Re: naming of wal-archives - Mailing list pgsql-general

>>> If your command does overwrite, then the server currently emitting the

>>> 8D files will become unrecoverable once those files start getting

>>> overwritten.  If it refuses to overwrite, but returns a zero status,

>>> then the server currently emitting 6D would become unrecoverable once

>>> it reaches 8D and its "archived" files are not actually being archived

>>> but are getting deleted from the local pg_xlog anyway.

>>

>>

>> Would it not be easier to archive the different servers to different

>> directories and eliminate the possibility of name collision between servers?

>Easier?  I would say that that is the only sane way of doing it.  I

>was pointing out the consequences of messing it up.  A proper

>archive_command will save you from some self-inflicted disasters, but

>that does not mean I'm recommending that you should invite those

>disasters on yourself.

>If the original author is in a production environment, he desperately

>needs to figure out what is going on, especially so if archive_command

>is not tested and verified to obey its contract.

>Cheers,

>Jeff

Thanks for your answers. 

Yes, we are in a production environment and there are definitely no two masters writing to the wal-archive directory, at least none that i am aware of. And i can assure you that we are not intentionally working on inviting disasters either :-). 

Point is, the ..6D.. line started showing up exactly after i set up the archive-command (which is btw: archive_command = 'test ! -f /data/backup/walbackup/%f && cp %p /data/backup/walbackup/%f', running on Ubuntu Linux Server)

The situation is as follows:

All concerned machines are running 9.2.2 64-bit on Ubuntu Linux Server 12.10, installed from source, all following exactly the same procedure. We have a hot-standby running to a different location over a rather thin line running since version 9.1 came out. That worked 

flawlessly, we only were bitten by autovacuums to prevent XID wraparounds that generated relatively high wal-volume and we 

were not sure whether the network connection could keep up with it before deleting wal-files. Since we had to physically transfer a backup once for other reasons, we set wal_keep_segments to 8192 to have enough fallback-time. 

The total size of the database is currently at 313 GB and we are in the process of rethinking our backup/emergency-strategy (currently daily full dumps + hot-standby for read-only queries in the other location + one hot-standby in the same office).

So we set up another machine, intended to become a another hot_standby later, but for now just use it to experiment with pg_receivexlog. Since our current backup-strategy does not allow for PIT-recovery and we have been thinking about wal-archiving all the time, we decided to implement it and thats what we did. Now we have the two "lines" in the archive-directory.

Could the the high number of wal_keep_segments have an impact ?

Does the fact that there already were a lot of existing wal-files when i set up archiving and the archive-command have an impact ?

Jeff, you wrote:

>> And how would i restore the needed file names for recovery

>> if i decide to keep one base-backup und then a very long chain of wal-files

>> ?

>There should be no need for that.

When you said there would be no need for that, did you mean restoring the files for recovery or keeping a base-backup and the chain of wal-files ?

I understand that the archive-command is responsible for not overwriting wal-files. But if that situation occurs, and if i understand you correctly it will, what do i do ? 

If the wal-archive files will be overwritten at some point in time, how is it possible to only have one single base-backup at time-point t and from then on only store the following wal-files ( however inconvenient that may be ) to be able to restore to any point in time after time t ? 

Thanks,

Neil