Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-115250ClosedFixed
Pause JWKSetRefreshJob during quiesce time
SERVER-101491 Stopped the fetcher from calling the IDP in a loop until successful. This fixed the issue of the server potentially getting banned from calling the IDP in a loop until successful if the IDP server was down.
We should also sleep the JWKSetRefreshJob for the quiesce period defined by the user, since right now JWKSetRefreshJob will still retry in a loop until successful, logging an error message on each attempt before the quiesce period has passed.
Even though we are not doing a
Fixed Versions
9.0.0-rc08.3.38.0.247.0.39
SERVER-101491ClosedFixed
Quiesce in JWKFetcherImpl when remote returns 5xx error
This tickets extends the quiesce functionality of clients making requests with an unkown kid. It now also adds a quiesce period for failed fetches to the IDP during key refresh intervals. If an IDP server is down the quiesce period must pass before a new request is made to the IDP server.
Fixed Versions
8.2.0-rc08.1.38.0.167.0.31
SERVER-138785ClosedFixed
Assert that insert succeeds in write_without_single_shard_key_single_shard_data_placement_change.js retry loop
Fixed Versions
9.1.0-rc09.1.0-rc1025
NODE-7941ClosedDone
node-mongodb-native - PR #5083: docs: generate docs from latest main [skip-ci]
[github-actions[bot]|https://github.com/apps/github-actions] has created [PR #5083: docs: generate docs from latest main [skip-ci]|https://github.com/mongodb/node-mongodb-native/pull/5083] in [node-mongodb-native|https://github.com/mongodb/node-mongodb-native]
Fixed Versions
7.8.1
GODRIVER-3947Backlog
Unified runner does not enforce runOnRequirements.csfle.minLibmongocryptVersion
h3. Detailed steps to reproduce the problem?
The Go driver parses [{{runOnRequirements.csfle.minLibmongocryptVersion}}|https://github.com/mongodb/specifications/blob/master/source/unified-test-format/unified-test-format.md#runonrequirement] from the Unified Test Format but never enforces it. Tests that should be skipped when the loaded libmongocrypt is too old instead run and fail with confusing low-level errors.
{{CSFLEOptions.MinVer}} ({{bson:"minLibmongocryptVersion"}}) is never read anyw
Components
Client Side EncryptionTesting
Affected Versions
2.6.1
NODE-7664Backlog
Flaky: socks5 test spinup hanging [4]
[https://parsley.corp.mongodb.com/evergreen/mongo_node_driver_next_windows_nodeless_test_socks5_csfle_95572f344ac4affb9feec978d01b8c6442a79165_26_07_07_17_59_16/0/task?bookmarks=0,456]
socks5test.py hangs on launch timing out after 2 hours.
{code:java}
[2026/07/07 14:23:39.281] ++ source secrets-export.sh
[2026/07/07 14:23:39.281] +++ set +x
[2026/07/07 14:23:39.281] + set -o errexit
[2026/07/07 14:23:39.281] + export TEST_CSFLE=true
[2026/07/07 14:23:39.286] + TEST_CSFLE=true
[2026/07
Components
Flaky Test
SERVER-138596ClosedFixed
Lower operationMemoryPoolBlockMaxSizeKB in resumable index build tests
The resumable primary-driven index build tests give each index a ~1 MB sorter budget, but after a spill the sorter keeps its newest memory-pool block (up to 2 MB by default) and counts its full size against that budget, so once that block alone exceeds the budget every later key immediately puts the sorter over budget and gets spilled on its own, making the tests slow and creating thousands of one-key spill runs.
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-136298ClosedFixed
Validate per collection hash on standby
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-138716ClosedFixed
Record prepared txn replicated metadata on standby
Today, the replicated fast count system on the primary tracks prepared transactions by parsing the 'm' field of the commitTransaction oplog entry. These contributions are subsequently checkpointed. However, the primary does not track prepared transactions hashes, so the collection's hash is set to boost::none, which indicates it should not be validated.
The standby SizeCountCheckpointValidator needs to match this behavior. It should ignore prepared transaction operations before they are commi
Fixed Versions
9.1.0-rc09.1.0-rc1025
GODRIVER-4165ClosedDone
Sample NumberSessionsInProgress in mtest.ResetClient
h3. *Context*
In the mtest.T.teardown function we count the number of sessions in progress on a client and verify that value is zero. But if a prose test resets the client, this assertion could be wrong.
Additional context: https://github.com/mongodb/mongo-go-driver/pull/2624#discussion_r4086038162
h3. *Definition of done*
Sample sessions onto mtest.T when a client is reset. In the teardown the assertion becomes this:
h3. *Pitfalls*
Uninvestigated
SERVER-138878ClosedFixed
Reject negative resumesWithoutProgress values in index build resume metadata
Components
Index Builds
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-61184ClosedFixed
Compound 2dsphere indexes can generate a massive number of index keys per document
When generating keys for 2dsphere indexes we calculate the [cartesian products of all keys|https://github.com/mongodb/mongo/blob/2d5a16adde2b17c851675c6a95f84afa33347cd1/src/mongo/db/index/expression_keys_private.cpp#L106-L128].
If documents contain locations with a large number of location coordinates compounded with an array of a large number of elements the resulting cartesian product can be massive in size.
When generating all these keys they are inserted into a single std::vector tha
Fixed Versions
5.3.0
SERVER-138222ClosedFixed
user_defined_roles should test all subsets of privileges for auth failures
The inline documentation of the `privileges` option in jstests/auth/lib/commands_lib.js states that the custom roles test runner will check "that having only a subset of the
privileges causes an authorization failure". However, in jstests/auth/commands_user_defined_roles.js:runOneTest(), the runner checks each privilege individually but not any larger subsets of privileges.
Either change the runOneTest() logic to test every subset of the full set of privileges for an auth failure, or update
Fixed Versions
9.1.0-rc09.1.0-rc1025
CSHARP-5913Backlog
.NET/C# Driver v4.0
{panel:bgColor=#fafbfc}
h3. *Summary*
_The MongoDB .NET/C# driver needs a new major release for capabilities that encapsulate breaking changes and help move towards a path of better performance, maintainability and developer experience._
h3. *Motivation*
h4. Who is the affected end user?
_All users of the .NET/C# driver_
h4. How does this affect the end user?
_N/A_
h4. How likely is it that this problem or use case will occur?
_N/A_
h4. If the problem does occur, what are the c
Fixed Versions
4.0.0
CSHARP-6299ClosedDone
Use a local NuGet source instead of MyGet in CI tests
The Evergreen tasks that test the driver packages against other projects ({{test-efcore}}, {{test-odata}}, {{test-analyzer}}) and the smoke tests ({{test-smoke-tests-*}}) currently depend on {{push-packages-myget}}. That task publishes every build to the public MyGet feed so the tasks can restore the freshly built driver package from there. As a result:
* every CI build is published to a public feed, even though only these tasks need the package;
* these tasks cannot run in any Evergreen project
Components
Build
Fixed Versions
4.0.03.next
CSHARP-6298ClosedFixed
Fix tests when MONGO_URI is set to 127.0.0.1
The following tests are failing after mongo orchestration script started setting MONGO_URI as "mongodb://127.0.0.1:27017" instead of "mongodb://localhost:27017":
* {{ServerDiscoveryAndMonitoringProseTests.ConnectionPool_cleared_on_failed_hello}}
* {{ServerDiscoveryAndMonitoringProseTests.Monitor_sleep_at_least_minHeartbeatFrequency}}
Fixed Versions
4.0.03.next
CSHARP-6296ClosedDone
Remove resolve_s3 from the Lambda deploy configuration
The test-aws-lambda-deployed task on main fails during sam deploy with "Cannot use both --resolve-s3 and --s3-bucket parameters". drivers-evergreen-tools v1.4.0 and later pass --s3-bucket with a tagged artifact bucket to comply with the Cloud Tag Policy (DRIVERS-3717), while tests/FaasTests/LambdaTests/samconfig.toml still sets resolve_s3 = true for deploy. Remove resolve_s3 from the deploy parameters so only the bucket supplied by drivers-evergreen-tools is used.
Fixed Versions
4.0.03.next
SERVER-138558ClosedFixed
Delete persistedSizeCount API on Collection
All callers should be using latestSizeCount instead
Fixed Versions
9.1.0-rc09.1.0-rc1025
PYTHON-6227ClosedDone
Enable ruff rule UP035 to remove deprecated import aliases
h3. *Context*
Ruff's rule [UP035|https://docs.astral.sh/ruff/rules/deprecated-import/] replaces deprecated import aliases with their modern versions.
h3. *Definition of done*
Remove the suppression of UP035.
h3. *Pitfalls*
{color:#505f79}What should the implementer watch out for? What are the risks?{color}
SERVER-138791ClosedFixed
Keep the $replaceRoot new-root check of SBE from looking dead
{{$replaceRoot}} and {{$replaceWith}} must fail when the new root is not a document, even when no later stage reads the new root. With {{featureFlagSbeBlueprintRewrites}} on, SBE blueprint dead slot elimination removes the unread new-root entry and the check with it, so on {{{x: 1}}} the pipeline {{[{$replaceWith: "$x"}, {$replaceWith: {ok: 1}}]}} returns {{{ok: 1}}} instead of failing.
* Mark the new-root slot with {{SbBlueprint::keepSlot()}} in {{buildReplaceRoot()}}, as the sort's parallel a
Fixed Versions
9.1.0-rc09.1.0-rc1025