Re: recreating point-in-time recovery when tables are in non-default tablespace

Mark Steben wrote:
> I have in the past been successful in creating a hot standby environment as all my tables were in one cluster on one disk.
> The base backup (I used tar) easily copied all tables and then the tar extract easily restored them on our standby server.
> Since the last time this process was invoked I moved about 14 of these tables to a non-default tablespace on another disk
> (to alleviate space issues) Now the base backup does NOT pick up these tables and I am getting errors on the restore side.
>
> Here is a copy the error:
>
> : WARNING: page 36107 of relation pg_tblspc/1057768125/PG_9.4_201409291/16469/2231796047 is uninitialized
> Nov 14 13:46:20 ardbc01dr postgres[58399]: [572-2] @: CONTEXT: xlog redo delete: index 1057768125/16469/2231796047; iblk 36107, heap 1663/16469/2202113342;
> Nov 14 13:46:20 ardbc01dr postgres[58399]: [573-1] @: PANIC: WAL contains references to invalid pages
> Nov 14 13:46:20 ardbc01dr postgres[58399]: [573-2] @: CONTEXT: xlog redo delete: index 1057768125/16469/2231796047; iblk 36107, heap 1663/16469/2202113342;
> Nov 14 13:46:21 ardbc01dr postgres[58398]: [3-1] @: LOG: startup process (PID 58399) was terminated by signal 6: Aborted
>
> Am I out of luck here? Do I need to somehow move these tables back to the original cluster? Or is there another way?

I'd guess that you forgot to run

SELECT pg_start_backup('your_label');

before you copied the file and

SELECT pg_start_backup();

afterwards.

That will cause the backup to be corrupt.

Yours,
Laurenz Albe