Tom Lane wrote:
> This sounds like corrupt data --- specifically, 1664639562 showing
> up where a variable-width field's length word ought to be. It
> may or may not be relevant that the ASCII equivalent of that bit
> pattern is Jb8c ... do you work with data that contains such
Not specifically but I cannot rule it out entirely, it 'could' be in one
of the column which may have a combination of uppercase/lowercase/number
otherwise all other text entries would be lowercase.
> If it's not 100% repeatable I'd start to wonder about flaky hardware.
> Have you run memory and disk diagnostics on this machine recently?
I did extensive tests a month or two ago (long crunching queries running
non stop for 24h) which were ok but honestly cannot say I'm not very
trusty in this particular hardware. Would need to put it offline and
memtest it for good obviously.
I moved some data (and a bit more, same table has 35M rows) to another
machine (also 8.1.10 on ubuntu with 2.6.17-10 smp, 2 Xeon 2GHZ instead
of P4 3GHz) and it passes with flying colors (though it's much much
slower with the same settings so I need to check a few things in there,
I had tmp dir topping 3GB so not sure if I could have more in memory)
Thanks for the insight, Tom.
In response to
pgsql-performance by date
|Next:||From: Andrew Sullivan||Date: 2007-11-13 15:21:11|
|Subject: Re: Curious about dead rows.|
|Previous:||From: Jens-Wolfhard Schicke||Date: 2007-11-13 10:24:42|
|Subject: random_page_cost etc. per tablespace?|