Checkpoint error on startup after clean shutdown
When tranlogs are on raw partitions, if a clean shutdown is done shortly
after starting a new transaction log (so there are very few records in
the final tranlog before shutdown), the subsequent startstone will incorrectly
report a checkpoint mismatch.
This error message is incorrect; there is actually no problem with the
extents, all data was written correctly. The startup checks include a
check for filesize, as well as the block id checks; both checks use the
same error message. The filesize check is failing due to filesize not
being recorded correctly for very small raw partition tranlogs.
Workaround
(1) Verify that the stone was cleanly shutdown by checking the stone log.
A clean shutdown includes via stopstone, System shutDown, or kill -TERM.
The stone log will show messages of either of these forms:
SHUTDOWN command was received from user ...
... < other messages >
Now stopping GemStone. or
SHUTDOWN command was generated by SIGTERM handler.
... < other messages >
Now stopping GemStone.
(2) Archive and remove the current tranlogs
(3) Restart stone with -N -R
(2) Execute SystemRepository commitRestore. This will warn of restoring
without previously execution restoreFromCurrentLogs, but you can ignore
this; all of the data was checkpointed as part of the shutdown.
Last updated: 8/16/12