Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
PYTHON-6167In Code Review
Split the test-non-standard task family into per-suite task families
h3. *Context*
{color:#505f79}The {{test-non-standard}} task family is shared by the encryption, load balancer, doctest, and PyOpenSSL suites, and the build variant decides which suite a task actually runs. Task names and tags do not identify the suite they run, which makes failures harder to attribute and keeps every suite from handling its own matrix.
PYTHON-6112 split the encryption runs by sync and async at the variant level. With per-suite task families, that mode belongs in the tasks
Components
Infrastructure
NODE-7905ClosedDone
Update dependencies for mongodb-connection-string-url
h3. AC
* Merge dependabot PRs that are safe to merge (semver-compatible updates, including any standalone package-lock-only PRs)
* When updating the dev dependencies to latest available semver compatible versions ANY reformatting MUST be done in a separate pull request and should ONLY be done if the team agrees that it is an important improvement
* Check dev dependency major version updates
** Update the major version if and ONLY if this can be done with minimal code changes; otherwise, file
Fixed Versions
mongodb-connection-string-url-7.0.3
NODE-7906ClosedDone
Update dependencies for dbx-js-tools
h3. AC
* Merge dependabot PRs that are safe to merge (semver-compatible updates, including any standalone package-lock-only PRs)
* When updating the dev dependencies to latest available semver compatible versions ANY reformatting MUST be done in a separate pull request and should ONLY be done if the team agrees that it is an important improvement
* Check dev dependency major version updates
** Update the major version if and ONLY if this can be done with minimal code changes; otherwise, file
SERVER-139333Needs Scheduling
Re-evaluate risk of zeroCE
After we have implemented histograms/NDV/MCV statistics, we should re-evaluate the risk of CE returning 0. And if its low, we should remove the fallback put into place by SERVER-136192/SERVER-136193
SERVER-139293In Progress
Incorrect output in $bucketAuto's $mergeObjects
In $bucketAuto's $mergeObjects, the value for each field should come from the document that is last in the input, tracked by its input position. When a value arrives, we compare its position with the largest position recorded for that field and only overwrite the output if it is greater. However, the recorded position is not updated when we overwrite the value. So the stored position stays at the first one seen for that field, and later values are compared against it instead of against the lates
SERVER-135420In Progress
Create PSAMEQueue, a single worker and PSAMEService
Create the PSAME task queue (a {{MultiProducerSingleConsumerQueue}}, producers use non-blocking {{tryPush()}}), a single worker thread running on low-priority tickets ({{TaskType::Background}}), and {{PSAMEService}} ({{ReplicaSetAwareService}}: {{onStartup}} initializes the queue and worker; {{onStepUpComplete}}/{{onStepDown}} notify primary state). {{PSAMEService}} is instantiated in {{mongod_main.cpp}}'s {{_initAndListen}}.
The queue has a single consumer, so there is one worker, not a pool.
SERVER-134963In Code Review
Lower the number of initial sync attempts before crashing
Based on an investigation into how many attempts an initial sync node requires before succeeding, 99.677% of initial syncing nodes succeed after 0 or 1 retries. Out of 83,323 initial sync attempts, only 5 initial syncs failed the first three attempts and succeeded on some later attempt (4-9).
As a result, it seems we can safely lower the initial sync attempt threshold. 10 seems to be too large of a number and we can increase observability by failing sooner.
SERVER-137247ClosedDone
Prototype a Query Admission Control and run against OOM reproducers
Help us to evaluate the impact of QAC for OOM reproducers.
SERVER-139123ClosedFixed
Move task ownership to CAR / CS
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-138010Open
Committed transactions report a spurious writeConcernError when a session kill interrupts the majority write-concern wait
A transaction that commits successfully can still report a {{{}writeConcernError{}}}(WCE) if a session kill delivered an interrupt to the committing operation before its majority write-concern wait completed. The interrupt is sticky and cannot be retracted, so the client receives {{ok: 1}} with WCE on a transaction that is durably committed. Retrying the whole operation in response is unsafe because the write already committed.
h3. Why the error is not recoverable for clients as-is?
[{{Tempo
SERVER-95324ClosedFixed
Disable chunkMigrationConcurrency
{\{chunkMigrationConcurrency}} can led to excessive resource consumption. In some instances, it also results in an increase in orphaned documents, complicating data management. For this reasons we are deprecating and removing it. Setting this parameter won't have any effect on the cluster, we are making it a no-op operation
Components
Sharding
Fixed Versions
8.1.0-rc08.0.57.0.23
SERVER-139266Needs Scheduling
Remove knob internalQuerySamplingCEMethodForPersistentSamples
Remove the query knob. Make sure that api compatibility is maintained, check with Atlas and Atlas Infinite to make sure knob is not in use anywhere what will break when it's removed.
Fixed Versions
9.1 Required
PYTHON-6262ClosedFixed
getMore retries and multi-mongos retryable writes fail in load-balanced mode
h3. *Context*
PYTHON-6167 makes the load-balancer Evergreen tasks run the whole default suite under {{loadBalanced=true}}. The first run failed 4 tests per task, and the PR now skips them in LB mode. Two underlying gaps:
* Pinned-cursor getMores never retry on {{SystemOverloadedError}} and {{RetryableError}} errors; the retryable reads and Client Backpressure specs expect retries in LB topology. Failing: {{test_retry_overload_error_getMore}} and the two {{TestClientBackpressureGetMoreRetri
Components
Load Balancer
Fixed Versions
4.20.1
SERVER-137994Open
Join optimization INLJ over multikey compound index returns duplicate row per array element
createIndexProbeQSN() in reorder_joins.cpp builds the IndexEntry with isMultikey=false regardless of the catalog, so buildIndexedJoinIndexProbe() never adds the dedup stage. Probing \{y: 1, tags: 1} with bounds [x, MinKey]..[x, MaxKey] returns the foreign document once per element of tags. The join field itself is scalar, so path arrayness does not block the plan; INLJ is chosen naturally with a large foreign collection.
SERVER-130790Backlog
Handle bumping CollectionTag/index fingerprints if multikeyness flips due to document write
As stated in SPM-3143, multikeyness flips from a document write operation are not subject to the same strict transactional guarantees as DDL operations. This means that updating the necessary versions and fields due to such an operation will require careful synchronization.
SERVER-121215Backlog
Cost of INDEX_PROBE_NODE does not take into account number of index keys visited per 1 probe
This plan is costed incorrectly:
The cost of the INDEX_PROBE_NODE is not reported but the cost of the overall join is 1.02, and the cost of the left join input is 0.92. This leaves a cost of 0.10 for the right input and the join itse
SERVER-139139ClosedFixed
Extract model-agnostic FSM walk into FsmCommandGenerator base class
Extract the randomized graph walk of {{ShardingCommandGenerator}} into a model-agnostic base class, so a transaction-centric generator can subclass it instead of copying ~200 lines of walk.
h2. Scope of Work
* Create {{jstests/libs/util/change_stream/change_stream_fsm_command_generator.js}} — {{{}FsmCommandGenerator{}}}: {{{}constructor(seed){}}}/{{{}getSeed(){}}}, {{generateCommands(model, params)}} (BFS all-pairs shortest paths precompute, unvisited-edge iteration: random unvisited self-loo
Fixed Versions
9.1.0-rc09.1.0-rc1026
PYTHON-6210In Progress
[Rust core] Audit PyMongo's platform and wheel matrix against Atlas driver metadata and Rust toolchain support
h3. *Context*
PyMongo ships binary wheels with a compiled C extension for a fixed platform matrix. A Rust-core PyMongo (PYTHON-6162) must ship on the same platforms.
Every PyMongo client sends an {{os}} document in the handshake {{client}} metadata; Atlas records it. Rust tiers 1 and 2 ship a prebuilt standard library; tier 3 does not (https://doc.rust-lang.org/rustc/platform-support.html).
Current wheel matrix ({{.github/workflows/dist.yml}}, {{[tool.cibuildwheel]}} in {{pyproject.toml
SERVER-137945Backlog
Store MCV and NDV in persistent stats
Fixed Versions
9.1.0-rc09.1.0-rc1025