Resize is no longer only due to avg tablet size. Log avg tablet size as an
information, not the reason, and log the true reason for target tablet
count.
Hints have common meaning for all strategies, so the logic
belongs more to make_sizing_plan().
As a side effect, we can reuse shard capacity computation across
tables, which reduces computational complexity from O(tables*nodes) to
O(tables * DCs + nodes)
The limit is enforced by controlling average per-shard tablet replica
count in a given DC, which is controlled by per-table tablet
count. This is effective in respecting the limit on individual shards
as long as tablet replicas are distributed evenly between shards.
There is no attempt to move tablets around in order to enforce limits
on individual shards in case of imbalance between shards.
If the average per-shard tablet count exceeds the limit, all tables
which contribute to it (have replicas in the DC) are scaled down
by the same factor. Due to rounding up to the nearest power of 2,
we may overshoot the per-shard goal by at most a factor of 2.
If different DCs want different scale factors of a given table, the
lowest scale factor is chosen for a given table.
The limit is configurable. It's a global per-cluster config which
controls how many tablet replicas per shard in total we consider to be
still ok. It controls tablet allocator behavior, when choosing initial
tablet count. Even though it's a per-node config, we don't support
different limits per node. All nodes must have the same value of that
config. It's similar in that regard to other scheduler config items
like tablets_initial_scale_factor and target_tablet_size_in_bytes.
This makes decisions made by the scheduler consistent with decisions
made on table creation, with regard to tablet count.
We want to avoid over-allocation of tablets when table is created,
which would then be reduced by the scheduler's scaling logic. Not just
to avoid wasteful migrations post table creation, but to respect the
per-shard goal. To respect the per-shard goal, the algorithm will no
longer be as simple as looking at hints, and we want to share the
algorithm between the scheduler and initial tablet allocator. So
invoke the scheduler to get the tablet count when table is created.
This is in preparation for using the sizing plan during table creation
where we never have size stats, and hints are the only determining
factor for target tablet count.
Resize plan making will now happen in two stages:
1) Determine desired tablet counts per table (sizing plan)
2) Schedule resize decisions
We need intermediate step in the resize plan making, which gives us
the planned tablet counts, so that we can plug this part of the
algorithm into initial tablet allocation on table construction.
We want decisisons made by the scheduler to be consistent with
decisions made on table creation. We want to avoid over-allocation of
tablets when table is created, which would then be reduced by the
scheduler. Not just to avoid wasteful migrations post table creation,
but to respect the per-shard goal. To respect the per-shard goal, the
algorithm will no longer be as simple as looking at hints, and we want
to share the algorithm between the scheduler and initial tablet
allocator.
Also, this sizing plan will be later plugged into a virtual table for
observability.
Logic is preserved since target tablet size is constant for all
tables.
Dropping d.target_max_tablet_size() will allow us to move it
to the load_balancer scope.
Currently the scale is applied post rounding up of tablet count so
that tablet count per shard is at least 1. In order to be able to use
the scale to increase tablet count per shard, we need to apply it
prior to division by RF, otherwise we will overshoot per-shard tablet
replica count.
Example:
4 nodes, -c1, rf=3, initial_tablets_scale=10
Before: initial_tablet_count=20, tablet-per-shard=15
After: initial_tablet_count=14, tablets-per-shard=10.5
This will result in new tables having at least 10 tablet replicas per
shard by default.
We want this to reduce tablet load imbalance due to differences in
tablet count per shard, where some shards have 1 tablet and some
shards have 2 tablets. With higher tablet count per shard, this
difference-by-one is less relevant.
Fixes#21967
In some tests, we explicity set the initial scale to 1 as some of the
existing tests assume 1 compaction group per shard.
test.py uses a lower default. Having many tablets per shard slows down
certain topology operations like decommission/replace/removenode,
where the running time is proportional to tablet count, not data size,
because constant cost (latency) of migration dominates. This latency
is due to group0 operations and barriers. This is especially
pronounced in debug mode. Scheduler allows at most 2 migrations per
shard, so this latency becomes a determining factor for decommission
speed.
To avoid this problem in tests, we use lower default for tablet count per
shard, 2 in debug/dev mode and 4 in release mode. Alternatively, we
could compensate by allowing more concurrency when migrating small
tablets, but there's no infrastructure for that yet.
I observed that with 10 tablets per shard, debug-mode
topology_custom.mv/test_mv_topology_change starts to time-out during
removenode (30 s).
Currently, we can not have more than one global topology operation at the same time. This means that we can not have concurrent truncate operations because truncate is implemented as a global topology operation.
Truncate excludes with other topology operations, and has to wait for those to complete before truncate starts executing. This can lead to truncate timeouts. In these cases the client retries the truncate operation, which will check for ongoing global topology operations, and will fail with
an "Another global topology request is ongoing, please retry." error.
This can be avoided by truncate checking if the ongoing global topology operation is a truncate running for the same table who's truncate has just been requested again. In this case, we can wait for the ongoing truncate to complete instead of immediately failing the operation, and
provide a better user experience.
This is an improvement, backport is not needed.
Closes#22166Closesscylladb/scylladb#22371
* github.com:scylladb/scylladb:
test: add test for re-cycling ongoing truncate operations
truncate: add additional logging and improve error message during truncate
storage_proxy: wait on already running truncate for the same table
storage_proxy: allow multiple truncate table fibers per shard
Return after executing the global metadata barrier to allow the topology
handler to handle any transitions that might have started by a
concurrect transaction.
Fixes#22792
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
Closesscylladb/scylladb#22793
Refs #22628
Adds exception handler + cleanup for the case where we have a bad config/env vars (hint minio) or similar, such that we fail with exception during setting up the EAR context. In a normal startup, this is ok. We will report the exception, and the do a exit(1).
In tests however, we don't and active context will instead be freed quite proper, in which case we need to call stop to ensure we don't crash on shared pointer destruction on wrong shard. Doing so will hide the real issue from whomever runs the test.
Adds some verbosity to track issues with the network proxy used to test EAR connector difficulties. Also adds an earlier close in input stream to help network usage.
Note: This is a diagnostic helper. Still cannot repro the issue above.
Closesscylladb/scylladb#22810
* github.com:scylladb/scylladb:
gcp/aws kms: Promote service_error to recoverable + use malformed_response_error
encryption_at_rest_test: Add verbosity + earlier stream close to proxy
encryption: Add exception handler to context init (for tests)
Replace boost::accumulate() with the standard library's alternatives
to reduce external dependencies and simplify the codebase. This
change eliminates the requirement for boost::range and makes the
implementation more maintainable.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22856
Disable seastar's built in handlers for SIGINT and SIGTERM and thus
fall-back to the OS's default handlers, which terminate the process.
This makes tool applications interruptable by SIGINT and SIGTERM.
The default handler just terminates the tool app immediately and
doesn't allow for cleanup, but this is fine: the tools have no important
data to save or any critical cleanup to do before exiting.
Fixes: scylladb/scylladb#16954Closesscylladb/scylladb#22838
The config variable `components_memory_reclaim_threshold` limits the
memory available to the sstable bloom filters. Any change to its value
is not immediately propagated to the sstable manager, despite it being
a LiveUpdate variable. The updated value takes effect only when a new
sstable is created or deleted.
This PR first refactors the reclaim and reload logic into a single
background fiber. It then updates the sstable manager to subscribe to
changes in the `components_memory_reclaim_threshold` configuration value
and immediately triggers the reclaim/reload fiber when a change is
detected.
Fixes#21947
This is an improvement and does not need to be backported.
Closesscylladb/scylladb#22725
* github.com:scylladb/scylladb:
sstables_manager: trigger reclaim/reload on `components_memory_reclaim_threshold` update
sstables_manager: maybe_reclaim_components: yield between iterations
sstables_manager: rename `increment_total_reclaimable_memory_and_maybe_reclaim()`
sstables_manager: move reclaim logic into `components_reclaim_reload_fiber()`
sstables_manager: rename `_sstable_deleted_event` condition variable
sstables_manager: rename `components_reloader_fiber()`
sstables_manager: fix `maybe_reclaim_components()` indentation
sstables_manager: reclaim components memory until usage falls below threshold
sstables_manager: introduce `get_components_memory_reclaim_threshold()`
sstables_manager: extract `maybe_reclaim_components()`
sstables_manager: fix `maybe_reload_components()` indentation
sstables_manager: extract out `maybe_reload_components()`
The config variable `components_memory_reclaim_threshold` limits the
memory available to the sstable bloom filters. Any change to its value
is not immediately propagated to the sstable manager, despite it being
a LiveUpdate variable. The updated value takes effect only when a new
sstable is created or deleted.
This patch updates the sstable manager to subscribe to any changes in
the above mentioned config value and immediately trigger the
reclaim/reload fiber when a change occurs. Also, adds a testcase to
verify the fix.
Fixes#21947
Signed-off-by: Lakshmi Narayanan Sreethar <lakshmi.sreethar@scylladb.com>
Refs #22628
Mark problems parsing response (partial message, network error without exception etc
- hello testing), as "malformed_response_error", and promote this as well as
general "service_error" to recoverable exceptions (don't isolate node on error).
This to better handle intermittent network issues as well as making error-testing
more deterministic.
Refs #22628
Adds some verbosity to track issues with the network proxy used to test
EAR connector difficulties. Also adds an earlier close in input stream
to help network usage.
Note: This is a diagnostic helper. Still cannot repro the issue above.
Adds exception handler + cleanup for the case where we have a
bad config/env vars (hint minio) or similar, such that we fail
with exception during setting up the EAR context.
In a normal startup, this is ok. We will report the exception,
and the do a exit(1).
In tests however, we don't and active context will instead be
freed quite proper, in which case we need to call stop to ensure
we don't crash on shared pointer destruction on wrong shard.
Doing so will hide the real issue from whomever runs the test.
Serializing `raft::append_request` for transmission requires approximately the same amount of memory as its size. This means when the Raft library replicates a log item to M servers, the log item is effectively copied M times. To prevent excessive memory usage and potential out-of-memory issues, we limit the total memory consumption of in-flight `raft::append_request` messages.
Fixesscylladb/scylladb#14411Closesscylladb/scylladb#22835
* github.com:scylladb/scylladb:
raft_rpc::send_append_entries: limit memory usage
fms: extract entry_size to log_entry::get_size
Reorder member variable initialization sequence to ensure `pw` is accessed
before being moved. While the current use-after-move warning from clang-tidy
is a false positive, this change:
- Makes the initialization order more logical
- Eliminates misleading static analysis warnings
- Prevents potential future issues if class structure changes
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22830
This patch adds to the fetch_scylla.py script, used by the "--release"
option of test/{cqlpy,alternator}/run, the ability to download the new
2025.1 releases.
In the new single-stream releases, the number looks like the old
Scylla Enterprise releases, but the location of the artifacts in the
S3 bucket look like the old open-source releases (without the word
"-enterprise" in the paths). So this patch introduces a new "if"
for the (major >= 2025) case.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Closesscylladb/scylladb#22778
This change adds two log messages. One for the creation of the truncate
global topology request, and another for the truncate timeout. This is
added in order to help with tracking truncate operation events.
It also extends the "Another global topology request is ongoing, please
retry." error message with more information: keyspace and table name.
Currently, we can not have more than one global topology operation at
the same time. This means that we can not have concurrent truncate
operations because truncate is implemented as a global topology
operation.
Truncate excludes with other topology operations, and has to wait for
those to complete before truncate starts executing. This can lead to
truncate timeouts. In these cases the client retries the truncate operation,
which will check for ongoing global topology operations, and will fail with
an "Another global topology request is ongoing, please retry." error.
This can be avoided by truncate checking if we have a truncate for the same
table already queued. In this case, we can wait for the ongoing truncate to
complete instead of immediatelly failing the operation, and provide a better
user experience.
Both, repair and streaming depend on view builder, but since the builder is started too late, both keep sharded<> reference on it and apply `if (view_builder.local_is_initialized())` safety checks.
However, view builder can do its sharded start much earlier, there's currently nothing that prevents it from doing so. This PR moves view builder start up together with some other of its dependencies, and relaxes the way repair and streaming use their view-builder references, in particular -- removes those ugly initialization checks.
refs: scylladb/scylladb#2737Closesscylladb/scylladb#22676
* github.com:scylladb/scylladb:
streaming: Relax streaming::make_streamig_consumer() view builder arg
streaming: Keep non-sharded view_builder dependency reference
streaming: Remove view_builder.local_is_initialized() checks
repair: Keep non-sharded view_builder dependency reference
repair: Remove view_builder.local_is_initialized() checks
main: Start sharded<view_builder> earlier
test/cql_env: Move stream manager start lower
Replace legacy shell test operator (-o) with more portable OR (||) syntax.
Fix fragile file handling in find loop by using while read loop instead.
Warnings fixed:
- SC2166: Replace [ p -o q ] with [ p ] || [ q ]
- SC2044: Replace for loop over find with while read loop
While no issues were observed with the current code, these changes improve
robustness and portability across different shell environments.
also, set the pipefail option, so that we can catch the unexpected
failure of `find` command call.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22385
It is redundant with reader_permit::impl::_ttl_timer. Use the latter for
TTL of inactive reads too. The usage of the two exclude each other, at
any point in time, either one or the other is used, so no reason to keep
both.
Closesscylladb/scylladb#22863
Demote do-nothing decisions to debug level, but keep them at info
if we did decide to do nothing (such as migrate a tablet). Information
about more major events (like split/merge) is kept at info level.
Once log line that logs node information now also logs the datacenter,
which was previously supplied by a log line that is now debug-only.
Closesscylladb/scylladb#22783
Currently, the tablet repair scheduler repairs all replicas of a tablet. It does not support hosts or DCs selection. It should be enough for most cases. However, users might still want to limit the repair to certain hosts or DCs in production. https://github.com/scylladb/scylladb/pull/21985 added the preparation work to add the config options for the selection. This patch adds the hosts or DCs selection support.
Fixes https://github.com/scylladb/scylladb/issues/22417
New feature. No backport is needed.
Closesscylladb/scylladb#22621
* github.com:scylladb/scylladb:
test: add test to check dcs and hosts repair filter
test: add repair dc selection to test_tablet_metadata_persistence
repair: Introduce Host and DC filter support
docs: locator: update the docs and formatter of tablet_task_info
result_set_row is a heavyweight object containing multiple cell types:
regular columns, partition keys, and static values. To prevent expensive
accidental copies, delete the copy constructor and replace it with:
1. A move constructor for efficient vector reallocation
2. An explicit copy() method when copies are actually needed
This change reduces overhead in some non-hot paths by eliminating implicit
deep copies. Please note, previously, in `create_view_from_mutation()`,
we kept a copy of `result_set_row`, and then reused `table_rs` for
holding the mutation for `scylla_tables`. Because we don't copy
the `result_set_row` in this change, in order to avoid invalidating
the `row` after reusing `table_rs` in the outer scope, we define a
new `table_rs` shadowing the one in the out scope.
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Closesscylladb/scylladb#22741
The existing test measures latencies of object GET-s. That's nice (though incomplete), but we want to measure upload performance. Here it is.
refs: #22460Closesscylladb/scylladb#22480
* github.com:scylladb/scylladb:
test/perf/s3: Add --part-size-mb option for upload test
test/perf/s3: Add uploading test
test/perf/s3: Some renames not to be download-centric
test/perf/s3: Make object/file name configurable
test/perf/s3: Configure maximum number of sockets
test/perf/s3: Remove parallelizm
s3/client: Make http client connections limit configurable
Given two sets of equivalent types, return the set
intersection.
This is a generic function which adapts to the actual
input type.
A unit test is added.
Closesscylladb/scylladb#22763
this PR is propper(pythonic) chance of commit 288a47f815
Creating an own folder used to be needed for two reasons:
we want a separate test suite, with its own settings
we want to structure tests, e.g. tablets, raft, schema, gossip.
We've been creating many folders recently. However, test suite
infrastructure is expensive in test.py - each suite has its own
pool of servers, concurrency settings and so on.
Make it possible to structure tests without too many suites,
by supporting subfolders within a suite.
As an example, this PR move mv tests into a separate folder
custom test.py lookup also works.
tests can be run as:
1. ./tools/toolchain/dbuild ./test.py --no-gather-metrics --mode=dev topology_custom/mv/tablets/test_mv_tablets_empty_ip
2. ./tools/toolchain/dbuild ./test.py --no-gather-metrics --mode=dev topology_custom/mv/tablets
3. ./tools/toolchain/dbuild ./test.py --no-gather-metrics --mode=dev topology_custom/mv
Fixes https://github.com/scylladb/scylladb/issues/20570Closesscylladb/scylladb#22816
* github.com:scylladb/scylladb:
test.py: move mv tests into a separate folder
test.py: suport subfolders
Seems tox is not used anywhere, so there is no need to have it then.
Especially when it messes with pytest. In some cases it can change the
config dir in pytest run.
Closesscylladb/scylladb#22819
