Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-139555In Code Review
Remove unused MakeObjSpec displayOrder
Remove the unused {{MakeObjSpec::displayOrder}} vector and its copy/move constructor initializers. Simplify {{MakeObjSpec::toString()}} to iterate in field order, preserving existing behavior because {{displayOrder}} is never populated.
SERVER-139556Needs Scheduling
Make sure slow changing colls have stats refreshed too
Atlas restarts servers approximately once a month to patch them. Since our staleness counters are in memory that means collections changing slowly enough to not cross the threshold for one month will never get updated.
SERVER-137525In Code Review
Refactor sampling estimator logic into a samples PersistentStatsManager
Extract the persistent-sample read/write logic from {{sampling_estimator_impl.cpp}} into a samples {{PersistentStatsManager}} reusable by the query planner, the analyze command and PSAME.
The manager implements the PersistentStatsManager API:
* {{load(StatSpec) -> Optional<Stat>}}
* {{refresh(StatSpec) -> Status}}
* {{delete(StatSpec) -> Status}}
* {{findForColl(UUID) -> Vector<StatSpec>}}
* {{collsWithStats() -> Vector<UUID>}}
See WRITING-11150, PersistentStatsManagers.
SERVER-139551Needs Scheduling
Measure time taken in type inference
We should measure the time taken applying type inference from $match stages in the pipeline. This is the same code block that we will allow to be disabled in SERVER-139549 $match narrowing / branch on forEachPossiblyNarrowingStage and in the future computed field type info).
We need to measure the time in that block and report it in some way (the task includes deciding how).
SERVER-114730Backlog
Allow $merge on field to have explicitly specified _id + shard key
Currently on a sharded collection on field of $merge if not present, defaults to _id + shard key:
[https://www.mongodb.com/docs/manual/reference/operator/aggregation/merge/#std-label-merge-on]
However, if we explicitly state _id + shard key on a collection that doesn't have _id as a part of it's shard key, $merge validation will fail, because it won't find an index that is unique and matches all on fields.
It would be nice to make the default value of "on" behave consistently if it is
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-134187In Code Review
Arrayness of dotted path after inclusion projection is incorrect
The dependency graph reports prefix of $project path as non-array after the $project stage.
This affects the following simple cases:
The prefix "a" is reported as non-array AFTER the $project.
This is wrong: an inclusion projection applies the computed field element-wise:
{a: [{}, {}]} becomes {a: [
{b: 1}
, \{b: 1}]} and a is still an array.
This was discovered using the PBT test from SERVER-134177 which is not merged yet.
*Consequence*
Only the MATCH_PUSHDOWN optimizat
Affected Versions
9.0.0
SERVER-139552Needs Scheduling
Limits testing for type inference
We already have benchmarks which test some worst-case scenarios. We need to review them and see if there's any missing cases. We should identify cases where our algorithms or data storage has quadratic or higher space or time complexity.
Then we should determine reasonable limits to apply inside of our type inference and type representation logic.
It is always valid to not perform type inference and bail, so the goal is to choose limits that are low enough such that type inference time is
SERVER-139438ClosedFixed
Define field stats data structures
As described in the design
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139354In Progress
Respect collMod sample size setting in read path
SERVER-137869 will update the write path (`analyze`) to choose sample sizes set in collMod over the knob configuration if set.
The read path needs to mirror that logic and respect samples set via collMod as well. If not, samples that are smaller than the knob configuration will be rejected even when they are the correct size (set by collMod). This means the read path will silently fall back to on-the-fly sampling, wasting resources.
For this ticket, we want to update the read path to respe
SERVER-127539Open
Provide API for adding more information to dependency graph
The pipeline dependency graph code tracks field provenance as well as:
- missing fields
- constants from ExpressionConstant
- arrayness from PathArrayness API
In the future, we may want to add type inference and other field-level properties.
This ticket is about refactoring the existing field-level properties, so they are owned by policies.
We saw need for that in SKUNK-32.
For illustration, a refactor of the constant propagation out from the graph code into a policy:
SERVER-139548Needs Scheduling
Avoid MP fallback if there is a single zeroCE
We currently pass to MP all zeroCE plans and the winning estimable plan, even if there is a single zeroCE plan.
We can instead treat the zeroCE plan as the unambiguous CBR winner, if there is a single one.
Repro:
Returns (with ~90% probability):
{noformat}
winningPlan: {
isCached: false,
stage: 'FETCH',
filter: { a: { '$eq': 0 } },
nss: 'test.b',
inputStage: {
stage: 'IXSCAN',
nss: 'test.b',
keyPattern: { b: 1 },
SERVER-133362ClosedFixed
Fall back into MP in the case of 0 CE
In some cases the sampling is insufficient to determine the proper CE of a predicate. This can be the case for highly selective predicates on large collections or simply a "bad" sample. This ticket should implement a fallback into MP in the case of a 0 estimate instead of using heuristics.
This ticket was intended as a fix to address HELP-98249 and was targeted only at 9.0
Fixed Versions
9.0.0-rc4
SERVER-139547Needs Verification
Fix evergreen_remote_exec value in the trimmed main-staging-poc Evergreen config
The trimmed {{etc/evergreen\_main\_staging\_poc.yml}} added in SERVER\-138693 sets {{evergreen\_remote\_exec: true}} on 22 variants. The build scripts only recognize the literal {{on}} \({{evergreen/bazel\_evergreen\_shutils.sh}} {{compute\_local\_arg}} and {{evergreen/generate\_evergreen\_bazelrc.sh}} both test {{\!= "on"}}\), so with {{true}} remote execution is silently disabled: {{\-\-jobs=auto}} is added, the Bazel timeout rises to 7200s and the build runs with {{Remote execution is disable
SERVER-139546In Code Review
New class for standbyORPRecovery
Extract standby ORP logic into a class that owns all the state.
SERVER-132800Open
Support configurable multi-threaded oplog writes on secondaries (oplogWriterThreadCount)
On MongoDB 8.0+ (verified against {*}9.0.0-alpha0{*}, {{{}master{}}}), when {{featureFlagReduceMajorityWriteLatency}} is enabled, a dedicated {{OplogWriter}} thread writes fetched oplog entries into the local {{rs.oplog}} collection before handing them off to the oplog applier. In steady-state replication this writer runs on a *single thread* ({{{}OplogWriter-0{}}}). Under high-throughput / large-document workloads this single thread becomes the replication bottleneck: it saturates one CPU core
SERVER-139545Backlog
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
SERVER-139544Backlog
Documentation Updates
This ticket is to update all types of documentation API Guide,
Programming Guide, and Architecture Guide related to the project.
SERVER-139331ClosedFixed
Re-enable lagged_config_secondary.js in no_passthrough_disagg_override_sharded
{{The lagged_config_secondary.js}} test verifies that mongos times out when the only reachable config-server secondary is behind the required metadata timestamp. The test currently uses {{stopServerReplication()}} to lag one secondary. Under DSC, this helper stalls primary-side replication across the replica set instead of pausing only the selected standby, so it does not preserve the intended scenario.
As part of this ticket, we should 1) Adapt the test for DSC, and 2) Add the test to {{no_p
Fixed Versions
9.1.0-rc09.1.0-rc1026