exponential bucketing:
- test_histogram_min_1_bucket_limits: Validates bucket lower limits
- test_histogram_min_1_basic: Tests value insertion and bucket distribution
- test_histogram_min_1_statistics: Tests min(), max(), quantile(), and mean()
- test_histogram_min_2_precision_4: Test min == 2 and precision 4.
These tests cover the new Min<Precision mode with Precision=4, verifying both the
linear range and exponential range.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
approx_exponential_histogram is a pseudo exponential histogram implementation
that can insert and retrieve values into and from buckets in O 1 time.
The implementation uses power of two ranges and splits them linearly into
buckets. The number of buckets per power of two range is called Precision.
The original implementation aimed at covering large value ranges had a
limitation. The histogram Min value had to be greater than or equal to
Precision. As a result code that needs histograms for small integer values
could not use this implementation efficiently.
This change addresses that gap by handling the case where Min is less than
Precision. For Min smaller than Precision the value is scaled by a power of
two factor during indexing so the existing exponential math can be reused
without runtime branching. Bucket limits are scaled back to the original
units which can lead to repeated bucket limits in the Min to Precision
range for integer values.
Example with Min 2 and Precision 4
Buckets 2 2 3 3 4 5 6 7 8 10 12 14 and so on
Implementation details
Introduce SHIFT based on log2 Precision minus log2 Min when positive
Scale Min and Max by SHIFT for all exponential calculations
Compute NUM_BUCKETS using the standard log2 Max over Min formula
Use scaled value in find_bucket_index to avoid fractional bucket steps
Return bucket limits by scaling back to original units
Constraint relaxed from Min greater or equal to Precision to allow any Min
less than Max still power of two
This change maintains backward compatibility with existing histograms
while enabling efficient tracking of small integer values.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Add missing tests for per-table Alternator latency metrics to ensure recent
per-table latency accounting is actually validated.
Changes in this patch:
Refactor latency assertion helper into check_sets_latency_by_metric(),
parameterized by metric name.
Keep existing behavior by implementing check_sets_latency() as a wrapper
over scylla_alternator_op_latency.
Add test_item_latency_per_table() to verify
scylla_alternator_table_op_latency_count increases for:
PutItem, GetItem, DeleteItem, UpdateItem, BatchWriteItem,
and BatchGetItem.
This closes a test gap where only global latency metrics were checked, while
per-table latency metrics were not covered.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Batch operations were updating only global latency histograms, which left
table-level latency metrics incomplete.
This change computes request duration once at the end of each operation and
reuses it to update both global and per-table latency stats:
Latencies are stored per table used,
This aligns batch read/write metric behavior with other operations and improves
per-table observability.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Add a new docs/dev document for the strongly consistent tables feature.
For now, it only contains information about the Raft metadata persistence,
but it should be updated as more of the strong-consistency components are
added.
In this patch we add various tests for checking how strongly consistent
tables work while allowing their tablets to reside on non-0 shards and
while using the new persistent storage for their raft metadata.
The tests verify that:
- strongly consistent tables' tablets can be allocated on different shards
and we can write/read from them
- the raft metadata is persistent across restarts even with disruptions
- the sharder correctly routes metadata queries to specified shards
- we can correctly perform multi-shard reads from the metadata tables
- we can read using just the group_id (without shard) using ALLOW FILTERING
For the tests we add logging to the sharder and partitioner and we add
some extra logs for observability.
In this patch we allow strongly consistent tables to have tablets on
shards different than 0.
For that, we remove the checks for shard 0 for the non-group0 raft
groups, and we allow the tablet allocator to place tablets of
strongly consistent tables on shards different than 0.
We also start using the new storage (raft::persistence) for strongly
consistent tables, added in the preceding commits.
Most functions of the new storage for raft groups for strongly
consistent tables are the same as for the system raft table
storage, so we reuse the tests for them to test the new storage.
We add additional tests for checking the new raft groups partitioner
and sharder, and for verifying that writes using storages for different
shards do not affect the data read on different shards.
We also add a test for checking the snapshot_descriptor present after
the storage bootstrap - for both system and strongly consistent storages
we check that the storage contains the initial descriptor.
Add raft_groups_storage, a raft::persistence implementation for
strongly consistent tablet groups.
Currently, it's almost an exact copy of the raft_sys_table_storage that
uses the new raft tables for strongly consistent tables (raft_groups,
raft_groups_snapshots, raft_groups_snapshot_config) which have
a (shard, group_id) partition key.
In the future, the mutation, term and commit_idx data will be stored
differently for for strongly consistent tables than for group0, which
will differentiate this class from the original raft_sys_table_storage.
The storage is created for each raft group server and it takes a shard
parameter at construction time to ensure all queries target the correct
partition (and thus shard).
Add three new system tables for storing raft state for strongly
consistent tablets, corresponding to the tables for group0:
- system.raft_groups: Stores the raft log, term/vote, snapshot_id,
and commit_idx for each tablet's raft group.
- system.raft_groups_snapshots: Stores snapshot descriptors
(index, term) for each group.
- system.raft_groups_snapshot_config: Stores the raft configuration
(current and previous voters) for each snapshot.
These tables use a (shard, group_id) composite partition key with
the newly added raft_groups_partitioner and raft_groups_sharder, ensuring
data is co-located with the tablet replica that owns the raft group.
The tables are only created when the STRONGLY_CONSISTENT_TABLES experimental
feature is enabled.
Add a custom partitioner and sharder that will be used for Raft tables
for strongly consistent tables. These tables will have partition keys
of the form (shard, group_id) and the partitioner creates tokens that
encode the target shard in the high 16 bits.
Token layout:
[shard: 16 bits][partition key hash: 48 bits]
This encoding guarantees that raft group data will be located on the
same shard as the tablet replica corresponding to that raft group as long
we use the tablet replica's shard as the value in the partition key.
Storing the shard directly in the partition key avoids additional lookups
for request routing to the incoming new raft tables.
For even more simplicity, we avoid biasing between uint64_t and int64_t
by limiting the acceptable shard ids up to 32767 (leaving the top bit 0),
which results in the same value of the token when interpreting either as
uint64_t or int64_t.
The sharder decodes the shard by extracting the high bits, which is
shard-count independent. This allows the partition key:shard mapping
to remain the same even during smp changes (only increases are allowed,
the same limitation as for tablets).
To handle RPC from other nodes, we need to be able to redirect the
requests for each raft group to the shard that owns it. We need to
be able to do the redirection on all shards, so to achieve that, on
all shards we need to store the information about which shard is
occupied by each Raft group server.
For that we add a group_id -> shard mapping to the raft_group_registry.
The mapping is filled out when starting raft servers, it's emptied
when we abort raft servers. We use it when registering RPC verb handlers,
so that regardless of the shard handling the RPC, the work on the raft
group can be performed on the corresponding shard.
SNI works only with DNS hostnames. Adding an IP address causes warnings
on the server side.
This change adds SNI only if it is not an IP address.
This change has no unit tests, as this behavior is not critical,
since it causes a warning on the server side.
The critical part, that the server name is verified, is already covered.
Fixes: VECTOR-528
In order to simplify troubleshooting connection problems, this patch
adds an extra warn log that prints the error for the vector search
request whenever it fails.
in s3::Range class start using s3 global constant for two reasons:
1) uniformity, no need to introduce semantically same constant in each class
2) the value was wrong
Extract the commonly used `open_flags::wo | open_flags::create |
open_flags::exclusive` into a reusable constant
`sstable_write_open_flags` to reduce duplication.
Introduce new methods to write SSTable components while calculating
and returning their CRC32 checksums. This adds:
- make_digests_component_file_writer(): creates a crc32_digest_file_writer
for component writing with checksum tracking
- write_simple_with_digest() and do_write_simple_with_digest(): write
components and return the full checksum value
Introduce template parameter to checksummed file writer to support
digest-only calculation without storing chunk checksums.
This will be needed for future to calculate digest of other components.
Add a schema_builder::with_sharder() overload that accepts a const
reference to dht::static_sharder. This allows schemas to use custom
sharder instances instead of only static sharder configurations.
This is needed to support tables that use custom partitioning and
sharding strategies, such as the incoming raft metadata tables for
strongly consistent tables.
The method is called from storage_proxy::mutate_hint() which is in turn
called from hint_mutation::apply_locally(). The latter is either called
from directly by hint sender, which already runs in streaming group, or
via RPC HINT_MUTATION handler which uses index 1 that negotiates streaming
group as well.
To be sure, add a debugging check for current group being the expected one.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Currently sender only switches group for hints sending on start. It's
worth doing the same on stop too for consistency. There's nothing to
compete with at this point.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
2026-02-09 08:54:51 +03:00
2742 changed files with 21563 additions and 14125 deletions
seastar::metrics::make_histogram("batch_item_count_histogram",seastar::metrics::description("Histogram of the number of items in a batch request"),labels,
seastar::metrics::make_histogram("batch_item_count_histogram",seastar::metrics::description("Histogram of the number of items in a batch request"),labels,
throwexceptions::configuration_exception(format("The experimental feature 'logstor' must be enabled in order to use the 'logstor' storage engine."));
}
if(!db.get_config().enable_logstor()){
throwexceptions::configuration_exception(format("The configuration option 'enable_logstor' must be set to true in the configuration in order to use the 'logstor' storage engine."));
}
}else{
throwexceptions::configuration_exception(format("Illegal value for '{}'",KW_STORAGE_ENGINE));
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
