On Tue, Sep 9, 2008 at 8:50 AM, Frederik Ramm
<frederik.ramm@geofabrik.de> wrote:
> Dear PostgreSQL community,
>
> I hope you can help me with a problem I'm having - I'm stuck and don't
> know how to debug this further.
>
> I have a rather large nightly process that imports a lot of data from the
> OpenStreetMap project into a PostGIS database, then proceeds doing all sorts
> of things - creating spatial indexes, computing bounding boxes, doing
> simplification of geometries, that kind of stuff. The whole job usually
> takes about five hours.
>
> I'm running this on a Quad-Core Linux (Ubuntu, PostgreSQL 8.3) machine with
> 8 GB RAM.
>
> Every other night, the process aborts with some strange error message, and
> never at the same position:
>
> ERROR: invalid page header in block 166406 of relation "node_tags"
>
> ERROR: could not open segment 2 of relation 1663/24253056/24253895 (target
> block 1421295656): No such file or directory
>
> ERROR: Unknown geometry type: 10
>
> When I continue the process after the failure, it will usually work.
>
> I know you all think "hardware problem" now. Of course this was my first
> guess as well. I ran a memory test for a night, no results; I downgraded do
> "failsafe defaults" for all BIOS timings, again no change. Ran "cpuburn" and
> all sorts of other things to grill the hardware - nothing.
You definitely are suffering from db corruption, and given the number
and differing type of errors, it would seem unlikely that pgsql has a
load of bugs only you are seeing. OTOH, if the bug is hidden deep in
postgis or something, then who knows...
I'd definitely run something like bonnie++ for a few days and see if
it gets HD errors or not.
And definitely run memtest86 for a day or so and make sure you're not
getting any errors there.