Details
-
Bug
-
Resolution: Gone away
-
Major - P3
-
None
-
3.2.11
-
None
-
ALL
-
Description
I got a rollback after stepping down a primary in a mongo cluster with 3 nodes. From mongo's documentation:
The method then initiates a catchup period where it waits up to secondaryCatchUpPeriodSeconds, by default 10 seconds, for a secondary to become up-to-date with the primary. The primary only steps down if a secondary is up-to-date with the primary during the catchup period to prevent rollbacks.
No networking issues were observed during that time. So, I wouldn't expect a rollback to happen.
It seems the rollback happened due to a primary election, which happened right after the step down (in this particular cluster the nodes have different priorities). However, I don't see anything in the documentation saying that primary elections may cause rollbacks either. In fact, I was not able to reproduce the problem only doing step downs or only swapping priorities.