Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-137782ClosedFixed
Fix timeseries_geonear_measurements.js for multi-shard explain
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-137296ClosedFixed
Fix IBM cross-toolchain Windows root substitution
IBM s390x and PowerPC builds fail during Bazel analysis because the shared `mongo_toolchain.BUILD.tmpl` leaves `{windows_unix_root}` unresolved, producing `syntax error at '}': expected :` and cascading toolchain-registration errors (BF-46826).
Supply a quoted empty string in `linux_cross_toolchain_substitutions()` to disable Windows dependency-path aliases for Linux cross toolchains. Add Starlark regression coverage for both IBM architectures with composed and override archives.
Components
Bazel
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-116978ClosedFixed
Tests that expect server crashes will spuriously time out on Windows
We own a number of tests that expect the server to crash ([index_build_killop_secondary_after_commit.js|https://github.com/10gen/mongo/blob/master/jstests/noPassthrough/index_builds/index_build_killop_secondary_after_commit.js], [index_secondary_awaiting_primary_abort_crash_on_commit.js|https://github.com/10gen/mongo/blob/master/jstests/noPassthrough/index_builds/index_secondary_awaiting_primary_abort_crash_on_commit.js] but that will cause time outs after crashing when run on Windows variants.
Fixed Versions
8.3.0-rc0
SERVER-97380ClosedFixed
Prevent index_build_killop_secondary_before_commit.js from running on Windows
index_build_killop_secondary_before_commit.js can hang trying to terminate the process after writing the minidump in Windows. We should prevent this test from running on that variant to generate fewer BFGs.
Fixed Versions
8.1.0-rc0
SERVER-135978ClosedFixed
Implement expression::traversePath
The signature should be something like:
traversePath should implement the traversal semantics used in the evaluation of the Expression classes and return the Type of the result.
There's a few different ways in which a path can be traversed in a pipeline:
* FieldPathExpression - the $a.b syntax
* GetFieldExpression - the $getField expression
* Document::getNestedField - method called in aggregation pipelines in places, not strictly within Expressions.
* there might be more!
W
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-130130In Code Review
Migration recipient corrupts the persisted orphan count by decrementing it for xferMods deletes that remove no document
h2. Problem
On the migration recipient, the apply step for xferMods deletes processes
each {{_id}} in the delete batch by issuing a justOne delete and then
decrementing an in-flight orphan delta. That delta is later persisted and
feeds the recipient's orphan accounting and range-deleter bookkeeping for
the migrated range.
h2. Root cause
The orphan delta is decremented once per delete entry, *unconditionally*,
even when the delete removed nothing:
The recipient legitimately r
Components
Sharding
Affected Versions
6.0.07.0.08.0.0
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-138281ClosedDuplicate
getType should not blindly assume FieldPath semantics for all paths
Quoting from [this thread|https://app.graphite.com/github/pr/10gen/mongo/65508#comment-PRRC_kwDOADUBOM8AAAABB5MXRA]:
{quote}For example if we have:
{$match: \{x: number and not array}}
{$project: \{'p.a': '$x'}}
Then for unknown 'p' once $project does the reshaping operation, the type of 'p' will be object or array.
And the method of traversing 'p.a' (where p might be array) would be unclear. $p.a as an ExpressionFieldPath will produce a number|array, but other logic might not, and
SERVER-138065ClosedFixed
Robustify change stream oplog rewrite test
When running with multiple routers, the cluster time stored in one router can be far behind the latest cluster time in the cluster. If we want to observe only the events created within the test, we should not rely on the cluster time cached in the current router. This can be done by polling the cluster time from all routers or even advancing the shard oplogs with oplog note records.
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-138173ClosedFixed
Add golden tests for timelib
Add tests to detect changes in the behavior and error messages from timelib. The goal is to reduce the risks of updating timelib.
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-136192ClosedFixed
Apply SERVER-133362 into master
SERVER-133362 was only applied to 9.0. Apply it to master now that we know SERVER-135741 cannot be fixed only with sampling
Fixed Versions
9.1.0-rc09.1.0-rc1024
SERVER-137374ClosedFixed
Enable join plan cache by default
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-138171ClosedFixed
Add missing <charconv> include for std::from_chars
Adds missing {{<charconv>}} includes to files using {{std::from_chars}}.
Compiling with Clang 23 detected this issue - it seems that the current code relied on another header including {{<charconv>}}which is not guaranteed by the standard.
Fixed Versions
9.1.0-rc09.1.0-rc1025
JAVA-6302Investigating
Server monitor handshake write has no effective timeout — MongoClient never recovers after frozen-flow TCP disruption (async transports)
h3. Issue + Relevant Info
Reported via MongoDB Support case. After a transient network disruption to the cluster, affected {{MongoClient}} instances *never recover* even once the cluster is fully reachable again: every replica-set member stays {{type=UNKNOWN, state=CONNECTING}} indefinitely, and only a JVM restart restores connectivity.
Root cause (confirmed with the Java driver team on Slack): the server-monitor handshake *write* leg has no effective timeout. This violates the CSOT spec,
Components
SDAM
SERVER-135162ClosedFixed
Expose appropriate API from inferred type constraints
Create and test the API using the inferred data types from the validator MatchExpression.
Fixed Versions
9.1.0-rc09.1.0-rc1025
NODE-7659ClosedFixed
Invalid op index reference during bulkWrite
h3. Expected Behavior
As a driver user
I expect unordered bulk insert operations to consistently match operation results
When calling getSuccessfullyInsertedIds with interspersed op types
h3. Actual Behavior and Impact
* When calling getSuccessfullyInsertedIds, it's possible for the insertedIds list to include indexes that refer to unrelated operations durig a bulkwrite.
Expected behaviour: the correct operation index should be used
* No customer reports
h3. Dependencies
* N
Components
Bulk API
Fixed Versions
7.8.0
SERVER-132514ClosedFixed
Consolidate Evergreen retry setup and fix retries never applying
`evergreen_activate_gen_tasks.py` tried to raise the Evergreen API retry count by reassigning `evg_api._http_retry` after building the client, but that never took effect — the client mounts its HTTP adapter at construction, so the reassignment was ignored. Task generation has been running on default retries this whole time, and the POST that activates tasks was never retried at all.
`evergreen_activate_result_tasks.py` already worked around this by re-mounting the adapter. Extract that into `
Components
buildscripts
Fixed Versions
9.1.0-rc08.3.169.0.78.0.377.0.48
SERVER-123024ClosedFixed
Remove devcontainer tests from 8.3 staging branch
Fixed Versions
8.3.0-rc59.0.7
SERVER-134003ClosedFixed
Remove redundant _recipientCtx.getState() read in resharding recipient change-streams-monitor completion path
h2. Overview
There is a data race (found via TSAN, BF-45859) between an unguarded read of {{_recipientCtx.getState()}} and the mutex-guarded write to {{_recipientCtx}} in {{{}ReshardingRecipientService::RecipientStateMachine{}}}. However, the read is itself redundant and can be safely removed.
h2. Background
see BF-45859.
h2. Scope of Work
* {{src/mongo/db/s/resharding/resharding_recipient_service.cpp}} — remove read on {{_recipientCtx.getState()}}
Fixed Versions
9.1.0-rc09.1.0-rc10229.0.7
SERVER-129387ClosedFixed
Optimize $queryStats for top-k metrics.workingTimeMillis.sum sort + limit pattern
We have an observed production pattern where Atlas observability runs a {{$queryStats}} aggregation of the form:
This is one of the two {{$queryStats}} pipelines run every 60 seconds, and it has been called out as a recurring expensive shape on large clusters with high query-shape cardinality.
In the customer investigation discussed in Slack, a slow query log for this exact pattern showed about 9.8s of CPU / working time for a single invocation. On the same cluster, {{serverStatus.metr
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7
SERVER-135600ClosedFixed
Queued operations should participate in query memory load shedding
h2. Summary
Query memory load shedding cannot shed operations that are waiting for an execution ticket. On the spill_to_disk availability workload this is the entire gap between the in-server shedder (arm E, 0.41 availability, crash-looping) and the mongotune killOp actor (arm D, 0.97 availability, zero failovers) as visible in the attached [^qmls-spill-to-disk-ab-report.html]
h2. Evidence
Analyzed 2026-09-23 from Atlas M50 A/B runs (base {{{}472d159296b6{}}}, patches 6ab2d999 killOp / 6a
Fixed Versions
9.1.0-rc09.1.0-rc10249.0.7