Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-137011ClosedFixed
Tighten skipping metadata consistency check only if snapshot has no authoritative shards enabled
SERVER-135517 already solved BF-46774 but the edge case when the snapshot was created right during the FCV change was not taken into consideration.
This change rather focuses on the status of the feature flag instead of the FCV change timestamp.
Fixed Versions
9.0.7
SERVER-135621ClosedDone
Add context metadata to GenerateAndCheckPerfResults failures
As of SERVER-129590, the {{GenerateAndCheckPerfResults}} hook no longer fails the task when it cannot retrieve the base commit's performance results. Instead it logs an error and emits it to OpenTelemetry. Set up observability/alerting so the Product Performance team is notified when these errors occur, rather than blocking server engineers.
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-135732ClosedFixed
[v9.0] Make internal_search_id_lookup_ddl_concurrency.js align with current behavior
This fix is targeted at recently-backported $idLookup changes (SERVER-134686 and SERVER-134067 that needed minor test fixes on v9.0
Fixed Versions
9.0.7
SERVER-135196ClosedFixed
Remove patch-only kernel 6.19 buildvariants (other than rseq-check)
These variants run MongoDB in a configuration where it intentionally fails to start. The rseq-check variant specifically runs a very small set of tests intended to test that behavior. The other kernel 6.19 variants just run a bunch of tests that fail because mongod fails to start. They should be removed.
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-135564ClosedFixed
Address hang in BlockingResultsMerger unit test
The test is occasionally hanging. When debugging, it is shown to be a test-only issue where we are incorrectly coordinating between threads to make sure progress is being made.
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-136721ClosedFixed
Fix race in TaskRunnerTest
There's a race between scheduling a task and checking that the task runner is active: if the task finishes running between when it's scheduled and when {{active}} is checked, then the task runner will not be active. We must guarantee that the task will not complete before we check.
Fixed Versions
9.1.0-rc09.1.0-rc10248.3.169.0.7
SERVER-134716ClosedFixed
Consume Mothra teams via S3
h2. Description
Migrate the consumers confirmed in DEVPROD-41443 off clone-and-parse of {{10gen/mothra}} and onto a single GET against the S3 export. Because the export ships a compatibility view in the legacy YAML shape, each migration should be a change to the fetch step only — the parsing stays as-is.
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-136649ClosedFixed
Convert a test-only RSTL call to IR
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-132823ClosedFixed
Harden remaining FCV concurrency workload teardowns after SERVER-125400
h2. Overview
Harden the remaining FCV concurrency workload teardowns after SERVER-125400.
h2. Background
After SERVER-125400, concurrent index builds can cause setFeatureCompatibilityVersion to return transient errors:
* {{12587}} ({{{}BackgroundOperationInProgressForNamespace{}}}): an index build is still running with a stale FCV; retry setFCV.
* {{7428200}} (Incomplete FCV downgrade cleanup): a previous FCV downgrade stopped during internal metadata cleanup and must be completed bef
Fixed Versions
9.1.0-rc09.0.7
SERVER-131654ClosedFixed
Tolerate calling cmc without secondaryCheckMode for multiversion suites
Fixed Versions
9.1.0-rc09.0.7
SERVER-134081ClosedFixed
Express user-index plan spins forever when an index entry references a missing record
h3. Problem
{{LookupViaUserIndex::consumeOne}} loops forever when an index entry points to a record that does not exist. A single orphan index key turns any point query that takes the express user-index path into an unbounded CPU spin, bounded only by the operation's deadline.
Observed on HELP-98643: a dev cluster on rc1018 ran at 100% CPU on the primary with read tickets fully consumed and roughly 350k/sec WiredTiger search calls sustaining a fraction of the normal throughput.
h3. Mechanism
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-135348ClosedFixed
rewrite TicketHolderTestTick.WaitForTicketDeadlineBetweenTimeoutWindows to make it not timing depe
The unit test keeps failing when test running in parallel with other tests and host are under pressure. Need a rewrite to make sure the test is deterministic and won't be hit again
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-137027ClosedFixed
ObservableMutex's Timer::getTime() performs a syscall on non-x86 and non-arm64 builds
As seen in HELP-101554, for non-x86 and non-arm64 builds, Timer::getTime() in {{ObservableMutex}} [calls std::clock()|https://github.com/10gen/mongo/blob/6b5787c43d011981395c54179e4017f3c230a526/src/mongo/util/observable_mutex.h#L61] which performs a syscall which has performance implications.
This was flagged by a customer's performance testing running on s390x which showed significant regressions when upgrading to 8.0.30, the first 8.0 version containing ObservableMutex. Flamegraphs from th
Fixed Versions
9.1.0-rc09.1.0-rc10258.3.169.0.78.0.37
SERVER-95479ClosedFixed
Allow skipping rebuilding indexes during startup on failed initial syncs
In cases where we drop unfinished index builds including _id, we rebuild the _id index for collections as a part of startup. If the node then undergoes initial sync, we end up dropping the index and rebuilding it again. We should avoid duplicating the index build effort from building in startup, dropping in initial sync, and possibly building it again in initial sync. We could detect needing to enable this optimization when retrying failed initial syncs.
Fixed Versions
9.1.0-rc09.0.7
SERVER-137256ClosedFixed
retry connections when verifying repl dbhash
When we make a new connection to check a dbhash in tests, especially tests with background hooks that kill nodes, we should retry on retryable errors. Hash mismatches are not retryable and will still fail the test, but the test should not fail because a single connection failed while a node was being killed.
Fixed Versions
9.1.0-rc09.1.0-rc10259.0.7
SERVER-136318ClosedDone
timeseries_insert_mixed_schema_docs.js bucket count assertion can fail under stepdown + viewless timeseries FCV upgrade/downgrade
[v9.0 only]
[Test only issue]
*Issue*
[The timeseries_insert_mixed_schema_docs.js inserts two measurements and asserts they both fall into the same bucket|https://github.com/mongodb/mongo/blob/e7465432fb44e31872bc6a1fd5ac35befdc90d4f/jstests/core/timeseries/write/timeseries_insert_mixed_schema_docs.js#L27-L43], this assertion can fail under this very specific interleaving:
# The first measurement is inserted, creating a new bucket.
# The node steps down, clearing the in-memory
Components
Upgrade/Downgrade
Affected Versions
9.0.0
Fixed Versions
9.0.7
SERVER-134485ClosedFixed
QueuedAdmissionWithLargeMaxTimeMSSucceeds may start sleeping after mock clock is advanced, causing a hang
In SERVER-132777, we replaced the logic that checked to see if opCtx was currently waiting for interrupt with the more specific check on the AdmissionContext. However, this introduced a separate race, since the advancer thread may wake up before the request thread has actually computed its napTime in the RateLimiter. As a result, the request may compute a napTime with a deadline that is after the clock has been advanced, causing it to hang forever.
To fix this, we should reintroduce the check
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-136123ClosedFixed
transaction_aborted_under_cache_pressure.js still flakes after SERVER-135094 fix: assertion misses NoSuchTransaction outcome
h3. Problem
After the SERVER-135094 fix (PR #63037 / backport #40b79dca66f to v9.0), jstests/noPassthrough/txns_cache_errors/transaction_aborted_under_cache_pressure.js still flakes intermittently, failing with:
Seen in patch build: https://spruce.corp.mongodb.com/task/mongodb_mongo_v9.0_amazon_linux2023_arm64_static_compile_display_burn_in_tests_patch_daf6e0739c52e0285c2306e42fdd1701cd5a5597_6ab470629f924d000739a920_26_09_24_00_36_06
h3. Root cause
In the drain loop, the test ab
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-134519ClosedFixed
Mongos may fail to populate CollectionUUIDMismatch properly if dbPrimary returns a different error
CollectionUUIDMismatch today is populated with the actualCollection namespace almost universally by contacting the dbPrimary shard in the event of a failure.
However, the only place that doesn't do this is the find command on mongos. If it knows that the request targeted the dbPrimary it assumes that the error has been already populated by the dbPrimary response and [skips the call to it|https://github.com/mongodb/mongo/blob/40a624b8e02075d780f72b7142f437f8df3a2f40/src/mongo/s/query/planner/c
Affected Versions
6.0.299.0.18.3.118.0.327.0.43
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-135598ClosedFixed
[v9.0] Fix failing tests from revert of SERVER-113184
Components
Replication
Fixed Versions
9.0.7