There have been several occurrences in Atlas where one or more secondary nodes already have an index built that the primary does not have. We think this is due to interrupted rolling index builds. Creating the same index on the primary may result in the index build stalling indefinitely as the secondary nodes with the index will return early and skip voting for the index's commit quorum.
The workaround is to adjust the commit quorum using the setIndexCommitQuorum command.
There's no indication of when this happens, so it would be helpful if we can add a log message when a secondary node is skipping the index build to allow the user to take action or for Atlas to detect this automatically via log ingestion.
- is related to
-
SERVER-45828 write js test for rolling index builds
- Closed
-
SERVER-77311 Add a new log message when a secondary node is skipping a two-phase index build with a subset of indexes built
- Closed
- related to
-
SERVER-73256 [v4.4] Disable rolling_index_builds_interrupted.js from running on single phase index build variants
- Closed
-
SERVER-71768 Have secondary nodes vote for the index build commit quorum if the index already exists
- Closed