Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-137727Open
Relax invariant on idIndexIdent
the idIndexIdent field is optional (see [here|https://github.com/10gen/mongo/blob/f7749b41a6832d96f2271324e62acca928324605/src/mongo/db/repl/create_oplog_entry.idl#L23-L26]), but during the execution of the command, there is an [invariant for that field|https://github.com/10gen/mongo/blob/f7749b41a6832d96f2271324e62acca928324605/src/mongo/db/shard_role/shard_catalog/database_impl.cpp#L861].
SERVER-137839Open
Complete TODO listed in SERVER-124472
There is a TODO in code owned by your team which references a resolved ticket.
The TODO appears in:
* buildscripts/modules/atlas/suites/no_passthrough_disagg_override_replsets.yml:128
* jstests/libs/replsettest.js:1887
* src/mongo/db/modules/atlas/jstests/disagg_storage/libs/disagg_storage_override.js:153
You can also search for all references to the resolved ticket here:
https://github.com/10gen/mongo/search?q=SERVER-124472&type=Code
The next steps for this ticket are to either remove the ou
SERVER-137801Open
Ensure applyOps delete cannot crash the server if it fails to find preimage
The delete branch of applyOperation_inlock in src/mongo/db/repl/oplog.cpp decides whether to record a change-stream pre-image *before* it runs the delete, then asserts that a pre-image was produced. The decision helper shouldRecordChangeStreamPreImage() (src/mongo/db/repl/oplog.cpp:2579-2591) looks only at whether the collection has changeStreamPreAndPostImages enabled and at the oplog application mode; it never checks that the document being deleted exists. When the delete matches nothing, dele
SERVER-137674Open
Add upper bound to heartbeatTimeoutSecs
[heartbeatTimeoutSecs|https://github.com/10gen/mongo/blob/31f7deeb74b47ca9372001f40ac0e8a7835a301f/src/mongo/db/repl/repl_set_config.idl#L79-L85] has no upper bound, so sufficiently large values are accepted, but overflow when [converted to milliseconds|https://github.com/10gen/mongo/blob/31f7deeb74b47ca9372001f40ac0e8a7835a301f/src/mongo/db/modules/atlas/src/disagg_storage/disagg_repl_set_config.h#L219] (multiplied by 1000). We should add an upper bound to the IDL validator so these configs are
SERVER-135127Open
Investigate and remediate recurring server-side causes of majority replication lag
h3. Action item
Investigate and remediate the recurring server-side causes of majority replication lag seen in repeat-offender clusters.
h3. Context
Provides the engineering disposition for customers that remain affected after the Stream 1 prevention rollout — that is, clusters where the released pacing and prevention mechanisms are active and majority replication lag still recurs.
h3. Ownership
* *Stream:* Stream 4 — Create a repeat-offender process with CSM and TSE support
* *Team(s):* Re
SERVER-133165Open
Update HELP Playbook for HELP-98001
The information and context for HELP-98001 was flagged for adding to the HELP Playbook. As owner of the original HELP issue (Assigned Teams), please review the linked issue and update the playbook accordingly.
SERVER-137858Open
Direct client inserts into local.oplog.rs invisible to oplog scans until restart — intentional in 9.0?
On mongod 9.0.4+ (bracket tested: clean on 8.3.13), acknowledged client inserts into a manually-created capped {{local.oplog.rs}} on a standalone node are persisted (unfiltered metadata count reflects them) but no scan returns them — {{find()}} / {{countDocuments}} / {{$natural}} all return 0 — until the mongod restarts. Restart rebuilds visibility and all entries appear. Looks like oplog visibility (OplogVisibilityManager, SERVER-106810 project area) is now advanced only through the replication
Components
Replication
SERVER-112402Open
Capture RSTL acquisition time distribution
Currently we have an [RSTL acquisition timeout of 1 minute|https://github.com/mongodb/mongo/blob/830b73a8895d10c53a7445ebe95d337dbbc76d0e/src/mongo/db/repl/repl_server_parameters.idl#L564] on step up / step down.
It used to be 30 seconds, but we changed it to 1 minute because we suspected that in many cases we weren't seeing a deadlock but slowness, and that the slowness was too slow be to resolved in 30 seconds but fast enough to be resolved in under a minute.
We made that decision kind o
SERVER-134803Open
Logging under mutex stalled mongod for the 4 seconds
`ReplicationCoordinatorImpl` emits log records *{*}while holding `_mutex`{*}*, and the logv2 sink
serialises on its own mutex *{*}across the backend write{*}*. A log consumer that is slow or blocked
therefore does not merely delay logging: it holds `_mutex` for the full duration of the write,
stalling every other user of that mutex — inbound `replSetHeartbeat`, the `oplog` `serverStatus`
section, and the heartbeat scheduler itself.
The BF-45880 is a reproduction of the problem - mongod g
SERVER-135733Open
Add multiversion testing to initial sync clean shutdown behavior
Some of the tests that were added in SERVER-132245 are not compatible with multiversion tests where we randomly assign versions to binaries. We should add some targeted multiversion tests to ensure that the behavior works as intended with mixed binary workloads.
GODRIVER-2765Backlog
Add an "omitzero" BSON struct tag
To support the use case of encoding Go structs as MongoDB documents, we will add an {{omitnil}} BSON struct tag that only omits {{nil}} values but includes empty values. The {{$push}} and {{$addToSet}} update operators work the same when modifying document fields that are empty arrays and document fields that do not exist (the only problem is document fields that are BSON null), so the behavior of {{omitnil}} maps well to that use case. Additionally, other MongoDB drivers (e.g. Java and Rust) of
Components
BSON
SERVER-138909Open
Lint jstests for one-shot commands
Commands like {{replSetStepDown}} can fail due to race conditions. We should therefore disallow
and instead use something like
For the rare cases that require that a single command succeed, we should provide a {{{}// NOLINT{}}}.
This would have caught BF-44771 and BF-46860.
SERVER-139005ClosedDuplicate
Consider wrapping replSetStepUp and replSetStepDown in retry helpers in jstests
BF-46612 (deduped to BF-44771) was an issue in which {{node.adminCommand(\{replSetStepUp: 1\})}} can transiently fail. The fix was simple: wrap the command:
We should consider this type of change more broadly, along with a linter.
SERVER-130148ClosedFixed
Report counters of connections according to client backpressure version
Backpressure-aware drivers report in their handshake that they support interpreting backpressure from the server. We should gather this metadata and report it for future analytical and policy development use.
Fixed Versions
9.1.0-rc0
GODRIVER-2116ClosedWon't Do
Add MongoDB Error Codes as Constants
MongoDB has a list of error codes here: https://github.com/mongodb/mongo/blob/master/src/mongo/base/error_codes.yml#L36. I think it would be helpful to have these error codes exposed as constants so that when handling `ServerErrors` when using the Go Driver, those constants can be used to handle specific errors. Otherwise, code that uses the go driver needs to copy + paste the error codes manually which can be error prone or might become out of sync. In addition, non-MongoDB employees might not
Components
Error Handling
SERVER-138757ClosedDone
ConflictDetector API and key model
h2. Overview
Define the internal API and key model for a conflict detection module that is separate from the storage engine, as part of the SPM-4946 prototype. This header is the seam that the detector implementations, commit-path integration, staged writes, and log-and-apply work all build on.
h2. Background
Today WiredTiger detects write-write conflicts while writes are staged. The proposed design stages writes in the RecoveryUnit and asks an in-memory detector at commit time. The detector mu
GODRIVER-3137Backlog
Gossip cluster time from internal MongoClient to session entities
This ticket was split from {{DRIVERS-2816}}, please see that ticket for a detailed description.
Components
Unified Test Runner
PYTHON-6031Ready for Work
Consolidate non-IO code in _AggregationCommand into shared base class
All methods except {{get_cursor}} can be lifted out of {{_AggregationCommand}} and its related classes into shared base classes. Note that {{_CollectionAggregationCommand}} and its ilk cannot be easily lifted due to multiple inheritance, but the class's small size makes it an acceptable duplication.