Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-136637ClosedFixed
Tolerate collection drop races in CheckMultikeyConsistencyInBackground hook
## Description
`CheckMultikeyConsistencyInBackground` failed in BF-46711 (`disagg_concurrency_replication_high_yielding`): the workload under test intentionally drops and re-creates its pool collections, and one such drop raced the hook's doc-sampling read.
In `assertWildcardMultikeyConsistent` (`jstests/hooks/run_check_multikey_consistency.js:144`), the `findOne({}, {_id: 0})` used to sample a document only tolerated `NamespaceNotFound`. A drop that lands \*while the read is running\* kills t
Components
Concurrency
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-137128ClosedFixed
Skip DDL FailFast for drop collection, drop database, drop index
Those operations are stopping the running index builds, so no need for avoid the concurrency.
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-136961ClosedFixed
Resolve FCBIS crash loop after restarting from unclean shutdown
h2. Context
FCBIS works by copying WT files from its sync source to a local temp directory called {{{}.initialsync{}}}, then moving these files to the original dbPath that the node is running on. In order to perform this move, FCBIS must [restart the storage engine|https://github.com/10gen/mongo/blob/18f971a4374e8f041e5d22ee38705bfe0d4177c7/src/mongo/db/modules/enterprise/src/fcbis/file_copy_based_initial_syncer.cpp#L1712-L1725] because we cannot safely move or delete WT files while they are
Fixed Versions
9.0.9
SERVER-133486ClosedFixed
Stop mongod hanging in shutdown under TSAN
Fixes a mongod hang during storage engine shutdown under TSAN, seen as `task-timed-out` on the TSAN fuzzer variants (BF-45518, and the same signature as BF-38299 which was closed on a suspected-TSAN theory without a root cause).
## The deadlock
Read directly off a core dump of a wedged mongod. Two threads deadlock over glibc's dynamic loader lock:
**`SignalHandler`** holds the global lock for shutdown and is running the WiredTiger shutdown checkpoint:
```
__wt_txn_global_shutdown -> __wt_che
Components
Build
Fixed Versions
9.1.0-rc09.0.48.3.18
SERVER-133979ClosedFixed
Anchor the libzstd TSAN suppression
## What
Changes the zstd entry in `etc/tsan.suppressions` from `called_from_lib:libzstd.so` to `called_from_lib:libzstd.so$`.
## Why
SERVER-133486 added `called_from_lib:libzstd.so` to stop a mongod shutdown deadlock under TSAN. That pattern is matched as a substring against every loaded module's full path, and TSAN requires a `called_from_lib` pattern to resolve to exactly one module — if it matches two, `LibIgnore::OnLibraryLoaded` reports and calls `Die()`.
The antithesis images have two:
Components
buildscripts
Fixed Versions
9.1.0-rc09.1.0-rc10229.0.48.3.18
SERVER-137819ClosedFixed
Retry CSRS demotion related InvalidOptions for collMod in replica_sets_transition_to_csrs suite
There are some places where a collMod sent *directly* to a shard (i.e. a mongod with sharding Grid initialized) may try to contact the CSRS: [1|https://github.com/mongodb/mongo/blob/3b35b031787090e1598d08028cfcf74bdf5d2b13/src/mongo/db/shard_role/shard_catalog/coll_mod.cpp#L140-L159], [2|https://github.com/mongodb/mongo/blob/3b35b031787090e1598d08028cfcf74bdf5d2b13/src/mongo/db/shard_role/ddl/collmod_cmd.cpp#L120-L137].
This is problematic during demotion of config server to replica set, as n
Affected Versions
8.3.09.1.0-rc09.0.0
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.98.3.18
SERVER-137890In Progress
Move remaining s390x Bazel compilation and linking to off-host cross compilation
h2. Problem
SERVER-136176 moved s390x release builds to ARM64-hosted cross compilation, but remaining Bazel build tasks on v9.0 and older branches still compile and link on native zSeries hosts. Extend the off-host cross-compilation approach to all remaining s390x Bazel builds on v9.0, v8.3, and v8.0.
One confirmed example is {{rhel83-s390x-bazel-compile}} in the v9.0 and v8.3 Evergreen configurations: it runs {{compile_bazel_TG}} on {{rhel83-zseries-large}}. Native compilation also remains in
Fixed Versions
9.0.98.3.18
SERVER-132433ClosedFixed
Retryable findAndModify remove by _id is broadcast to all shards and tasserts on retry after chunk migration
{{targetUpdate\(\)}} excludes findAndModify from the {{WithoutShardKeyWithId}} broadcast protocol, but {{targetDelete\(\)}} does not: {{src/mongo/s/query/exec/target\_write\_op.cpp:670\-671}} sets {{isNonTargetedRetryableWriteWithId = isExactId && multipleEndpoints && isRetryableWrite\(opCtx\)}} with no {{\!isFindAndModify}} term, unlike the update path at lines 540\-541 and unlike the {{useTwoPhaseWriteProtocol}} expression immediately above at line 668. Because {{write\_op\_analyzer.cpp:102\-1
Fixed Versions
9.1.0-rc09.0.0-rc48.3.18
SERVER-137865ClosedFixed
Disable DDLFailFast on MigrationBlockingOperationCoordinator
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-126499ClosedFixed
Express write path tasserts on _id $eq with DBRef-shaped object
h2. Problem
The Express update/delete fast path can tassert for an exact {{_id}} equality query where the explicit {{$eq}} value is an object whose first field starts with {{{}${}}}, such as a DBRef-shaped object.
h2. Reproduction
The request is routed to the Express write path and mongod hits a tassert:
* update: {{tassert(9248801)}}
* delete: {{tassert(9248804)}}
h2. Root Cause
{{isSimpleIdQuery()}} accepts {{{}_id: {$eq: <object>{}}}} as an exact {{_id}} query. The Express wr
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.98.3.18
SERVER-138066ClosedFixed
$shardedDataDistribution is slow when the chunk count is large
The current {{$shardedDataDistribution}} performs a {{$lookup}} operation on {{config.chunks}} to determine which shards contain chunks for each sharded collection.
The observed execution plan used the \{uuid: 1, min: 1} index instead of the existing \{uuid: 1, shard: 1, min: 1} index. Therefore, it examines every chunk document matching the collection's UUID to collect the distinct shards ids.
Consequently, the work grows with the total number of chunks across the collections being process
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-136903ClosedFixed
Investigate lifetime issues with the replication worker pool owned by ReplicationCoordinatorExternalStateImpl
h3. Background
Both issues came up in review comments (Copilot) on a PR that made {{ReplicationCoordinatorExternalStateImpl::_workerPool}} a {{const std::unique_ptr[ThreadPool]}} constructed in the constructor via {{{}makeReplWorkerPool(){}}}. They have not been independently verified, so this ticket is to confirm whether they are real and decide on fixes.
h3. Issue 1: Pool can be resized after replication shutdown
* {{getDbWorkThreadPool()}} now stays non-null after {{_stopDataReplication
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-137763ClosedFixed
Regex operators that match repeatedly re-validate UTF-8 on every match
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-135020ClosedFixed
Memory leak: per-response CancellationSource child states accumulate unboundedly on long-lived parent tokens in exhaust command readers (ExhaustResponseReaderTL / AsyncDBClient::_withCancellation)
h3. Summary
Every response received on a long-lived exhaust command session (oplog fetcher getMore, awaitable hello, change-stream getMore) constructs a new child CancellationSource from the reader's long-lived parent token. The child's onCancel() continuation is registered into the parent token's SharedPromise children list and strongly holds the child CancellationState. When the child source is destroyed after the response completes, only its own promise is dismissed; {*}the continuation en
Components
ConcurrencyNetworkingReplication
Affected Versions
8.1.08.2.08.3.28.3.89.0.0
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-113458ClosedFixed
CheckMetadataConsistency should report if FCV upgrade have left legacy timeseries collections
During upgrade to FCV 9.0 timeseries collections will be converted automatically to viewless timeseries . The conversion will start after the cluster (including the config server) have already transitioned to FCV 9.0 and before the setFeatureComatibilityVersion command returns the response to the user. This means that if the conversion is interrupted by an error (e.g. network error, primary stepdown etc..) it will be the responsibility of the user or automation agent to keep retrying the setFeat
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-136366ClosedFixed
$lookup binds a view onto its own definition stages unless the view starts with a source stage
_Filed by the AI query bug hunter (qhunt), an agent that reviews query code for corner-case bugs. A differential oracle and the repro below confirmed the bug._
h2. Problem
When a view is the {{from}} of a {{$lookup}}, the execution-time view binder binds the view onto every stage of the resolved pipeline, including the stages of the view's own definition. Reading the same view directly or through {{$unionWith}} binds an empty namespace to the definition stages. A definition stage that reacts to
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-135301ClosedFixed
Query settings not eligible/applied for cluster-wide change streams
h2. Overview
A whole-cluster change stream (\{\{$changeStream: \{allChangesForCluster: true\}\}\}) always runs its \{\{aggregate\}\}/\{\{getMore\}\} commands against the \{\{admin\}\} database. Any \{\{querySettings\}\} passed inline on that command (query knob overrides, \{\{maxTimeMS\}\}, index hints, \{\{reject\}\}) is silently dropped instead of being applied, regardless of topology (replica set, sharded cluster via mongos, or a direct connection to a shard).
h2. Root Cause
\{\{QuerySetting
Components
Change streams
Affected Versions
9.0.28.0.33
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-138003ClosedFixed
Investigate apiStrict behavior for extension agg stages
DocumentSourceExtensionOptimizable::registerStage registers every extension stage with AllowedWithApiStrict::kAlways, but the built-in fallback parsers for $search, $vectorSearch and $listSearchIndexes (as of SERVER-137305) all use kNeverInVersion1. This means the same strict query can be accepted/rejected depending on whether the extension is loaded and its feature flag is enabled.
One simple approach would be to change the default to kNeverInVersion1 (kInternal for internal extension stages
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9
SERVER-138463In Progress
Move IBM crypt builds to ARM64-hosted cross compilation
h2. Overview
Extend the IBM cross-compilation work in DEVPROD-43701 to the crypt build pipeline for Z Series (s390x) and Power (PPC64LE). Move crypt compilation and linking to dynamically allocated ARM64 Linux hosts using the cross-build approach already used for server builds.
h2. Scope
* Audit the IBM crypt build variants and tasks, including {{crypt_create_lib}}, for compilation, linking, and packaging that still depend on native IBM hosts.
* Reuse the appropriate cross toolchains and target
Components
Build
Fixed Versions
9.0.98.3.18
SERVER-138484ClosedFixed
Always report QMLS counters in serverStatus, even when load shedding is disabled
{{serverStatus().queryMemory.loadShedding}} is omitted whenever load shedding is disabled ({{{}queryMemoryLoadSheddingLowMarkPercent{}}} = -1), because of the early return in {{{}appendQueryMemoryLoadSheddingStats(){}}}. Mongotune switches shedding on and off over time, so consumers see gaps in the cumulative {{operationsShed}} / {{idleCursorsShed}} / {{idleCursorBytesShed}} counters.
Atlas auto-scaling (TUNE-1578) will compute a rate from {{{}operationsShed{}}}. It treats missing samples as
Components
DiagnosticsQuery Execution
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.9