-
Type:
Task
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Workload Resilience
-
Minor Change
-
v9.0
-
Workload Resilience 2026-11-23
-
0
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Problem
similar to a issue in https://jira.mongodb.org/browse/SERVER-131934 we previously seen in delinquent_ops.js. deprioritization_stats_and_observability.js flakes in should report 'priorityLowered: true' in $currentOp while active.
The test waits for setPreYieldWait to be entered 3 times, then takes a single $currentOp snapshot and asserts priorityLowered: true. With batchSize: 3 and internalQueryExecYieldIterations: 1, those three pre-yield pauses are the entire find. The find returns a cursor, toArray() issues a getMore (new OperationContext, flag reset to false), and waitForFailPoint (100ms poll) can return after the find has already finished. $currentOp then observes the getMore still on its first ticket, so the flag is false.
This is a test race. The original find's slow log shows priorityLowered: true, and the following slow-query-log case in the same file passes.
Seen on Windows in BF-47067.
Fix
Pin the original find after heuristic deprioritization has run, then read $currentOp:
- Enable setPreYieldWait (comment-filtered) so setPreYieldWaitDeferred can run.
- Hang with setPreYieldWaitDeferred skip: 1: skip the first pre-yield, let the op yield and
reacquire (admissions >= 1 sets priorityLowered), then hang on the next pre-yield while the
find is still active. - Filter $currentOp with command.find so a getMore cannot match.
- Turn both failpoints off in finally before joining the parallel shell.
- is related to
-
SERVER-131934 delinquent_ops.js sleep hit internalQueryJavaScriptFnTimeoutMillis under CPU pressure
-
- Closed
-