Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-136809Open
Server parameters for ORP fallback values
The design has a few numbers that we've handwaved, for eg. loop limit before we check for a newer checkpoint, ORP connect retry backoff, ORP stream retry limit, no progress timeout etc.
Make these values runtime settable server parameters with reasonable defaults so we can tune them at runtime
Also consider if we should make ORP behaviour enabled behind a server parameter so we can quickly kill it at runtime if needed.
SERVER-136804Open
Make startup ORP catchup exempt from the secondary lag watchdog
A standby catching up from ORP at startup is by definition definitely more than 2 minutes behind so SecondaryLagWatchdog will crash it mid catchup and we end up in the same restart loop
Either we should exempt the watchdog from running during ORP catchup or we should increase its limit during this time until we cutback to LS. I'm not sure how easy it is to pick a reasonable limit to increase to, so I'd think the less risky option here is to exempt.
Should also check oplogCatchUpForStepUp
SERVER-136471Open
Suppress page server warnings in offline validation
Reconsider page server logging for offline validation.
In the linked validation log, I observed 14,548 of the 20,489 lines (71%) were one repeated warning (logid: 13067400):
"No candidate page server is known to have materialized the requested LSN, continuing with existing routing"
Another warning, (logid: 13295600) "ClosedSegmentCompletionTracker: materialization frontier has not advanced", fires once a minute per cell for the whole run with frontier: "none". StalledMillis begins
SERVER-136866Open
clean observer/provider interfaces for mat tracker
in SERVER-136190 and SERVER-136120 observer interfaces were added but the relative lifetime of the observer/observation are unclear, and we just use smart ptrs/reference wrappers.
We should create proper reusable observer infrastructure to facilitate things like this.
SERVER-138893ClosedFixed
Normalize set types when materializing objects in SBE
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139514Needs Scheduling
Add tests that prepared transactions work across backup / restore in Atlas Infinite
SERVER-139429Open
Make transaction_aborted_under_cache_pressure.js deterministic with failpoints instead of relying on real cache pressure
h3. Background
jstests/noPassthrough/txns_cache_errors/transaction_aborted_under_cache_pressure.js has been patched three times in a row. Each patch followed a WiredTiger eviction change and tuned the test to the new timing.
||Ticket||Trigger||What it changed||
|SERVER-132736|WT-18120 improved eviction, so the app-thread stall dropped (about 0.73 s/s to 0.44 s/s) and the detector rarely fired. The test also leaked sessions (BF-45315).|endSession() in the drain loop; cachePressureEvictionStallTh
Components
Testing Infrastructure
SERVER-128157Open
Handle new WT cache overload error
See WT-17582.
As part of the agreement for exposing and using these WT metrics, WT has agreed to add a new error code (along the lines of WT_OVERLOAD_REJECT or similar). Server must now handle this and return some exception.
Also to think about: how different callers with different priorities might want to handle this error code return differently.
Note: this ticket must be merged in conjunction with the WT ticket that changes it to return the new error code.
SERVER-137509Open
Complete TODO listed in SERVER-122306
There is a TODO in code owned by your team which references a resolved ticket.
The TODO appears in:
* jstests/disk/wt_repair_orphaned_idents.js:84
You can also search for all references to the resolved ticket here:
https://github.com/10gen/mongo/search?q=SERVER-122306&type=Code
The next steps for this ticket are to either remove the outdated TODO or follow the steps in the TODO if it is correct. If the latter, please update the summary and description of this ticket to represent the work you'
SERVER-132653Open
Complete TODO listed in SERVER-127039
There is a TODO in the codebase referencing a resolved ticket which is assigned to you.
Please follow this link to see the lines of code referencing this resolved ticket:
https://github.com/10gen/mongo/search?q=SERVER-127039&type=Code
The next steps for this ticket are to either remove the outdated TODO or follow the steps in the TODO if it is correct. If the latter, please update the summary and description of this ticket to represent the work you're actually doing.
SERVER-135242Open
Handle all errors in the PeriodicThreadToRollbackUnderCachePressure job runner to avoid crash
Any error outside of {{CancellationError}} or {{Interruption}} will escape the lambda, eventually crashing the server ([ref|https://github.com/10gen/mongo/blob/655ca324f52260e3988e4f790dea3c3da7f340bc/src/mongo/db/periodic_runner_cache_pressure_rollback.cpp#L134-L138]). We observe this in a particular case where we receive an [exception|https://github.com/10gen/mongo/blob/655ca324f52260e3988e4f790dea3c3da7f340bc/src/mongo/db/storage/wiredtiger/wiredtiger_connection.cpp#L310-L314] due to hitting
SERVER-135913ClosedFixed
Use layered tables in more disagg unit tests
DisaggOplogApplierImplTest and DisaggOplogTruncateMarkersTest installed DisaggregatedPersistenceProvider after the storage engine was created, resulting in them creating tables with type=file because that setting from the provider happens to be read at engine creation time. This had a few problems: it left WiredTigerKVEngine with a dangingly reference to a deleted persistence provider, it relied on specific implementation details of when WiredTigerKVEngine happened to use the provider, and unint
Components
Storage
Fixed Versions
9.1.0-rc09.1.0-rc1024
SERVER-123894In Code Review
Review TypeChecker::operator() handling of empty TypeSignature
Dave fixed a related bug in SERVER-123790. This might not be all that important, but I don't think we're handling the empty type signature correctly in the TypeChecker code, generally speaking.
For instance, in this unit test
we remove the FillEmpty here
which speaks to a class of bug where we're using isSubset without considering the empty type signature case.
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-84645ClosedFixed
OCSP stapling log messages should indicate response validity
After an OCSP response acquisition attempt, several events might have occurred:
* A network error
* One or more responder errors
* A response which failed signature validation
* A response which failed to meet policy
* A response which indicates the subject certificate was revoked
* A response which indicates the subject certificate was valid
There are probably a few other edge cases. Currently, the main OCSP stapling loop dispatches requests, then logs message {{577163}} which includes
Fixed Versions
8.3.0-rc0
SERVER-139180In Progress
Update join path resolver to query types after $match
In agg_join_model.cpp there is this code which determines which pipeline stage to pass into the graph's canPathBeArray:
For a subpipeline, we see that the first subpipeline stage is selected.
So the call ends up being canPathBeArray("path", <first stage>).
{quote}canPathBeArray: Returns false if the path as seen at the input of 'stage' can be proven to not be an array.
{quote}
This means that the first stage (the $match) will not be considered for type inference, instead returning the t
SERVER-139391In Code Review
Implement type narrowing for comparisons expressions $allElementsTrue
Final ticket in the agg ops narrowing.
SERVER-138470In Progress
Preserve the LiteParsed value overrides for extension $listSearchIndexes
Specify query-settings-rejection bypass, isSearchStage, and canRunOnTimeseries override values (true).
We currently do not allow extensions to provide these per-stage LP replacement values (ex: override LPDS::isSearchStage() = true) to avoid unexpected behavior that could stem from a extension stage wrongly declaring itself as a $search stage, for example.
But, the server could instead provide the policy with the desired overrides, something like this.
For canRunOnTimeseries:
'canRunO
SERVER-138475Backlog
Add host-side first-demand target resolution for the extension $listSearchIndexes stage
Add a host path that resolves the target for the extension $listSearchIndexes stage on first demand and delivers the backing namespace, UUID, and view to the extension, calling fetchCollectionUUIDAndResolveView with failOnTsColl=false to match native. This makes a time-series target return an empty result rather than erroring, while failOnTsColl=true stays the default for the search-index commands and any other extension stage. Depends on SERVER-138470 admitting the stage on time-series targets.
SERVER-127760ClosedFixed
Use shorter license header in source code
Legal has approved replacing our current 28 line header comment with:
This also removes the year, which was frequently forgotten about when copy-pasting to a new file.
Plan of attack with a week or so between steps to minimize disruption of in-progress PRs, and avoid mixing real changes that need to be reviewed with wide, mechanical ones:
# Update the license check in mongolint.py to support both forms, and update code generators to use the new one. (Announce the change as soon as thi
Fixed Versions
9.1.0-rc0
SERVER-98423ClosedWorks as Designed
Results differ depending on the order of operands in $sum inside $group with optimizations on
This came up while investigating BF-35904. The minimized query looks like this
The explain output for optimizations turned has the operands swapped like so
while with optimizations off the operands remain in the same order
this change comes from [ExpressionNary::optimize()|https://github.com/10gen/mongo/blob/27d91db841953662a6780bcc4faba34dc257bd5a/src/mongo/db/pipeline/expression.cpp#L3066] as it will hold the ExpressionConstant operand in a [separate array|https://github.com/10gen/mon