Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-138704In Progress
Rework join node serialization in explain
In the explain {{queryPlan}} of join-optimized plans:
* Rename stages: {{NESTED_LOOP_JOIN_EMBEDDING}} to {{NESTED_LOOP_JOIN}}, {{HASH_JOIN_EMBEDDING}} to {{HASH_JOIN}}, {{INDEXED_NESTED_LOOP_JOIN_EMBEDDING}} to {{INDEXED_NESTED_LOOP_JOIN}}, {{INDEX_PROBE_NODE}} to {{INDEX_PROBE}}.
* Replace {{leftEmbeddingField}}/{{rightEmbeddingField}} on join nodes with {{embedAs}} on the root of each embedded collection's access path (no {{'none'}} sentinel).
* Serialize {{joinPredicates}} with output-documen
SERVER-134476Backlog
Address client CPU errors on resharding sys-perf workloads
From https://github.com/10gen/dsi/pull/5887.
After running patches to verify the changes in the above CR, I found that both the development branch and the master branch had various flaky test failures, including the following.
CPU Error
* reshard_collection_writes_to_other_coll_locust_test_execution
* reshard_collection_failover_test_execution
System Failed
* reshard_collection_burst_locust
* reshard_collection_2d2r_with_users_with_connection_storm_rate_limited_locust
Run a seri
SERVER-139076In Code Review
Report all estimated NDVs and rename selectivity in join explain ndvEstimates
In the detailed join explain output, change each {{ndvEstimates}} entry to:
* {{ndvs}} lists the NDV estimate of every side that was estimated: the smaller collection's side always, and the other side when it is available without sampling (unique index or persisted NDV statistics). It replaces {{assumedPkSide}}/{{ndv}}/{{ndvSource}}, which only show the larger of the two.
* Rename {{selectivity}} to {{joinPredicateSelectivity}} (= 1 / max(ndv)).
{{joinPredicateSel}} in {{cardinality_estimator.
SERVER-138703In Code Review
Rename detailed join estimate fields in explain
Rename {{cardinalityRHSBeforeJoinPred}} to {{innerCardinalityBeforeJoinEstimate}}, and in {{joinCostComponents}}: {{numDocsTransmitted}} to {{docsTransmitted}}, {{sequentialIOPages}} to {{sequentialIOs}}, {{randomIOPages}} to {{randomIOs}}. These fields stay behind {{internalEnableDetailedJoinExplain}}.
SERVER-134702ClosedFixed
Add connection pool metrics for connections created shortly after cancelation
To get a better idea of how often connection churn occurs, we should add a metric which tracks the number of connections that were created shortly after a connection in the same pool was closed due to cancelation. This will help us understand better the cost of canceling commands and may be used to help motivate something like inline cancelation.
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-138179Backlog
Write conflict can escape moveChunk command
It appears that the {{moveChunk}} command can return a write conflict error rather than retrying it internally.
SERVER-138728Backlog
Add a building-index phase to the resharding coordinator
Recipients build indexes after cloning, but the coordinator stays in the cloning phase for that entire duration. Donor history is currently unpinned when the coordinator advances to applying, even though the retained history is only needed until all recipients finish cloning.
Long-running index builds can extend the time donor history remains pinned, increasing cache pressure and adding system pressure.
The coordinator needs a distinct building-index phase that begins once every recipient
SERVER-139332ClosedFixed
Remove locals from structlog exception tracebacks
{{structlog}}'s default console renderer shows locals in exception tracebacks, which is noisy and can leak secrets.
This updates a logging utility to always configure logging without locals, by default.
Components
Testing Infrastructure
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-134183ClosedFixed
Investigate: multi-statement transaction may read silently incomplete results after shard catalog history filtering
h3. Suspected problem
When {{featureFlagShardCatalogExpiredHistoryFiltering}} is enabled, authoritative metadata disk recovery drops chunk documents unless {{shard == <me>}} or {{onCurrentShardSince > <WT oldest timestamp>}} ( {{readCollectionMetadataFromDisk()}} in +[collection_metadata_synchronizer | [https://github.com/10gen/mongo/blob/master/src/mongo/db/shard_role/shard_catalog/collection_metadata_synchronizer.cpp#L37-L144]), and rebuilds the routing table with {{{}RoutingTableHistory::m
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-137928ClosedFixed
Add a microbenchmark for key_string::readValue()
The SBE hot path now uses readValue() instead of toBson(), so add a variant of the toBson benchmark that measures it.
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-134424In Code Review
Remove the temporary disagg no-schema-epochs test variant
SERVER-133713 added enterprise-amazon-linux2023-arm64-disagg-no-schema-epochs, a temporary Evergreen variant that runs the disagg suite with featureFlagEnableSchemaEpochs set to false. It exists so a failure in the disagg suite can be attributed to schema epochs rather than to something else.
Remove the variant once schema epochs are stable enough that the comparison no longer earns its CI cost.
The variant is defined in etc/evergreen_yml_components/variants/amazon/test_dev_master_branch_only.
SERVER-81782ClosedFixed
Waiting for journalling is redundant with waiting for replication for replicated writes
See [this doc|https://docs.google.com/document/d/1TaN4VhYHU6Abpc2ccvAn65Sa9Im8ZtDymX7Ol6PwiJs/edit] for a more detailed description of how waiting for journaling works today.
Right now, for (implicit and explicit) {{j:1}} writes we wait for durability on the current node (which also includes waiting for no oplog holes if voters=1) prior to waiting for replication. However, the mechanism used for waiting for replication is redundant with the mechanism for waiting for durability now that the re
Fixed Versions
8.0.0-rc0
SERVER-139268ClosedFixed
Make SBE efficient boundary materialization flag an IFR
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-138517Needs Scheduling
Disable test_api_version_compatibility
This task is not compatible with our BF infrastructure, security process, and breaks frequently with little value derived.
SERVER-123456ClosedGone away
Cleanup legacy classes for collection size and count
This [may include|https://github.com/mongodb/mongo/blob/6efa48969872ee9a29ed082b26e63c83ab6a6fe6/src/mongo/db/replicated_fast_count/replicated_fast_count_uncommitted_changes.h#L40], but is not limited to, {{UncommittedFastCountChange}}
SERVER-139208ClosedFixed
Refactor CheckpointHashValidationTest::appendCheckpoint()
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139190ClosedFixed
Refactor isRepairReplicatedMetadataO2()
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-137710ClosedFixed
Handle repair oplog entry in SizeCountCheckpointValidator
If we repair any replicated metadata for a collection, the standby must expect the repaired value(s). Without this change, we cannot repair any collection metadata while Phase 2 validation is enabled. There are at least a couple interesting cases:
# Repairing a {{boost::none}} hash to a real hash. If we parse the repair command correctly in the validator, we should be able to validate the collection’s hash during repair.
# Collection validation failed, so we are repairing incorrect metadata
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-136627Open
SBE ignores internalQueryLegacyDottedPathNullSemantics for MinKey/MaxKey comparisons on dotted paths through arrays
_Found by an AI bug hunt and confirmed by a differential oracle._
h2. Problem
For a comparison against {{MinKey}} or {{MaxKey}} on a dotted path that crosses arrays, the SBE engine returns the same documents whatever {{internalQueryLegacyDottedPathNullSemantics}} is set to, and differs from the classic engine under both settings. Comparisons against {{null}} agree between the engines under both settings.
h2. Repro
Collection {{c}} holds these documents; the query is run for each predicate belo
SERVER-139016Backlog
Use CompactHashAggStage for eligible $group stages behind featureFlagSbeCompactHashAgg
Make the SBE stage builder run an eligible scalar {{$group}} on {{CompactHashAggStage}}, behind {{featureFlagSbeCompactHashAgg}}, which stays off by default. The stage builder picks {{CompactHashAggStage}} only when all of these hold; otherwise it builds {{HashAggStage}} as before.
* {{featureFlagSbeCompactHashAgg}} is enabled.
* {{allowDiskUse}} is false, since the compact stage cannot spill yet.
* Every accumulator has a compact implementation ({{$sum}} and {{$push}} for now). A {{$group}} wit
Fixed Versions
9.1.0-rc09.1.0-rc1026