Known Issues
5279 tracked MongoDB server issues from Jira with status, affected versions, and resolution details.
SERVER-138124Open
Membership-based recordId emission and honoring in the oplog path
Flip both ends of the {{rid}} pipeline from the per-collection {{recordIdsReplicated}} catalog option to {{groupId}} membership in one commit: a primary emits rids iff it holds a {{groupId}}, and an applier honors them iff it holds a {{groupId}}.
h2. Details
- Emission: the op observer already sets the {{rid}} ([oplogEntry.setRecordId in op_observer_impl.cpp|https://github.com/10gen/mongo/blob/5fce3b99dea39b4f9453477819cedd935fba9a2c/src/mongo/db/op_observer/op_observer_impl.cpp#L811]). The
SERVER-138309Open
Clear the groupId on standalone document-data writes (C4)
Standalone inserts diverge the rid space, so they must clear the groupId. Decided design: clear at the first standalone insert into an in-scope collection, plus storage-level record store rebuilds. Only new rid assignment can create a wrong rid-to-document mapping: updates are in-place and never change a rid, deletes create missing mappings but never wrong ones. Rolling Index Builds and read-only standalone maintenance (including TTL deletes) preserve membership.
h2. Details
- A one-shot gua
SERVER-138186Open
TTL resume skips owned documents that share an index key with more than ttlIndexDeleteTargetDocs unowned documents
After SERVER-92779, a batched TTL delete on a secondary index resumes from the index key of the last staged document. If an attempt stages only entries equal to that key and deletes none of them, the next attempt starts after the key, because SortedDataInterface::Cursor cannot seek to a (key, RecordId) position. Owned expired documents that share that exact key and sort after the first {{ttlIndexDeleteTargetDocs}} unowned documents by RecordId are skipped for the rest of the pass. Every pass rep
SERVER-139146ClosedFixed
transaction_aborted_under_cache_pressure.js sets CachePressureAbortSessionKillLimitPerBatch at runtime, triggering a ThreadSanitizer data race
Description:
h3. Problem
jstests/noPassthrough/txns_cache_errors/transaction_aborted_under_cache_pressure.js fails deterministically on the ThreadSanitizer variant (enterprise-rhel8-debug-tsan). The test sets CachePressureAbortSessionKillLimitPerBatch via a runtime setParameter, which races with the periodic rollbackUnderCachePressure thread reading the same global. TSAN aborts the server, the connection drops, and the test fails at load:
Reproduced on v8.3 patch 4213 (13/13 runs fa
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139415ClosedFixed
Fix MongoShellProgramSkipExtensionsSignatureVerificationTestCase setup of config defaults
Test case class MongoShellProgramSkipExtensionsSignatureVerificationTestCase directly modifies module level variables in module `buildscripts.resmokelib.config`.
The values are not restored at teardown, thus breaking the isolation of each test.
The purpose of this ticket is to modify the testcase so that it uses a context manager to modify the value, and to limit the change only to the variables which are actually needed.
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-139505Needs Scheduling
Store genId as a metadata field in persistent sample docs
Currently genId is only stored in a sample's _id, not in the metadata fields on each page. For ease of filtering $persistedStats output it would be nice to represent as a metadata field.
SERVER-136023ClosedWon't Do
Prevent interrupt delinquency in $facet and $lookup by checking for interrupt in the sub-pipeline drain loops
$facet and $lookup each run entire sub-pipeline executions inside one {{doGetNext()}} call. Sub-pipeline stages check interrupts every 128 {{getNext()}} calls via {{Stage::getNext}} -> {{{}ExpressionContext::InterruptChecker{}}}, but (a) sub-pipelines processing fewer than 128 docs may never trip the amortized checker (each sub-pipeline has its own {{ExpressionContext}} counter), (b) per-input plan build/optimize/translate cost is check-free, and (c) blocking sub-stages expose their full span in
SERVER-139188In Code Review
Implement type narrowing for the remaining comparisons expressions
Follow up from SERVER-135984. After $eq works, the logic should be expanded to: $lt, $lte, $gt, $gte, $ne. Not $cmp - it will not be supported and it returns a number.
$lt:["$a", "string"] constrains 'a' to be of the type string and types that sort before string. Should probably look like 'minKey|missing|null|undefined|number|string'.
$ne does nothing unless the type is a singleton type - $ne: ["$a", "string"] cannot tell us the type of "a" but a != null can tell us it's not null.
Fixed Versions
9.1.0-rc09.1.0-rc1026
PYTHON-6282In Code Review
shrub.py silently drops allowed_requesters from task definitions
h3. Detailed steps to reproduce the problem?
Pass a documented Evergreen task field to {{shrub.v3.evg_task.EvgTask}} that the model does not declare, for example {{allowed_requesters}}:
Output with shrub.py 3.11.0 and pydantic 2.14.3:
The constructor accepts the keyword, pydantic drops it because the model ignores extra fields, and no error surfaces. The field never reaches the generated YAML.
h3. Impact
Configurations that set {{allowed_requesters}} on a task silently los
Components
Infrastructure
SERVER-41263ClosedFixed
Disallow the empty string as an index key type
At present it is possible to generate an index with an empty string as the key type (eg a definition of \{x:""}). The optimizer seems to be able to use this index when the key is a predicate in the query:
However it is unable to use the index to sort:
{noformat}
> db.repro.find({x:123}).sort({x:1}).explain()
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "repro.repro",
"indexFilterSet" : false,
"parsedQuery" : {
"x" : {
"$eq" : 123
}
},
"que
Components
Index Maintenance
Fixed Versions
4.3.14.2.3
SERVER-96766ClosedFixed
Verify behavior of pre-3.4 index key specs on 3.4+ versions
*Background*
The server only applies stricter index key pattern validation to v:2 indexes (as of SERVER-26659). This means that the server does not apply stricter validation on pre-3.4 legacy indexes. Pre-3.4 index key values could be 0, empty string, numbers other than -1 and 1, non-numeric / string values, etc. See SERVER-11064 for more information. 3.4+ server versions still handle pre-3.4 legacy indexes because these legacy indexes can persist across server upgrades.
*Solution*
We shoul
Fixed Versions
8.1.0-rc0
SERVER-137418In Code Review
Add pre-condition/post-conditions infrastructure for Resharding Recipient
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-137254Investigating
Call SBE fetch index key checks directly instead of via callbacks
Fixed Versions
9.1.0-rc09.1.0-rc1026
SERVER-133899ClosedFixed
Disable minidump in index_secondary_commit_after_scan_error.js
index_secondary_commit_after_scan_error.js is expected to fail by crashing the secondary; on Windows that crash writes a minidump which can hang the process, so disable it.
Fixed Versions
9.1.0-rc09.1.0-rc1022
SERVER-137997Backlog
Join optimization $expr equijoin treats BSON undefined as not equal to a missing field
*Minor* because undefined is a deprecated BSON type.
Aggregation $eq considers undefined equal to a missing field, and classic/SBE $lookup return the row. The JOO hash join buckets fillEmpty(path, null), so undefined never meets the null produced for missing; the index probe seeks undefined while the index stores missing as null.
SERVER-21992Backlog
Inconsistent results when grouping by possibly missing values
My database contains the following documents:
Note that the second document does not have a type field, it's missing. The aggregation pipeline produces inconsistent results sometimes showing 'type' :null in the id field, and sometimes not.
For example, the following aggregation does NOT return the missing type in the key:
Results in:
However, this aggregation, returns back a 'type' : null,
Results in:
Note that the only difference was the comment out of the 'col
Components
Aggregation Framework
SERVER-139502Open
Drop the overrideRecordIdsReplicatedFalse failpoint from apply_ops_mode.js
[{{apply_ops_mode.js}}|https://github.com/10gen/mongo/blob/20753a5d812/jstests/noPassthrough/replication/apply_ops_mode.js] verifies {{applyOps}} command semantics ({{alwaysUpsert}} and {{oplogApplicationMode}} defaults, mode parsing, FCV collection rules) on a standalone. It enables [{{overrideRecordIdsReplicatedFalse}}|https://github.com/10gen/mongo/blob/20753a5d812/src/mongo/db/shard_role/shard_catalog/database_impl.cpp#L107] only because its update ops carry no {{rid}}: with the flag on, the
SERVER-139236In Code Review
Add a blocking interface for Copybara branches
Adds an Evergreen interface for blocking Copybara branch sync through {{copybara_blocked_branches}}. Blocked branches and release tags are skipped successfully while other branches continue syncing.
Notifications default to {{#devprod-build-triager}} and {{#devprod-server-copybara}}. Every Copybara Slack message also reaches {{#devprod-server-copybara}}, including recipient overrides and failures, with duplicate destinations removed.
Documents configuration and tests filtering, tag identity, a
Components
Build
Fixed Versions
9.1.0-rc09.1.0-rc1026