-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Storage Execution
-
Fully Compatible
-
ALL
-
Storage Execution 2026-11-09
-
200
-
None
-
None
-
None
-
None
-
None
-
None
-
None
jstests/core/index/index_many2.js fails with expected 64 to equal 63: Expected 64 indexes. when its final createIndex({z: 1}) is interrupted by a replica set state change. Observed as BF-47033 on disagg_replica_sets_transition_to_csrs_jscore_passthrough, whose ContinuousTransition hook repeatedly restarts the replica set nodes while tests run.
Root cause
The test's final phase (drop one index, create z_1, assert the index count) does not tolerate interruptions:
- The transition hook terminated both nodes of the 2-node disagg replica set while the test was in its final steps. Disagg fixtures terminate the primary before stepup (see should_stop_primary_before_stepup, SERVER-131909).
- The z_1 index build was killed by the intent registry during the primary's termination (Cannot register WRITE intent due to ReplStateChange), and createIndexes failed with InterruptedDueToReplStateChange.
- The suite's network_error_and_txn_override retried the command once. The retry failed with a non-retryable error, most plausibly IndexBuildAlreadyInProgress or IndexBuildAborted, while the interrupted build was being cleaned up or recovered on the new primary. SERVER-110248
tracks the underlying build-cleanup race. - The shell's DBCollection.prototype.createIndex does not throw on command failure, and the test neither checked the response nor retried, so the assertion on the index count observed a stale count of 63 instead of 64.
The initial index-creation phase of the same test already tolerates this with an assert.soon retry loop. The final phase was missing that protection.
Fix
Wrap the final createIndex in an assert.soon retry that checks the command response and the resulting index count, so an index creation interrupted by the transition hook is retried until the index is actually created.