Database becomes corrupt doe to a failing unique index and consequent creation of duplicate entities

[andrii0lomakin]

Hi,

We have experienced serious melt-down of our graph when a unique index fails due to incomplete re-indexing triggered by a unclean shutdown.

What happens is this*:

A unclean shutdown occurs (we obviously try to eliminate these)
A re-indexing is triggered of an affected index
The indexing does not, for some reason (restart, error etc.), finish
The system restarts thinking the index has been re-created and moves on with an incomplete index
Duplicates are added and now there is no easy way back
This can be easily reproduce by forcing a shutdown during indexing

We are using 2.2.18-SNAPSHOT but we believe we have also seen this on older 2.2.X builds.

IS it possible that the index is marked healthy at the start of the re-indexing process?

Please take a look at this ASAP as it has become very urgent for us to move this 2.2.X based upgrade to production.