Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
NODE-7991ClosedDone
Fill out security advisory description
h3. _fyi [@sean|https://mongodb.enterprise.slack.com/team/U0B0JA4Q04Q] for the oidc CVE fix, we'll want to fill out the description in the security advisory covering the impact, workarounds, and PoC - doesn't have to be done immediately but is one of the todos for between now and thursday:_
_[https://github.com/mongodb/node-mongodb-native/security/advisories/GHSA-c5cf-gf4q-92vj]_
SERVER-136816ClosedWon't Do
Update sasl_authentication_session_gssapi_test ldaptest endpoint
The old test host {{ldaptest.10gen.cc}} is scheduled to be taken down on late October 2026. This unit test depends on this hostname, and so must be updated to {{ldaptest.build.10gen.cc.}}
Note that this unit test is currently only present in branches prior to v8.3, since it was replaced by a jstest in SERVER-109788.
SERVER-134196ClosedFixed
Add otel metrics for physical bytes in/out
Similar to how we have ingress and egress logical bytes otel counters defined [here|https://github.com/10gen/mongo/blob/b56102aa36ce45061f1bb5b71552f4368d3aa1d6/src/mongo/db/stats/counters.cpp], we should add counters for "physical" bytes which tracks the number of bytes actually sent through the network. This will be useful to include in internal dashboards.
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-137699Needs Scheduling
Monguard: expose host CPU core count for process utilization %
h2. Overview
Expose a gauge for online host CPU cores so SREs can compute _Monguard process CPU as % of host_, instead of absolute cores from {{monguard\_process\_cpu\_seconds\_total}} alone.
h2. Background
Fleet Overview currently shows _Peak Monguard CPU \(cores\) by Region_:
That metric is correct for _process load in cores_ \(from {{/proc/self/stat}} utime\+stime\), but it is _not_ host utilization. Without a core count \(or cgroup quota\), dashboards cannot answer \"how much of th
SERVER-137183In Progress
Extend join optimization to work with query settings indexHints
*this may also have some interaction with .hint(), pending syntax discussions
SERVER-127296In Code Review
Use consistent JSON string serialization functions in plan_stability_subjoin_cardinality_md
Currently, the golden test output is serialized using JSON.stringify() [here|https://github.com/10gen/mongo/blob/b7384c1e7723b9151328ae6780fe572107070834/jstests/query_golden/join_opt/plan_stability_subjoin_cardinality_md.js#L354] and [here|https://github.com/10gen/mongo/blob/b7384c1e7723b9151328ae6780fe572107070834/jstests/query_golden/join_opt/plan_stability_subjoin_cardinality_md.js#L455].
However, the error output produced when a reconstruction assertion fails is [serialized with tojson()
PYTHON-6245ClosedDone
Enable ruff INT rules
h3. *Context*
{color:#505f79}Describe the background behind the problem.{color}
Ruff 0.16.0 added the {{INT}} category of rules derived from the flake8-gettext linter.
h3. *Definition of done*
{color:#505f79}What must be done to consider the task complete?{color}
Remove the suppression of {{INT}} rules or add an explicit comment explaining why keeping it suppressed is correct.
h3. *Pitfalls*
{color:#505f79}What should the implementer watch out for? What are the risks?{color}
PYTHON-6259ClosedDone
Enable ruff TC003 rule
h3. *Context*
{color:#505f79}Describe the background behind the problem.{color}
Ruff's [TC003|https://docs.astral.sh/ruff/rules/typing-only-standard-library-import/] rule requires that stdlib imports used only for type-checking are imported within a {{if TYPE_CHECKING}} block for improved runtime performance.
h3. *Definition of done*
{color:#505f79}What must be done to consider the task complete?{color}
Enable the TC003 rule and make the required code changes for it to pass.
h3. *Pitfa
SERVER-136203Open
potential null pointer dereference in listSessionsParseSpec() in MongoDB 8.0.9:
I found a potential null pointer dereference in {{listSessionsParseSpec()}} in MongoDB 8.0.9
The return value of {{Client::getCurrent()->getOperationContext()}} is passed to {{LogicalSessionCache::get()}} without a prior null check:
{{uassert(
31111,
str::stream() << "The " << DocumentSourceListLocalSessions::kStageName
<< " stage is not allowed in this context :: missing a LogicalSessionCache",
LogicalSessionCache::get(Client::getCurrent()->getOperationCo
Affected Versions
8.0.9
SERVER-129715In Progress
Make query multiversion fuzzers in LTS deterministic and reduce execution time for 9.0
In order to reduce BFs, LTS branches will run the query fuzzers with a fixed seed and for a shorter time (SPM-4019). See prior tickets SERVER-113153 and SERVER-113151. Ping Philip Stoev or assign the ticket to him directly.
SERVER-97413Open
Improve spatial memory safety of BSON library in the face of corrupt BSON
Right now the BSON parsing library assumes that bson is well-formed. This is generally valid because we do validation when accepting messages over the network, our disk images are crc32c-checked, and all bson we generate internally should be inherently well formed. That said, we would benefit from some defense in depth to limit the damage if ill-formed BSON makes its way into a {{BSONObj}} or similar types. Here is a non-exhaustive list of things we can do to improve this:
First the things th
SERVER-106675Needs Scheduling
document ownership model for all BSON document & element types.
Goal: BSONObj, BSONElement, and any new classes introduced in this project are documented with explicit guidance on how to choose when to use each class and how types can be converted between one another.
This could be an arch guide.
SERVER-105887Open
Annotate BSONElement and BSONObj APIs with [[clang::lifetimebound]]
{{BSONElement}} instances are always views into a subset of a buffer containing a BSON document which is owned by a different object, and {{BSONObj}} instances are sometimes views of the entire documents to which they refer. We currently lack any compile-time protections against temporal memory misuses of these types due to referencing underlying buffers after their owning objects have been destroyed. Investigate solving these temporal memory misuses by annotating the {{BSONElement}} and {{BSONO
SERVER-139259In Code Review
Ignore FCV lower-bound override on v6.0
Backport the multiversion_service.py change from SERVER-67791: the generateFCVLowerBoundOverride read from master's releases.yml must not raise the FCV lower bound above this branch's last LTS.
Affected Versions
6.0.0
SERVER-106723ClosedFixed
Always set params.overwrite to false
Set the record store insert overwrite setting ({{{}allowOverwrite{}}}) to {{false}} for all non-clustered collections on attached storage, so a replicated-rid insert colliding with an existing key fails with {{WT_DUPLICATE_KEY}} instead of silently overwriting a different document.
h2. Details
- Today we use {{allowOverwrite=false}} only for clustered and recordIdsReplicated collections and {{true}} for everything else (see [getRecordStoreOptions|https://github.com/10gen/mongo/blob/53ac306590
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-67791ClosedFixed
Tests are wrongly being excluded from running
From slack
[https://mongodb.slack.com/archives/C0V79S1PY/p1656621686256279]
requires_fcv_52 <- tests with this tag should be running on the 6.0 branch but they are not
[https://github.com/mongodb/mongo/blob/831c5c4617312f901dfa3d70838a591365e12825/jstests/sharding/proxy_protocol_connect.js#L3]
[https://spruce.mongodb.com/task/mongodb_mongo_master_enterprise_rhel_80_64_bit_dynamic_required_display_sharding_multiversion_patch_add41e1ac8bf0dbc3be680bdc3e7d49266239e35_62bc8b042a60ed61722227e1_
Fixed Versions
6.1.0-rc0
SERVER-83841Backlog
Simplify the VectorClock implementations
With SERVER-64863, the _configTime_ is also gossiped-in on {_}configServers{_}. With that change in place, the _gossipIn_ and _gossipOut_ functions become symmetrical, which might yield an opportunity to simply the general VectorClock implementation:
# It might be possible to get rid of the separate _gossipIn_ / _gossipOut_ methods, checking at the place where they are called (the only thing left is the trivial vector clock which doesn't gossip in or out anything, so we can perhaps even just
SERVER-132659Open
Async replication compression pipeline: remove the single-thread compression bottleneck for large-document replication throughput
SERVER-130410Backlog
Allow disabling (or independently selecting) wire compression on the replication (sync-source) connection, decoupled from the client-facing listener controlled by net.compression.compressors.
MongoDB's {{net.compression.compressors}} setting controls wire-protocol compression {*}for every connection of the node{*}, including:
# Client → mongod connections.
# *Secondary → sync source (Primary or upstream Secondary)* connections used by {{{}OplogFetcher{}}}, {{InitialSyncer}} and {{{}SyncSourceResolver{}}}.
Snappy is enabled by default. While Snappy is cheap for small OLTP payloads, on *large-document workloads running on small-instance replica sets* the replication path becomes