- YugabyteDB - 2.4.4 or earlier
This issue only affects systems using colocated tables. See the YugabyteDB documentation section on colocated tables for more information about colocated tables.
After deleting a colocated table using the SQL command DROP TABLE colocated_table, restarting a tablet server (tserver) may cause the tserver to enter a crash loop. A fatal error message similar to the following is recorded in the tserver logs:
F1109 16:51:33.201021 7222 tablet.h:495] Check failed: _s.ok() Bad status: Not found (yb/tablet/tablet_metadata.cc:368): Table <unknown_table_name> (0000b8de00003000800000000000a803) not found in Raft group 7294fe7dbfb34a6788cc28299b3d8c2e
When a tablet server is restarted, the data for each tablet is read from the corresponding SST data files, then the write ahead logs (WALs) for any transactions that have not yet been flushed to the SST files are replayed to bring the tablet up to date.
If the WALs for a colocated table contain an ALTER TABLE operation and the table has since been deleted using DROP TABLE, the tserver will crash when the ALTER TABLE operation attempts to access the deleted table.
This issue is resolved in the following YugabyteDB releases:
- YugabyteDB 2.4.x starting in version 2.4.5
- YugabyteDB 2.6.x
- YugabyteDB 2.8.x
For further details, see YugabyteDB GitHub Issue 6096.
Follow the instructions in the Upgrade the YugabyteDB software section of the documentation to upgrade the affected Universe to version 2.4.5 or newer.
Upgrading the YugabyteDB software will correct the problem for currently affected Universes and prevent the issue from occurring in future.
Please sign in to leave a comment.