Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-121864ClosedFixed
Add memory usage to synthetic benchmarks
Memory usage is a potential concern for CBR due to the lack of specific memory limits for the planning stage. I only have anecdotal observations from doing the original CBR performance analysis, but I noticed a few cases where peakTrackedMemBytes (recorded in the explain output) was quite a bit higher in automaticCE mode than in the baseline
It might be relatively simple to add memory usage as a new metric for the synthetic benchmarks just by comparing that field to get some sense of CBR's im
Fixed Versions
9.1.0-rc0
SERVER-139166Needs Scheduling
fassert 12436504 "Attempted to reserve optime without a declared write intent" in LocalOplogInfo::getNextOpTimes via TransactionParticipant::performNoopWrite
h3. Summary
A {{mongod}} 9.0.5 shard primary aborted with fatal assertion 12436504, {{"Attempted to reserve optime without a declared write intent"}}, raised from {{LocalOplogInfo::getNextOpTimes}}. The node did not recover and stayed down until the test fixture tore the cluster down ~12 minutes later.
Found in MongoDB CI for the mongosync (Cluster-to-Cluster Sync) project, in a suite that repeatedly kills and restarts {{mongos}} processes against a 2-shard cluster while a migration is running
SERVER-134362ClosedFixed
Add test-only forceParallelism knob
By setting this knob we can bypass the QO size/cost gate and force parallelism.
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-137319ClosedFixed
Add the change stream error-conversion stage ($_internalChangeStreamConvertErrors)
mplement the top-level conversion stage as a registered {{DocumentSource}} and execution-stage pair. *Do not inject the stage into any pipeline yet.*
h2. Stage Definition
h3. {{DocumentSourceChangeStreamConvertErrors}}
* *Stage name:* {{$_internalChangeStreamConvertErrors}}
* Internal stage; excluded from the query shape.
* Constraints:
** {{kChangeStreamStage}}
** {{kReceivingHostOnly}}
** Streaming
** Not swappable with {{$match}} or single-document transforms
h3. {{ChangeSt
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-131449ClosedFixed
Dependency graph loses nested path arrayness after later $project of collection field
When the inclusion projection of a collection field is not at the first position in the pipeline, the field inclusion logic for nested paths can lose the path arrayness of collection fields.
This will not lead to an incorrect rewrite, but it is an omission and can be improved.
We lose arrayness for 'a.b' and 'a.b.c', just 'a' remains as known non-array.
canPathBeArray(nullptr, "a.b") -> true when it should be false
Fixed Versions
9.1.0-rc09.1.0-rc1024
SERVER-138422ClosedFixed
Convert oversized change stream events into error events when showErrorsAsEvents is set inside converting stage
With showErrorsAsEvents, a change stream event that exceeds the BSON size limit (~16MB) currently fails the stream exactly as on master. This ticket covers converting such oversized events into operationType:"error" events so the stream stays alive, and completing/hardening that path.
The core conversion landed in the change stream error-conversion stage:
- The stage serializes each delivered event through serializeChangeStreamEventForExecutor; an event exceeding the BSON size limit is con
Fixed Versions
9.1.0-rc09.1.0-rc1025
SERVER-87365Backlog
Execution control doesn't ramp up fast enough for slow yielding queries
When queries don't yield cooperatively for long periods of time, say due to blocking sorts and groups, execution control does not add tickets fast enough, or at all.
Imagine a workload where queries take longer than 1 second. The default probing interval is 200ms. We will never observe an increase in throughput even though progress is being made, and thus never increase concurrency.
We should consider the following improvements:
*Adjust probing interval based on query latency*
If the
SERVER-137779In Progress
Make PlanYieldPolicySBE thread-capable
The object used for performing yielding captures the point for the clock from the original OperationContext, so it cannot be used for the clones of the stages meant to run in a separate thread unless we double check that the clock is thread safe (the main OperationContext should already survive until all the threads have been stopped). Alternatively, change PlanYieldPolicySBE so that it uses the clock from the OperationContext that is provided at runtime and avoid providing one in the constructo
SERVER-135353ClosedFixed
Heap use-after-free reading a write command's sort pattern from the plan cache
h2. Summary
The per-statement {{sort}} field of the {{update}} and {{bulkWrite}} commands is declared as IDL type {{{}object{}}}, which is an unowned {{BSONObj}} view into the wire message buffer. The sort is threaded through the planner into {{{}MergeSortStats::sortPattern{}}}, which the plan cache retains in its debug info. Once the originating command completes its message buffer is freed, leaving the cached entry holding a dangling {{{}BSONObj{}}}. A later {{$planCacheStats}} aggregation
Fixed Versions
9.1.0-rc09.1.0-rc10258.3.169.0.78.0.377.0.48
SERVER-105234ClosedFixed
Ignore untracked files when checking codeowners
Use the PR [here|https://github.com/hmarr/codeowners/pull/43] to ignore untracked git files to make "bazel run //:codeowners" run faster locally.
Fixed Versions
8.2.0-rc0
SERVER-85145Backlog
ShardNotFound error should not be bubbled up when concurrently removing a shard and running operations
In HELP-54194, we discovered that there are some commands that may fail when a removeShard is taking place / draining a shard. As an example “listIndexes”. It is expected to hit ShardNotFound as a transient error triggered by a specific timing and in a specific window of time, and bubble up to the user application. The command failed can be perfectly retried and successfully executed after that. The exact reproducible test is attached to the comments.
The problem is that the user will be able
SERVER-129378ClosedFixed
Multi part BSONElements sub field sizes are checked
Fixed Versions
9.1.0-rc0
SERVER-120806ClosedWon't Do
Document performance of various BSON spatial memory checks
Run sys-perf patches and benchmark each change. The compilation of performance will help determine which checks should be merged.
SERVER-120276ClosedWon't Do
Check object is contained within the shared buffer when calling shareOwnershipWith
SERVER-120274ClosedFixed
BSONObj construction optionally accepts an explicit size
Used to validate objsize() and eoo()
Fixed Versions
9.1.0-rc0
SERVER-120273ClosedFixed
Validate that BSONElements with a variable size (like BSONType::string) never have negative sizes
Fixed Versions
9.1.0-rc0
SERVER-120272ClosedFixed
BSONElement size computation is also responsible for validating the element type
* valuesize() uses element type to check size.
* It needs to validate all bits of the type, not just the low 5 bits. One option (not the only one) is to use a 256-byte static array, rather than a 32-byte one.
Fixed Versions
9.1.0-rc0
SERVER-120270Open
BSONObjStlIterator uses C++20 range style range
* end() is allowed to return a different type (a sentinel), so operator== knows when it is doing an end check, and can do >= like BSONObj, rather than ==. This ensures that if we somehow end up past the end we exit the loop rather than continuing forever. Alternatively, we could throw in the > case, since something is very wrong if we get there.
* operator* doesn't need to return a `T&`, so we can return a BSONElement by value rather than storing one as a member. This gives us more flexibility
SERVER-119970ClosedFixed
BSONObj checks element ends before eoo during iteration or when handing out BSONElement
This would cover BSON iteration. getField(), which internally does its own iteration, should also require these bounds checking or be changed to use normal iteration.
Fixed Versions
9.1.0-rc0