Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-77560ClosedFixed
Update resmoke results to include log paths
Update the results.json output to link log paths and test names when using the file system logger.
Requires backport.
Fixed Versions
8.0.0-rc0
SERVER-77559ClosedFixed
Implement file system log handler for resmoke
Format and log lines to files in the reserved Evergreen logging directory.
Requires backport.
Fixed Versions
8.0.0-rc06.0.157.0.85.0.277.3.3
SERVER-77561ClosedFixed
Rollout resmoke file system logging
This will be a phased rollout:
* Patch build testing
* Single required build
* All required builds
* All builds
Requires backport.
Fixed Versions
5.0.278.1.0-rc06.0.167.3.38.0.0-rc57.0.11
SERVER-136254ClosedFixed
Upgrade WASM spidermonkey engine to ESR 153
Fixed Versions
9.1.0-rc09.1.0-rc1026
CSHARP-5974In Code Review
Test against .NET 11 in early September
Update tests to run against .NET 11 when RC1 ships, which is usually early September. Test again when RC2 ships.
SERVER-111883Backlog
Add ability for cursors to yield and return results early
even if it hasn't reached EOF or filled up the buffer. One possible implementation would be allowing callers to specify maximum time before yielding (this is different with maxTimeMs which will completely cause the command to error out).
This will allow caller to record progress (either with resume token if supported or just based on some fields from documents returned like oplog)
SERVER-50111ClosedFixed
Secondary stuck with old sync source for reporting up to 30 seconds
EDIT: Cancelling the reporter on changing sync source is a more robust solution.
syncSourceFeedbackNetworkTimeoutSecs is currently [hardcoded to 30s|https://github.com/mongodb/mongo/blob/eae31861e0f813f0099e1d490c4a622d75cd5a08/src/mongo/db/repl/sync_source_feedback.cpp#L57]. In case of network partition, the sync source feedback report might need to take 30s before timing out on the replSetUpdatePosition remote command against the old sync source even though the node has selected a new sync
Components
Replication
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-138978ClosedFixed
Wait for mongos primary discovery after failover in live_shard_logical_initial_sync
On master, the [test|https://github.com/10gen/mongo/blob/master/jstests/sharding/live_shard_logical_initial_sync.js#L29-L31] still [writes|https://github.com/10gen/mongo/blob/f826fc5db8671a18184620430bff15e446f722b3/jstests/sharding/libs/sharding_state_test.js#L107] immediately after the [failover helper|https://github.com/10gen/mongo/blob/master/jstests/sharding/libs/sharding_state_test.js#L82-L93] returns.
The helper waits for replica-set readiness, but not for mongos to recognize the new p
Affected Versions
9.1.0-rc09.0.0
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-76347ClosedFix Not Needed
Documentation Updates
This ticket is to update all types of documentation API Guide,
Programming Guide, and Architecture Guide related to the project.
SERVER-138166Backlog
Investigate perf of TPC-E locust q03
q03: classic order is industry, company, security,
joo picks same order, but uses a HJ at the end, which is more expensive; classic uses ~an INLJ on s_co_id. There's also a CE error.
SERVER-138164Backlog
Investigate CE error on TPCE q5
Looks like q5 performs worse with join opt enabled than classic. The exchange.ex_ad_id = address.ad_id predicate is mis-estimated at 27k rows when in fact it only returns one.
May possibly be fixed by enabling internalEnableJoinOptimizationUseIndexUniqueness for this variant- but maybe we should first look at why this estimate is so bad in the first place.
SERVER-138159Backlog
Add memory bounds on join-optimizer data structures
Proposed initial list:
* JoinPlanNodeRegistry::_allJoinPlans
* PlanEnumeratorContext::_joinSubsets
* ReorderedJoinSolution::rejectedSolns (probably needs a cap, unless explain work dec ides we're not going to support it)
* Sampling estimators: bound sum of samples (although, maybe this is implicitly bounded to max nodes * per-sample bound?)
This is a parent task, we'll probably want subtasks to add metrics for these fallbacks
SERVER-136738In Progress
Perform a trial run on join plan cache hit and make replanning decision
SERVER-136737, implemented the logic around inserting baseline metrics into the join plan cache entry. On a cache hit, these metrics (numReads, numIntermediateRowsProduced and numSpills) from the TrialRunTracker should be used to make a decision on whether to initiate replanning for the query. The plan from the cache will be run under a trial period for N number of docs (tuneable) and its metrics will be recorded.
The replanning decision will be based on:
# Did the query start spilling when
SERVER-137800Backlog
Handle wildcard/hashed/collated index scans on JOO path
SERVER-134637 is adding support for CE & costing of wildcard/hashed/collated index scans in CBR. That means that automatically these indexes can also be used for single table access plans in JOO. The problem is that our code *costIndexScanFragment()* is not prepared to cost I/O for any indexes beside BTrees. The options we have:
# Modify costing of IXSCAN fragment and give some cost for unknown index types (con: may be wrong)
# Use existing IXSCAN fragment IO cost for BTree and use same lo
SERVER-138912Needs Scheduling
Split DocumentSourceLookup resolvedPipeline into segmented sections
SERVER-136263In Code Review
$group producing MinKey/MaxKey followed by a classic stage turns the value into an object {"": MaxKey}
With the default engine selection, when {{$group}} runs in SBE and produces MinKey or MaxKey, and a later stage runs in the classic engine (here {{$project}}), the value arrives as an object with a single empty-named field holding MinKey/MaxKey instead of MinKey/MaxKey itself. With {{forceClassicEngine}} the same pipeline returns MinKey/MaxKey.
* *classic engine*: {{internalQueryFrameworkControl: forceClassicEngine}}. Check: the explain line in the output shows {{explainVersion 1}}, with {{$gro
Components
Query Execution
Affected Versions
8.3.11
PYTHON-5838Investigating
Agents consistently choose Motor over ASyncClient
h3. *Context*
h3. The following is part of a conversation with Claude Code about the Code Agent Developer Experience.
h3. A. LLMs consistently choose MotorClient over pymongo.AsyncMongoClient. How do we change this?
This is a pure training-data problem with a known shape. Motor has been the async answer in Python MongoDB code on the public internet since ~2015. AsyncMongoClient shipped in PyMongo 4.9 (late 2024). Every model whose training cutoff predates that — which is most of them, inclu
Components
AI/ML
PYTHON-6260ClosedDone
Opt into all ruff TC rules
h3. *Context*
{color:#505f79}Describe the background behind the problem.{color}
PYTHON-6243 enabled the small number of {{TC}} rules on by default. Opting into the full suite of {{TC}} rules will improve code consistency and performance.
h3. *Definition of done*
{color:#505f79}What must be done to consider the task complete?{color}
Enable all {{TC}} rules and make the required changes to pass them.
h3. *Pitfalls*
{color:#505f79}What should the implementer watch out for? What are the
SERVER-133852ClosedWon't Fix
Fine-tune slice count/size parameters for record store slicing
As part of validation parallelization, the current record-store slicing logic uses hard-coded values to determine how many slices to create. Update this to use configurable parameters (i.e target slice count, size), and investigate how we can derive reasonable defaults based on collection characteristics and workload.
Document the chosen parameters and provide heuristics for setting them.
PYTHON-6263ClosedDone
Drop the blacken-docs pre-commit hook
h3. Context
The blacken-docs pre-commit hook formats code samples with the pinned black 22.3.0, which is stale next to the repo's ruff-format hook, and it failed the static job on PR 3131 by rewriting CONTRIBUTING.md.
h3. Definition of done
The blacken-docs block is removed from .pre-commit-config.yaml, the dead skip markers are removed from CODING_STANDARDS.md, and the static job passes.
h3. Pitfalls
Doc code samples will no longer be auto-formatted, so example drift from ruff style must b