Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
4.6.0, 5.0.0
-
None
Description
A data loss was detected in CBSE-3576 (MB-24817) due to pre-mature closing of a DCP stream during takeover. We could detect that by manually digging through the logs.
Though the issue is now fixed, we could possibly find some data loss bug in future.
Hence it would be good to record the any data loss during rebalance/failover.
For example, during graceful failover or rebalance, during take-over if last_sent_seqno on an active vb < high_seqno while the stream is being closed, then certainly there is a data loss. This can be recorded in stats and reported to ns-server which in-turn can facilitate notification on UI.
We can also think of other ways of handling this. (More brutal way can be throwing an exception, which may not be preferred in production though)