Victor Hugo dos Santos <listas.vhs@gmail.com> writes:
> On Tue, Jan 4, 2011 at 1:05 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> It looks like a corrupted-data problem from here. �You need to isolate
>> and delete the bad row(s).
> # su - postgres -c "psql bacula -c 'SELECT md5 from public.file OFFSET
> 1417610 LIMIT 1'"
> server closed the connection unexpectedly
So you've got some rows that are corrupted badly enough to crash the
backend :-(.
> In others words.. if I run the same command two times, I get a error,
> and if I rerun the same command, work !!!
When you're running long seqscans like these, you need to turn off
"synchronize_seqscans" to get reproducible results. With that flag
turned on, scans may start from somewhere in the middle of the table
instead of always starting from the beginning.
regards, tom lane