Details
-
Improvement
-
Resolution: Unresolved
-
Critical
-
7.6.0
-
None
Description
If indexer restarts while rebalance is in progress, then both transfer token cleanup and stream reestablishment will race with each other.
Currently, each dropIndex would wait for stream request lock before returning the response. This means that if request to projector is stuck, then rebalance cleanup will also be stuck. Until cleanup happens, next rebalance can not happen blocking cluster wide progress.
The ask for this improvement is to de-couple dropIndex from stream request lock so that both these operations can happen asynchronously.