The following bug has been logged on the website:
Bug reference: 9187
Logged by: kevin
Email address: kevin.hughes@uk.fujitsu.com
PostgreSQL version: 9.2.3
Operating system: Centos 6.2 (64 bit)
Description:
I am unsure if this is a bug but other than a h/w problem I cannot explain
it (and there is no evidence of any h/w problem so far).
We have been running an instance of PostgreSQL on a cloud server for
approaching a year now. It provides the underlying DB for a Confluence wiki
installation. So far everything has gone well - postgres just did the job
:).
Sometime between the morning of the 31st of January and the morning of the
3rd of February the database was corrupted. During that period there was
little or no activity - other than the nightly postgres dump
Looking at various logs we found this error: ERROR: unexpected chunk number
110 (expected 106) for toast value 223972 in pg_toast_80768;
Looking at our dump we discovered they had been failing - they were short.
Rerunning these online showed the pg_dumpall failing with toast errors and
creating a small dump file
I used psql to look at various tables and found 3 corrupt table entries -
all were toast table entries.
Investigation showed that the entries had not been intentionally changed by
the DB s/w since 2013 (entries are time stamped by Confluence)
On the evening of the 10th of February the database problems were fixed â
the broken entries were removed.
However on dumping the db a short dump was created although no errors were
reported. On investigation a further corruption was located â a corruption
that was NOT identified previously although the new corruption was detected
in exactly the same way as the previous three corruptions. Again the
corrupted entry had not apparently changed since 2013. NOTE â the database
use had been limited to read only access since the first corruptions were
discovered
Based on the evidence the suggestions from the internet point to a h/w
fault.... however there is no other evidence that a h/w fault has occurred
I regret I don't know what evidence to provide - or what evidence remains
from when the corruption occurred.