Hi,
we have a streaming replication running and kind of suspect that the slave contains dead files caused by an abort of a
hugetransaction. I'd like to ask how we can be sure that those files are dead.
The details are:
* PostgreSQL 9.1.2 on x86_64-unknown-linux-gnu, compiled by gcc-4.6.real (Debian 4.6.2-9) 4.6.2, 64-bit
* streaming replication with hot-standby
* We moved a huge table from one tablespace to the other
* Unfortunately the disk-space on the master host exceeded during the transaction, which caused a transaction abort and
furthermorea server-crash because WAL files could not be written
* We added more disk-space and restarted the master
* After restarting the master, the slave continued to read the WAL until the transaction abort
* Now we have files on the slave which we suspect to be dead:
332166.27
...
332166.2
332166.1
* These files are located in the destination tablespace (where we tried to move the table to)
I suspect those files to be dead since the following query returns 0 rows:
# select * from pg_class where relfilenode=332166;
Is this a sufficient condition to delete all files $relfilenode.* ? Is relfilenode unique per database or per cluster?
Is there any explanation why this situation lead to dead files?
Thank you!
Regards,
Andreas