Loading LRU dump prevents shutdown

[Peter Zaitsev]

If LRU load is happening Innodb can't shutdown. This is especially bad as it looks like some variables are checked after Innodb initialization is done:

110202 12:46:14 InnoDB: Initializing buffer pool, size = 107.4G
110202 12:46:22 InnoDB: Completed initialization of buffer pool
110202 12:46:22 InnoDB: highest supported file format is Barracuda.
110202 12:46:33 Percona XtraDB (http://www.percona.com) 1.0.14-12.5 started; log sequence number 15021882671828
110202 12:46:33 [ERROR] /usr/sbin/mysqld: unknown option '--log_slave_slow_statements'
110202 12:46:33 [ERROR] Aborting

110202 12:46:33 InnoDB: Starting shutdown...
110202 12:47:53 InnoDB: reading pages based on the dumped LRU list was done. (requested: 163262, read: 162650)
110202 12:48:03 InnoDB: Shutdown completed; log sequence number 15021882671828
110202 12:48:03 [Note] /usr/sbin/mysqld: Shutdown complete

As you can see I used wrong Variable name. Because of this start aborted and started to shut down but it only could complete it after load have been prefetched, wasting few minutes here. For large buffer pools it could be hours.

[Alexey Kopytov (akopytov)]

Bug #920910 was marked as a duplicate of this one.

[Mrten (bugzilla-ii)]

Alright, here is a stab at a patch. This is from apt-get source, not bzr. Compiles, but since a seemingly unrelated test fails (main.percona_innodb_kill_idle_trx) I have not run with this yet.

Is this the right way to go about fixing this?

[Alexey Kopytov (akopytov)]

Yes, the patch looks good.

We should do the same in buf_LRU_file_dump() after bug #686392 is fixed, so we don't end up with an incomplete dump file.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-460