Details
-
Bug
-
Resolution: Fixed
-
Major
-
5.1.3, 5.5.6, 6.0.4, 6.5.1, 6.6.1
-
Triaged
-
1
-
No
-
KV-Engine 2021-Jan
Description
Currently when we do a rollback we roll back all unpersisted items and reset the checkpoint manager. When we reset the checkpoint manager we reset the VBucket::dirtyQueueSize but not the VBucket::dirtyQueueAge (and some others). This causes us to then miscount queue_age which is a vBucket/aggregated Bucket level stat that is shown in the UI as an indication of how long everything has taken to be persisted. Due to the way we calculate the stat, the numbers can be off by massive amounts and the returned value can "underflow" giving the indication in the UI that we have accumulated centuries worth of unpersisted mutations.