RAFT_TEST_CASE macro creates 2 test cases, one with random 20% packet
loss named name_drops. The framework makes hard coded assumptions about
leader which doesn't hold well in case of packet losses.
This short term fix disables the packet drop variant of the specified test.
It should be safe to re-enable it once the whole framework is re-worked to
remove these hard coded assumptions.
This PR fixes a bug. Hence we need to backport it.
Fixes: scylladb/scylladb#23816Closesscylladb/scylladb#25489
(cherry picked from commit a0ee5e4b85)
Closesscylladb/scylladb#25526
We're enabling the configuration option `rf_rack_valid_keyspaces`
in all Python test suites. All relevant tests have been adjusted
to work with it enabled.
That encompasses the following suites:
* alternator,
* broadcast_tables,
* cluster (already enabled in scylladb/scylladb@ee96f8dcfc),
* cql,
* cqlpy (already enabled in scylladb/scylladb@be0877ce69),
* nodetool,
* rest_api.
Two remaining suites that use tests written in Python, redis and scylla_gdb,
are not affected, at least not directly.
The redis suite requires creating an instance of Scylla manually, and the tests
don't do anything that could violate the restriction.
The scylla_gdb suite focuses on testing the capabilities of scylla-gdb.py, but
even then it reuses the `run` file from the cqlpy suite.
Fixesscylladb/scylladb#25126Closesscylladb/scylladb#24617
(cherry picked from commit b41151ff1a)
Closesscylladb/scylladb#25229
The flush api could not detect if the node is down and fail the flush
before the timeout. This patch detects if there is down node and skip
the flush if so, since the flush will fail after the timeout in this
case anyway.
The slowness due to the flush timeout in
compaction_test.py::TestCompaction::test_delete_tombstone_gc_node_down
is fixed with this patch.
Fixes#22413Closesscylladb/scylladb#22445
(cherry picked from commit 0682b1c716)
Closesscylladb/scylladb#25433
We're providing additional information in error messages when throwing
an exception related to data corruption: when a segment is truncated
and when it's content is invalid. That might prove helpful when debugging.
Closesscylladb/scylladb#25190
(cherry picked from commit 408b45fa7e)
Closesscylladb/scylladb#25459
This patch sets, for alternator test suite, all 'alternator-*' loggers and 'paxos' logger to trace level. This should significantly ease debugging of failed tests, while it has no effect on test time and increases log size only by 7%.
This affects running alternator tests only with `test.py`, not with `test/alternator/run`.
Closes#24645Closesscylladb/scylladb#25327
(cherry picked from commit eb11485969)
Closesscylladb/scylladb#25381
TRUNCATE TABLE performs a memtable flush and then discards the sstables of the table being truncated. It collects the highest replay position for both of these. When the highest replay position of the discarded sstables is higher than the highest replay position of the flushed memtable, that means that we have had writes during truncate which have been flushed to disk independently of the truncate process. We check for this and trigger an on_internal_error() which throws an exception, informing the user that writing data concurrently with TRUNCATE TABLE is not advised.
The problem with this is that truncate is also called from DROP KEYSPACE and DROP TABLE. These are raft operations and exceptions thrown by them are caught by the (...) exception handler in the raft applier fiber, which then exits leaving the node without the ability to execute subsequent raft commands.
This commit changes the on_internal_error() into a warning log entry. It also outputs to keyspace/table names, and the offending replay positions which caused the check to fail.
This PR also adds a test which validates that TRUNCATE works correctly with concurrent writes. More specifically, it checks that:
- all data written before TRUNCATE starts is deleted
- none of the data after TRUNCATE completes is deleted
Fixes: #25173Fixes: #25013
Backport is needed in versions which check for truncate with concurrent writes using `on_internal_error()`: 2025.3 2025.2 2025.1
- (cherry picked from commit 268ec72dc9)
- (cherry picked from commit 33488ba943)
Parent PR: #25174Closesscylladb/scylladb#25348
* github.com:scylladb/scylladb:
truncate: add test for truncate with concurrent writes
truncate: change check for write during truncate into a log warning
When a node shuts down, in storage service, after storage_proxy RPCs are stopped, some write handlers within storage_proxy may still be waiting for background writes to complete. These handlers hold appropriate ERMs to block schema changes before the write finishes. After the RPCs are stopped, these writes cannot receive the replies anymore.
If, at the same time, there are RPC commands executing `barrier_and_drain`, they may get stuck waiting for these ERM holders to finish, potentially blocking node shutdown until the writes time out.
This change introduces cancellation of all outstanding write handlers from storage_service after the storage proxy RPCs were stopped.
Fixes scylladb/scylladb#23665
Backport: since this fixes an issue that frequently causes issues in CI, backport to 2025.1, 2025.2, and 2025.3.
- (cherry picked from commit bc934827bc)
- (cherry picked from commit e0dc73f52a)
Parent PR: #24714Closesscylladb/scylladb#25168
* github.com:scylladb/scylladb:
storage_service: Cancel all write requests on storage_proxy shutdown
test: Add test for unfinished writes during shutdown and topology change
During a graceful node shutdown, RPC listeners are stopped in `storage_service::drain_on_shutdown`
as one of the first steps. However, even after RPCs are shut down, some write handlers in
`storage_proxy` may still be waiting for background writes to complete. These handlers retain the ERM.
Since the RPC subsystem is no longer active, replies cannot be received, and if any RPC commands are
concurrently executing `barrier_and_drain`, they may get stuck waiting for those writes. This can block
the messaging server shutdown and delay the entire shutdown process until the write timeout occurs.
This change introduces the cancellation of all outstanding write handlers in `storage_proxy`
during shutdown to prevent unnecessary delays.
Fixesscylladb/scylladb#23665
(cherry picked from commit e0dc73f52a)
This test reproduces an issue where a topology change and an ongoing write query
during query coordinator shutdown can cause the node to get stuck.
When a node receives a write request, it creates a write handler that holds
a copy of the current table's ERM (Effective Replication Map). The ERM ensures
that no topology or schema changes occur while the request is being processed.
After the query coordinator receives the required number of replica write ACKs
to satisfy the consistency level (CL), it sends a reply to the client. However,
the write response handler remains alive until all replicas respond — the remaining
writes are handled in the background.
During shutdown, when all network connections are closed, these responses can no longer
be received. As a result, the write response handler is only destroyed once the write
timeout is reached.
This becomes problematic because the ERM held by the handler blocks topology or schema
change commands from executing. Since shutdown waits for these commands to complete,
this can lead to unnecessary delays in node shutdown and restarts, and occasional
test case failures.
Test for: scylladb/scylladb#23665
(cherry picked from commit bc934827bc)
test_validate_truncate_with_concurrent_writes checks if truncate deletes
all the data written before the truncate starts, and does not delete any
data after truncate completes.
(cherry picked from commit 33488ba943)
Currently, repair_service::repair_tablets starts repair if there
is no ongoing tablet operations. The check does not consider global
topology operations, like tablet resize finalization.
Hence, if:
- topology is in the tablet_resize_finalization state;
- repair starts (as there is no tablet transitions) and holds the erm;
- resize finalization finishes;
then the repair sees a topology state different than the actual -
it does not see that the storage groups were already split.
Repair code does not handle this case and it results with
on_internal_error.
Start repair when topology is not busy. The check isn't atomic,
as it's done on a shard 0. Thus, we compare the topology versions
to ensure that the business check is valid.
Fixes: https://github.com/scylladb/scylladb/issues/24195.
Needs backport to all branches since they are affected
- (cherry picked from commit df152d9824)
- (cherry picked from commit 83c9af9670)
Parent PR: #24202Closesscylladb/scylladb#24778
* github.com:scylladb/scylladb:
test: add test for repair and resize finalization
repair: postpone repair until topology is not busy
`kmip_test_helper()` is a utility function to spawn a dedicated PyKMIP
server for a particular Boost test case. The function runs the server as
an external process and uses a thread to parse the port from the
server's logs. The thread communicates the port to the main thread via
a promise.
The current implementation has a bug where the thread may set a value
to the promise after its destruction, causing a segfault. This happens
when the server does not start within 20 seconds, in which case the port
future throws and the stack unwinding machinery destroys the port
promise before the thread that writes to it.
Fix the bug by declaring the promise before the cleanup action.
The bug has been encountered in CI runs on slow machines, where the
PyKMIP server takes too long to create its internal tables (due to slow
fdatasync calls from SQLite). This patch does not improve CI stability -
it only ensures that the error condition is properly reflected in the
test output.
This patch is not a backport. The same bug has been fixed in master as
part of a larger rewrite of the `kmip_test_helper()` (see 722e2bce96).
Refs #24747, #24842.
Fixes#24574.
Signed-off-by: Nikos Dragazis <nikolaos.dragazis@scylladb.com>
Closesscylladb/scylladb#25028
Currently, repair_service::repair_tablets starts repair if there
is no ongoing tablet operations. The check does not consider global
topology operations, like tablet resize finalization. This may cause
a data race and unexpected behavior.
Start repair when topology is not busy.
(cherry picked from commit df152d9824)
Interval map is very susceptible to quadratic space behavior when it's flooded with many entries overlapping all (or most of) intervals, since each such entry will have presence on all intervals it overlaps with.
A trigger we observed was memtable flush storm, which creates many small "L0" sstables that spans roughly the entire token range.
Since we cannot rely on insertion order, solution will be about storing sstables with such wide ranges in a vector (unleveled).
There should be no consequence for single-key reads, since upper layer applies an additional filtering based on token of key being queried.
And for range scans, there can be an increase in memory usage, but not significant because the sstables span an wide range and would have been selected in the combined reader if the range of scan overlaps with them.
Anyway, this is a protection against storm of memtable flushes and shouldn't be the common scenario.
It works both with tablets and vnodes, by adjusting the token range spanned by compaction group accordingly.
Fixes#23634.
We can backport this into 2024.2, 2025.1, but we should let this cook in master for 1 month or so.
- (cherry picked from commit 494ed6b887)
- (cherry picked from commit 59dad2121f)
- (cherry picked from commit 21d1e78457)
- (cherry picked from commit c77f710a0c)
- (cherry picked from commit d5bee4c814)
Parent PR: #23806Closesscylladb/scylladb#24012
* github.com:scylladb/scylladb:
test: Verify partitioned set store split and unsplit correctly
sstables: Fix quadratic space complexity in partitioned_sstable_set
compaction: Wire table_state into make_sstable_set()
compaction: Introduce token_range() to table_state
dht: Add overlap_ratio() for token range
Currently the service levels cache is unnecessarily updated in every
call of `topology_state_load()`.
But it is enough to reload it only when a snapshot is loaded.
(The cache is also already updated when there is a change to one of
`service_levels_v2`, `role_members`, `role_attributes` tables.)
Fixesscylladb/scylladb#25114Fixesscylladb/scylladb#23065Closesscylladb/scylladb#25116
(cherry picked from commit 10214e13bd)
Closesscylladb/scylladb#25303
The PyKMIP server uses an SQLite database to store artifacts such as
encryption keys. By default, SQLite performs a full journal and data
flush to disk on every CREATE TABLE operation. Each operation triggers
three fdatasync(2) calls. If we multiply this by 16, that is the number
of tables created by the server, we get a significant number of file
syncs, which can last for several seconds on slow machines.
This behavior has led to CI stability issues from KMIP unit tests where
the server failed to complete its schema creation within the 20-second
timeout (observed on spider9 and spider11).
Fix this by configuring the server to use an in-memory SQLite.
Fixes#24842.
Signed-off-by: Nikos Dragazis <nikolaos.dragazis@scylladb.com>
Closesscylladb/scylladb#24995
(cherry picked from commit 2656fca504)
Closesscylladb/scylladb#25297
In this PR, we adjust tests in the cqlpy test suite so they
only use RF-rack-valid keyspaces. After that, we enable
the configuration option `rf_rack_valid_keyspaces` in the
suite by default.
Refs scylladb/scylladb#23428
Fixes scylladb/scylladb#25306
Backport: backporting to 2025.1 so we can test the option there too.
- (cherry picked from commit 6bde01bb59)
- (cherry picked from commit 958eaec056)
- (cherry picked from commit a59842257a)
- (cherry picked from commit be0877ce69)
Parent PR: #23489Closesscylladb/scylladb#25307
* github.com:scylladb/scylladb:
test/cqlpy: Enable rf_rack_valid_keyspaces by default
test: Move test_alter_tablet_keyspace_rf to cluster suite
test/cqlpy: Adjust tests to RF-rack-valid keyspaces
test/cqlpy/cassandra_tests: Adjust to RF-rack-valid keyspaces
TRUNCATE TABLE performs a memtable flush and then discards the sstables
of the table being truncated. It collects the highest replay position
for both of these. When the highest replay position of the discarded
sstables is higher than the highest replay position of the flushed
memtable, that means that we have had writes during truncate which have
been flushed to disk independently of the truncate process. We check for
this and trigger an on_internal_error() which throws an exception,
informing the user that writing data concurrently with TRUNCATE TABLE is
not advised.
The problem with this is that truncate is also called from DROP KEYSPACE
and DROP TABLE. These are raft operations and exceptions thrown by them
are caught by the (...) exception handler in the raft applier fiber,
which then exits leaving the node without the ability to execute
subsequent raft commands.
This commit changes the on_internal_error() into a warning log entry. It
also outputs to keyspace/table names, the truncated_at timepoint, the
offending replay positions which caused the check to fail.
Fixes: #25173Fixes: #25013
(cherry picked from commit 268ec72dc9)
Right now, service levels are migrated in one group0 command and auth is migrated in the next one. This has a bad effect on the group0 state reload logic - modifying service levels in group0 causes the effective service levels cache to be recalculated, and to do so we need to fetch information about all roles. If the reload happens after SL upgrade and before auth upgrade, the query for roles will be directed to the legacy auth tables in system_auth - and the query, being a potentially remote query, has a timeout. If the query times out, it will throw an exception which will break the group0 apply fiber and the node will need to be restarted to bring it back to work.
In order to solve this issue, make sure that the service level module does not start populating and using the service level cache until both service levels and auth are migrated to raft. This is achieved by adding the check both to the cache population logic and the effective service level getter - they now look at service level's accessor new method, `can_use_effective_service_level_cache` which takes a look at the auth version.
Fixes: scylladb/scylladb#24963
Should be backported to all versions which support upgrade to topology over raft - the issue described here may put the cluster into a state which is difficult to get out of (group0 apply fiber can break on multiple nodes, which necessitates their restart).
- (cherry picked from commit 2bb800c004)
- (cherry picked from commit 3a082d314c)
Parent PR: #25188Closesscylladb/scylladb#25283
* github.com:scylladb/scylladb:
test: sl: verify that legacy auth is not queried in sl to raft upgrade
qos: don't populate effective service level cache until auth is migrated to raft
All of the tests in the suite have been adjusted so they only
use RF-rack-valid keyspaces, so let's start enabling the option
by default.
(cherry picked from commit be0877ce69)
We move the test `test_alter_tablet_keyspace_rf` from the cqlpy to the
cluster test suite. The reason behind the change is that the test cannot
be run with `rf_rack_valid_keyspaces` turned on in the configuration.
During the test, we make the keyspace RF-rack-invalid multiple times.
Since RF-rack-validity is a very strong constraint, adjust the test
otherwise is impossible.
By moving it to the cluster test suite, we're able to change the
configuration of the node used in the test, and so the test can work
again.
(cherry picked from commit a59842257a)
We adjust three existing Cassandra tests so that they don't create
RF-rack-invalid keyspaces. We modify the replication factor used
in the problematic tests. The changes don't affect the tests as
the value of the RF is unrelated to what they verify. Thanks to
that, we can run them now even with enforced RF-rack-valid keyspaces.
The drawback is that the modified ALTER statements do not modify
the RF at all. However, since the tests seem to verify that the code
responsible for VALIDATING a request works as intended, that should
have little to no impact on them.
(cherry picked from commit 6bde01bb59)
Adjust `test_service_levels_upgrade`: right before upgrade to topology
on raft, enable an error injection which triggers when the standard role
manager is about to query the legacy auth tables in the
system_auth keyspace. The preceding commit which fixes
scylladb/scylladb#24963 makes sure that the legacy tables are not
queried during upgrade to topology on raft, so the error injection does
not trigger and does not cause a problem; without that commit, the test
fails.
(cherry picked from commit 3a082d314c)
Right now, service levels are migrated in one group0 command and auth
is migrated in the next one. This has a bad effect on the group0 state
reload logic - modifying service levels in group0 causes the effective
service levels cache to be recalculated, and to do so we need to fetch
information about all roles. If the reload happens after SL upgrade and
before auth upgrade, the query for roles will be directed to the legacy
auth tables in system_auth - and the query, being a potentially remote
query, has a timeout. If the query times out, it will throw
an exception which will break the group0 apply fiber and the node will
need to be restarted to bring it back to work.
In order to solve this issue, make sure that the service level module
does not start populating and using the service level cache until both
service levels and auth are migrated to raft. This is achieved by adding
the check both to the cache population logic and the effective service
level getter - they now look at service level's accessor new method,
`can_use_effective_service_level_cache` which takes a look at the auth
version.
Fixes: scylladb/scylladb#24963
(cherry picked from commit 2bb800c004)
Note: The simplest approach to resolving `process_request_one` merge issues, since it has been refactored, was to include the three commits from before, and then the commits that are actually being backported.
`protocol_exception` is thrown in several places. This has become a performance issue, especially when starting/restarting a server. To alleviate this issue, throwing the exception has to be replaced with returning it as a result or an exceptional future.
This PR replaces throws in the `transport/server` module. This is achieved by using result_with_exception, and in some places, where suitable, just by creating and returning an exceptional future.
There are four commits in this PR. The first commit introduces tests in `test/cqlpy`. The second commit refactors transport server `handle_error` to not rethrow exceptions. The third commit refactors reusable buffer writer callbacks. The fourth commit replaces throwing `protocol_exception` to returning it.
Based on the comments on an issue linked in https://github.com/scylladb/scylladb/issues/24567, the main culprit from the side of protocol exceptions is the invalid protocol version one, so I tested that exception for performance.
In order to see if there is a measurable difference, a modified version of `test_protocol_version_mismatch` Python is used, with 100'000 runs across 10 processes (not threads, to avoid Python GIL). One test run consisted of 1 warm-up run and 5 measured runs. First test run has been executed on the current code, with throwing protocol exceptions. Second test urn has been executed on the new code, with returning protocol exceptions. The performance report is in https://github.com/scylladb/scylladb/pull/24738#issuecomment-3051611069. It shows ~10% gains in real, user, and sys time for this test.
Testing
Build: `release`
Test file: `test/cqlpy/test_protocol_exceptions.py`
Test name: `test_protocol_version_mismatch` (modified for mass connection requests)
Test arguments:
```
max_attempts=100'000
num_parallel=10
```
Throwing `protocol_exception` results:
```
real=1:26.97 user=10:00.27 sys=2:34.55 cpu=867%
real=1:26.95 user=9:57.10 sys=2:32.50 cpu=862%
real=1:26.93 user=9:56.54 sys=2:35.59 cpu=865%
real=1:26.96 user=9:54.95 sys=2:32.33 cpu=859%
real=1:26.96 user=9:53.39 sys=2:33.58 cpu=859%
real=1:26.95 user=9:56.85 sys=2:34.11 cpu=862% # average
```
Returning `protocol_exception` as `result_with_exception` or an exceptional future:
```
real=1:18.46 user=9:12.21 sys=2:19.08 cpu=881%
real=1:18.44 user=9:04.03 sys=2:17.91 cpu=869%
real=1:18.47 user=9:12.94 sys=2:19.68 cpu=882%
real=1:18.49 user=9:13.60 sys=2:19.88 cpu=883%
real=1:18.48 user=9:11.76 sys=2:17.32 cpu=878%
real=1:18.47 user=9:10.91 sys=2:18.77 cpu=879% # average
```
This PR replaced `transport/server` throws of `protocol_exception` with returns. There are a few other places where protocol exceptions are thrown, and there are many places where `invalid_request_exception` is thrown. That is out of scope of this single PR, so the PR just refs, and does not resolve issue #24567.
Refs: #24567
This PR improves performance in cases when protocol exceptions happen, for example during connection storms. It will require backporting.
* (cherry picked from commit 7aaeed012e)
* (cherry picked from commit 30d424e0d3)
* (cherry picked from commit 9f4344a435)
* (cherry picked from commit 5390f92afc)
* (cherry picked from commit 4a6f71df68)
Parent PR: #24738Closesscylladb/scylladb#25240
* github.com:scylladb/scylladb:
test/cqlpy: add cpp exception metric test conditions
transport/server: replace protocol_exception throws with returns
utils/reusable_buffer: accept non-throwing writer callbacks via result_with_exception
transport/server: avoid exception-throw overhead in handle_error
test/cqlpy: add protocol_exception tests
transport: remove redundant references in process_request_one
transport: fix the indentation in process_request_one
transport: add futures in CQL server exception handling
This commit:
- Extends the Drivers support table with information on which driver supports tablets
and since which version.
- Adds the driver support policy to the Drivers page.
- Reorganizes the Drivers page to accommodate the updates.
In addition:
- The CPP-over-Rust driver is added to the table.
- The information about Serverless (which we don't support) is removed
and replaced with tablets to correctly describe the contents of the table.
Fixes https://github.com/scylladb/scylladb/issues/19471
Refs https://github.com/scylladb/scylladb-docs-homepage/issues/69Closesscylladb/scylladb#24635
(cherry picked from commit 18b4d4a77c)
Closesscylladb/scylladb#25247
Tested code paths should not throw exceptions. `scylla_reactor_cpp_exceptions`
metric is used. This is a global metric. To address potential test flakiness,
each test runs multiple times:
- `run_count = 100`
- `cpp_exception_threshold = 10`
If a change in the code introduced an exception, expectation is that the number
of registered exceptions will be > `cpp_exception_threshold` in `run_count` runs.
In which case the test fails.
Fixes: #25273
(cherry picked from commit 4a6f71df68)
Replace throwing protocol_exception with returning it as a result
or an exceptional future in the transport server module. This
improves performance, for example during connection storms and
server restarts, where protocol exceptions are more frequent.
In functions already returning a future, protocol exceptions are
propagated using an exceptional future. In functions not already
returning a future, result_with_exception is used.
Notable change is checking v.failed() before calling v.get() in
process_request function, to avoid throwing in case of an
exceptional future.
Refs: #24567Fixes: #25273
(cherry picked from commit 5390f92afc)
Make make_bytes_ostream and make_fragmented_temporary_buffer accept
writer callbacks that return utils::result_with_exception instead of
forcing them to throw on error. This lets callers propagate failures
by returning an error result rather than throwing an exception.
Introduce buffer_writer_for, bytes_ostream_writer, and fragmented_buffer_writer
concepts to simplify and document the template requirements on writer callbacks.
This patch does not modify the actual callbacks passed, except for the syntax
changes needed for successful compilation, without changing the logic.
Refs: #24567Fixes: #25273
(cherry picked from commit 9f4344a435)
Previously, connection::handle_error always called f.get() inside a try/catch,
forcing every failed future to throw and immediately catch an exception just to
classify it. This change eliminates that extra throw/catch cycle by first checking
f.failed(), getting the stored std::exception_ptr via f.get_exception(), and
then dispatching on its type via utils::try_catch<T>(eptr).
The error-response logic is not changed - cassandra_exception, std::exception,
and unknown exceptions are caught and processed, and any exceptions thrown by
write_response while handling those exceptions continues to escape handle_error.
Refs: #24567Fixes: #25273
(cherry picked from commit 30d424e0d3)
Add a helper to fetch scylla_transport_cql_errors_total{type="protocol_error"} counter
from Scylla's metrics endpoint. These metrics are used to track protocol error
count before and after each test.
Add cql_with_protocol context manager utility for session creation with parameterized
protocol_version value. This is used for testing connection establishment with
different protocol versions, and proper disposal of successfully established sessions.
The tests cover two failure scenarios:
- Protocol version mismatch in test_protocol_version_mismatch which tests both supported
and unsupported protocol version
- Malformed frames via raw socket in _protocol_error_impl, used by several test functions,
and also test_no_protocol_exceptions test to assert that the error counters never decrease
during test execution, catching unintended metric resets
Refs: #24567Fixes: #25273
(cherry picked from commit 7aaeed012e)
This issue happens with removenode, when RBNO is disabled, so range
streamer is used.
The deadlock happens in a scenario like this:
1. Start 3 nodes: {A, B, C}, RF=2
2. Node A is lost
3. removenode A
4. Both B and C gain ownership of ranges.
5. Streaming sessions are started with crossed directions: B->C, C->B
Readers created by sender side exhaust streaming semaphore on B and C.
Receiver side attempts to obtain a permit indirectly by calling
check_needs_view_update_path(), which reads local tables. That read is
blocked and times-out, causing streaming to fail. The streaming writer
is already using a tracking-only permit.
Even if we didn't deadlock, and the streaming semaphore was simply exhausted
by other receiving sessions (via tracking-only permit), the query may still time-out due to starvation.
To avoid that, run the query under a different scheduling group, which
translates to the system semaphore instead of the maintenance
semaphore, to break the dependency. The gossip group was chosen
because it shouldn't be contended and this change should not interfere
with it much.
Fixes#24807Fixes#24925
- (cherry picked from commit ee2fa58bd6)
- (cherry picked from commit dff2b01237)
Parent PR: #24929Closesscylladb/scylladb#25052
* github.com:scylladb/scylladb:
streaming: Avoid deadlock by running view checks in a separate scheduling group
service: migration_manager: Run group0 barrier in gossip scheduling group
This issue happens with removenode, when RBNO is disabled, so range
streamer is used.
The deadlock happens in a scenario like this:
1. Start 3 nodes: {A, B, C}, RF=2
2. Node A is lost
3. removenode A
4. Both B and C gain ownership of ranges.
5. Streaming sessions are started with crossed directions: B->C, C->B
Readers created by sender side exhaust streaming semaphore on B and C.
Receiver side attempts to obtain a permit indirectly by calling
check_needs_view_update_path(), which reads local tables. That read is
blocked and times-out, causing streaming to fail. The streaming writer
is already using a tracking-only permit.
To avoid that, run the query under a different scheduling group, which
translates to the system semaphore instead of the maintenance
semaphore, to break the dependency. The gossip group was chosen
because it shouldn't be contended and this change should not interfere
with it much.
Fixes: #24807
(cherry picked from commit dff2b01237)
The references were added and used in previous commits to
limit the number of line changes for a reviewer convenience.
This commit removes the redundant references to make the code
more clear and concise.
(cherry picked from commit 9b1f062827)
Fix the indentation after the previous commit that intentionally had
a wrong indent to limit the number of changed lines
(cherry picked from commit 9c0f369cf8)
