De : mailto:dev@archonet.com
> Cyril VELTER wrote:
> >
> > Is length() supposed to return the very high length in case of corruption
?
>
> You'd have thought it would. The odd thing (if it is data corruption) is
> that you would expect to see something in the server logs about a
> failure to allocate 12345412234124 bytes of memory or some such. Whereas
> all you get is this winsock error.
I have another theory. The message printed by pg_dump :
"pg_dump: Error message from server: out of memory"
is printed in dumpTableData_copy (pg_dump.c) : write_msg(NULL, "Error message
from server: %s", PQerrorMessage(g_conn));
There are serveral places in libpq where the conn error might be set to "out
of memory". I've also discovered that the machine running pg_dump is pretty
tight in ram (256M) and that no swap file is active (oversight after a disk
upgrade). May be this error is simply pg_dump running out of memory and not the
server. This would also explain that the server only report a socket error
(though I would have expected WSAECONNRESET (10054) instead of WSAECONNREFUSED
(10061)).
The 8.0.13 pg_dump used to work just fine on the exact same machine. Maybe 8.2
pg_dump need more memory (or maybe a memory leak ?).
I'm running the dump again after adding some swap space and will monitor
memory usage.
What do you think ?
> > Is there anythning else i can do ?
>
> Could you try CREATE some_table AS SELECT * FROM c2674 WHERE ... to copy
> the large rows within the database. If that fails, the table is
> corrupted but you can identify the problem rows and work around them
> while you dump the data.
I will try that this week end if my theory prove wrong (I need to make disk
space available on the server for that, the table is 60GB).
Thanks,
Cyril
