Known Issues
5166 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-136764In Code Review
Complete TODO listed in SERVER-117943
There is a TODO in code owned by your team which references a resolved ticket.
The TODO appears in:
* src/mongo/db/modules/atlas/src/disagg_storage/pali/pali_callbacks.cpp:37
* src/mongo/db/modules/atlas/src/disagg_storage/pali/pali_callbacks.cpp:168
* src/mongo/db/modules/atlas/src/disagg_storage/pali/pali_callbacks.cpp:363
You can also search for all references to the resolved ticket here:
https://github.com/10gen/mongo/search?q=SERVER-117943&type=Code
The next steps for this ticket are to
NODE-7953ClosedDone
Document that cursors do not support overlapping hasNext() / next() / tryNext() calls
h3. Use Case
As a developer using the Node driver,
I want the cursor API docs to say that overlapping {{hasNext()}} / {{next()}} / {{tryNext()}} calls are not supported,
So that I don't fire several unawaited calls on one cursor and get duplicate {{find}} commands, leaked server cursors or session errors.
h3. User Experience
The {{AbstractCursor}} class comment, and the comments on {{{}hasNext(){}}}, {{next()}} and {{{}tryNext(){}}}, state that calls must not overlap.
Background in N
Components
CursorsDocs
Fixed Versions
7.9.1
SERVER-137761In Progress
Run WiredTiger's verify with verify_internals config for modal validation
Full validation calls WiredTiger verify once per table and per index, and each call used to re-scan the entire metadata for duplicate btree IDs. WiredTiger now gates that database-wide check behind an opt-in verify_internals config, so the server can this separately from WT verify. This ticket adds that single call to the validate path, it will be one extra verify call (verify("database:")) before looping through verify for each collections.
Full validation calls WiredTiger verify once per ta
SERVER-136313Investigating
Add validation that a node never sends a heartbeat to itself in DSC
SERVER-134162 fixed a bug where a DSC primary's skip-self check in {{_startHeartbeats()}} compared each member's {{_id}} to the node's array index in the config, assuming that the member's ID and index position would match. After a reconfig reordered the members, the primary skipped the other electable node and heartbeated itself instead. This ocurred silently, as a node that heartbeats itself gets a well-formed response back, so the heartbeat chain keeps running and the bug is invisible from in
CSHARP-6309Investigating
[Microsoft Agent Framework] .NET: code review and NuGet release
*Background*
[mongodb-developer/ms-agent-framework-mongodb|https://github.com/mongodb-developer/ms-agent-framework-mongodb] is our MongoDB integration for Microsoft Agent Framework. It has a Python package and a .NET package, {{{}MongoDB.AgentFramework{}}}. We want the .NET package ready to ship, and we need the .NET team's help to get it there. Part of the MAF epic PENGINT-57.
*Scope*
# *Code review:* review the .NET code on {{main}} and flag anything that should be fixed or tightened b
SERVER-139528In Code Review
Add the recipient kAwaitingFetchTimestamp -> kCreatingCollection transition validator (resharding recipient pre/post-condition validations)
SERVER-139053In Code Review
Load-shedding sweep reads another thread's QueryLifespan slot without synchronization
Item 1 of SERVER-138793. {{sweepQueuedOperationsForMemoryPressure}} ({{query_memory_load_shedding.cpp:475}}) runs on the monitor thread. For each client it takes the {{ClientLock}}, then calls {{QueryLifespan::getIfExists(opCtx)}} (line 504) and {{getOpState(opCtx)}} (line 507). Both read the lifespan slot, a plain {{std::shared_ptr}} decoration on the {{OperationContext}}.
The op's own thread writes that slot without the {{ClientLock}}: {{QueryLifespan::get()}} (lazy create), {{bind()}}, and t
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139122In Code Review
Query-memory load shedding: fix double count, log outside client locks, trim RSS read under mutex, correct "never throws" comment
Items 4, 5, 6 and 8 of SERVER-138793. All four are in {{query_memory_load_shedding.h}} and {{query_memory_load_shedding.cpp}}.
*Double count (item 6).* The queued-operation sweep kills an op with {{killOperation()}} then {{recordShed()}}. The op's own thread, in {{queryMemoryCheckLoadShedding}}, calls {{markKilled()}} then {{recordShed()}} without checking for a pending kill. If both paths hit one op, {{operationsShed}} and {{bytesShed}} count it twice. The callers run {{checkForInterrupt()}} b
SERVER-139493ClosedWon't Do
CBR limit propagation floors child limit shares, costing OR/AND subtrees as free under small limits
h2. Summary
{{CardinalityEstimator::propagateLimit}} ({{src/mongo/db/query/compiler/optimizer/cost_based_ranker/cardinality_estimator.cpp}}) splits a limit between the children of OR, AND_HASH, AND_SORTED and SORT_MERGE in proportion to their cardinality estimates, and passes a FETCH's limit on to its child. Both use {{CardinalityEstimate::toCount()}}, which floors:
Any child whose proportional share is below one row is therefore given a limit of 0. That sets its output estimate to 0 and scales
SERVER-137809Open
Split insert/delete opcounters by tier: cold insert and cold delete operation metrics
[Metric support sheet for cold collection|https://docs.google.com/spreadsheets/d/1nqLitHW0S62oZzTjKMhj1G4iE9K6E2PHsqe6pqbuAJ0/edit?gid=1093816893#gid=1093816893]
Atlas needs insert/delete throughput charts for cold collections (e.g. on the Metrics tab, alongside the existing Opcounters charts). mongod already exposes insert/delete operation counters via {{{}serverStatus.opcounters{}}}, but they are tier-blind: an insert or delete against a cold collection is indistinguishable from one against
SERVER-139302ClosedFixed
Fastcount can drift on index build attempts collection on clean shutdown
If a _clean_ shutdown and restart rolls back a commitIndexBuild or abortIndexBuild the sizeStorer's count for local.config.indexBuilds is left 1 higher than it should be. At the time of shutdown, (assuming no other index builds) the collection is empty and the count is correctly 0. On startup, the index build is restarted due to that the commit/abort has been rolled back, which creates a ledger document and sets the count to 1. Oplog replay then reapplies the commit/abort, which deletes the docu
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139444ClosedFixed
Give concrete FCV steps independent Bazel targets
h2. Objective
Modularize the build of the concrete steps used by the setFeatureCompatibilityVersion command. Give each concrete FCV step its own Bazel library, keep the registry/framework library independent of concrete implementations, and link the implementations into mongod through one aggregation target. Demonstrate that a concrete step can be linked and tested independently of all other steps and of the setFCV command implementation.
h2. Problem and motivation
Currently, the framew
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139525Backlog
Register bulkWrite query stats on the legacy mongos write path
SERVER-137436 registers bulkWrite ops for query stats on mongos only on the unified write executor path. When {{featureFlagUnifiedWriteExecutor}} is off, mongos runs bulkWrite through the legacy {{cluster::bulkWrite}} path, which registers nothing, so no bulkWrite query stats are recorded on mongos.
*Scope*
* Register bulkWrite ops for query stats on the legacy mongos bulkWrite path, matching the unified write executor path.
*Close-out condition*
If the unified write executor is fully rolled o
SERVER-139524Backlog
Report the bulkWrite op queryShapeHash in bulkWrite explain output
Explaining a bulkWrite with a single update or delete op reports the {{queryShapeHash}} of the equivalent standalone update/delete command, not the bulkWrite op shape hash recorded in {{$queryStats}}. The two hashes differ by design, so the hash from explain cannot be used to find the bulkWrite entry in {{$queryStats}}.
* mongos: {{cluster_bulk_write_cmd.h}} converts the op into a standalone update/delete request and runs the standalone explain path, which computes the update/delete shape hash.
SERVER-137436Investigating
Shapify bulkWrite ops on mongos
Register query shapes and stats keys for bulkWrite ops on mongos, so that mongos records a query stats entry per bulkWrite op the same way it already does for update and delete commands.
*Scope*
* In {{WriteCmdQueryStatsRegistrar::parseAndRegisterRequest}} ({{src/mongo/s/write_ops/write_cmd_query_stats_registrar.cpp}}), add a bulkWrite branch. Today it returns early for anything that is not an insert/update/delete command.
* For each op, build the bulkWrite op shape (update/delete/insert) and t
SERVER-139523Needs Scheduling
Provide ability to disable stack trace captures during tasserts in tests
Some tests might not care about tassert output but they pay a time cost for the stack trace capture that happens during it slowing them down considerably. For example bson spatial fuzz test doesn't care about tassert trips during fuzzing but it pays the time cost.
I found a roughly 44x speed up if I disable libunwind but I found a 300x speed up if I disable stack trace collection altogether.
SERVER-137438Backlog
Attach metrics to bulkWrite responses on shards
On shards, when a bulkWrite op carries {{includeQueryStatsMetricsForOpIndex}}, attach that op's query stats metrics (keys/docs examined, etc.) to the bulkWrite reply, as update/delete/insert commands already do.
Depends on the bulkWrite IDL fields from SERVER-137437. mongos aggregation of these metrics is SERVER-137437.
SERVER-137437Backlog
Collect query stats for bulkWrite ops in sharded
Collect shard-side query stats metrics for bulkWrite ops on mongos, building on the per-op registration from SERVER-137436.
*Scope*
* Add an {{includeQueryStatsMetricsForOpIndex}} field to bulkWrite update/delete/insert ops, and a per-op {{queryStatsMetrics}} field to the bulkWrite reply, mirroring the update/delete/insert command IDL in {{write_ops.idl}}.
* On mongos, set {{includeQueryStatsMetricsForOpIndex}} on shard-bound bulkWrite ops that were registered for query stats, and aggregate the
SERVER-139521Backlog
Multiversion and Upgrade/Downgrade Testing
This ticket is to ensure there is adequate multiversion testing
coverage for this project.
1. Test that this feature works correctly on mixed binary version
clusters.
2. Test how the feature behaves when upgrading FCV turns the
feature on
3. Test that the correct behavior occurs when downgrading FCV
completely and then upgrading FCV completely
4. Test that the correct behavior occurs if the FCV upgrade fails.
5. Test that the correct feature behavior occurs if the FCV downgrade
fails after only