Table data corruption with frequent restarts #4123
Description
With H2 version 2.3.232 we are observing a rare / non-deterministic behavior where with frequent restarts eventually data is lost from a table - see keycloak/keycloak#32785
The invalid data is not arbitrary - it contains the values in insertion order, just from a much earlier time. There can be 10-20 restarts between when the now invalid data was valid. Its usage in between is as expected - (values inserted, and selected correctly. Once the data is invalid, the table is still perfectly usable - selects and new insertions seem to work fine.
We have verified that our application is not issuing any sql that would delete the affected rows.
Most importantly this behavior does not seem to occur with version 2.3.230.
We have quite a bit of logging, but have not yet attempted to create a stand-alone reproducer. Let me know what may be of use in getting this addressed.