-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Query Execution
-
Fully Compatible
-
ALL
-
200
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Overview
IncrementalFeatureRolloutContext::set() trips tassert(13002310, "Refusing to replace an IFRContext whose egress metadata has already been cached") at src/mongo/db/ifr_context_decoration.cpp:75 — tracked as BF-45368. The tassert fires intermittently across many tasks/variants/projects but has resisted every attempted deterministic repro so far, and because a tassert logs its backtrace and then lets the process continue to a clean shutdown, no core dump is produced from any of the observed CI occurrences.
Background
BF-45368 has recurred at least 9 times across concurrency_sharded_replication_rate_limited and sharded_collections_jscore_passthrough_rate_limited on numerous variants/projects within a few days.
Scope of Work
* src/mongo/db/ifr_context_decoration.cpp — temporarily replace the tassert(13002310, ...) in IncrementalFeatureRolloutContext::set() with an invariant(...) carrying the same message, to force a crash (and core dump, where the variant collects one) instead of a tripwire-and-continue on hit.
Technical Notes
* Reverting to tassert(13002310, ...) before this PR (or a follow-up) is merged is required — this ticket exists purely to gather crash diagnostics, not to change production behavior.
- blocks
-
SERVER-133148 Avoid using semi-initialized opCtx when retrying getMores
-
- In Code Review
-