[Pgpool-general] online recovery by PITR - broken examples?

[Tomasz Chmielewski]

Jaume Sabater wrote:
> On Mon, Oct 26, 2009 at 10:32 PM, Tomasz Chmielewski <mangoo at wpkg.org> wrote:
> 
>> Yes, this is what I don't understand - these scripts seem to want to copy
>> the whole postgresql data directory. That can't be right, can it?
> 
> If you use rsync, only the differences.

Assuming your database files are really small, yes, that would work.

If your database files are relatively big (say, 10, 50 MB and bigger), 
even a slight change in them will result in rsync copying everything 
from scratch.

Not sure how PostgreSQL organizes that - should I assume it keeps all 
objects in < 10 MB files or so?


>> With 10 GB database, how long will the recovery take? What if the database
>> is even bigger?
> 
> Online recovery does not use the binary log to put the two databases
> in sync, therefore, considering the amount of time and transactions
> that could have happened between the failover and the online recovery
> attempt, how else would you expect it to work?

If we keep binary logs - it just takes to copy the logs from the 
up-to-date server and replay them on the crashed server.


I don't understand why everything should be copied *and* the binary 
logs, too.

It would make sense it should one or the other, but not both.

Could be I don't understand something here, but that's why I'm asking ;)


-- 
Tomasz Chmielewski
http://wpkg.org