Hi
One of our production systems was running 7.4.1 for a few months, when
suddenly some queries that used a specifiy table (a cache table) started
crashing the backend.
A colleague of mine "fixed" the problem by simply dumping and rebuilding
the affected table (That was possible since it was only a cache for the
results of slow queries).
About 2 weeks later, the problem reappeared - this time affecting more
tables. It started to analyze the problem, and found out about the
alignment-bug in 7.4.1. I upgraded to 7.4.2, and fixed the system-tables
according to the 7.4.2 release-note. But this didn't really help - the
"analyze table" issued after fixing the system-tables exited with an
error about an invalid page header in one of our tables. Dumping the
database was also impossible at that stage - some tables would cause pg_dump
to either abort, or to silently block (we had it running for about 10
minutes, and it didn't output a single line in that time).
I finally fixed the problem by dumping all relevant tables "by hand", and
restoring them into a clean install of 7.4.2.
Since that 7.4.2 release-note only talked about crashing queries due to the
7.4.1 bug, but not about data-corruption occuring, I wondered if the
symptoms I have seen are related to the alignment bug in 7.4.1 or not.
The affected tables where all updates very frequently, and were quite large
(about a million records each). The data is comes from daily imports, which
delete the old records, and insert the new ones inside a transaction.
I a backup of the corrupted postgres-data, so I could do further analysis
if necessary.
greetings, Florian Pflug
