[Bucardo-general] Swap replication latency

[David Christensen]

On May 1, 2011, at 8:36 PM, Greg Sabino Mullane wrote:

>> Last good:            40m 30s (time to run: 46m 19s)
>> Last good time:       May 01, 2011 16:15:24  Target: xxxxxxx_remote
>> Ins/Upd/Del:          8674 / 2501 / 11
>> 
>> Any ideas?
> 
> You need to check your logs and follow a single KID process to see 
> where all the time is being taken. Feel free to be aggresive in 
> purging the bucardo_delta table: that's a common cause of delay. 
> To be specific, run:
> 
> SELECT bucardo_purge_delta('5 minutes'::interval);
> 
> followed by:
> 
> VACUUM VERBOSE bucardo_delta;
> VACUUM VERBOSE bucardo_track;
> 
> You could also try a vacuum full on those, but that will block 
> all your replicated tables until it finishes, so be careful.
> 
> Also make sure your q table is getting cleaned up. You can run:
> 
> SELECT bucardo_purge_q_table('5 minutes'::interval);
> 
> But it's probably faster and simpler to just truncate the table 
> once Bucardo is stopped. You will temporarily lose some of 
> the "Last good" stats, but might be worth it.
> 
> However, the logs are really the best place to start to see 
> where all the time is being taken.


Is there any sort of log analysis tool for this, or are the logs specific enough per installation/schema that that makes it fairly impractical?  Seems like it'd be nice to be able to standardize some parser and perform any common analysis if possible.

Regards,

David
--
David Christensen
End Point Corporation
david at endpoint.com