Affects Version/s: 5.6.41-84.1
Fix Version/s: None
This could be a duplicate of PS-4283 and I doubt we'd be able to reproduce it, but we might have some information to help you fix this.
Our application database is split among several TokuDB and within each TokuDB, it's always the same table (~32e7 rows, index file is almost 4GB, compressed) that's corrupt. It happened after heavy operations so trying to reproduce would take time. No sensitive data so we could send it, or give an SSH access to the DB. But we must free space on the server and we'd like to know quickly if we should save anything.
Here, the corruption always affects the PK, the secondary index is fine. We have different kinds of corruptions:
- PK is not sorted
- PK has duplicate and unsorted keys
- PK is sorted and unique but differs from secondary index: missing records
That's the first time we get such a critical issue with TokuDB and this is a blocker for us. Although TokuDB is faster than InnoDB & RocksDB for our use, we're going to check again the performance of RocksDB.
This is with MariaDB, version 10.2.22