Re: [HACKERS] vacuum analyze problem

> Just did full and clean cvs'ing, cleaning, recompiling on 2 Linux
> machines.
> The problem with vacuum analyze on regression db dissapears, but it's
> still exists when I run postmaster with -B 1024 option.
> usually I start postmaster as
>
> /usr/local/pgsql/bin/postmaster -i -B 1024
> -S -D/usr/local/pgsql/data/ -o '-Fe'
>
> and 6.3.2 is rock solid, no problem at all with vacuuming.
> > Your problem would be easier to diagnose if someone could reproduce
> > it on another machine, but if the test is as simple as trying a
> > vacuum on the regression database then it's easy to test and we
> > aren't finding it happening everywhere :(
>
> I could reproduce the problem myself :-) on 3 Linux machines, running
> different kernels, different compilers etc. This is an excerption from
> debug messages:
> il. Pages 0/0. Elapsed 0/0 sec.
> DEBUG: Index bt_txt_index: Pages 27; Tuples 10000. Elapsed 0/0 sec.
> ERROR: cannot write block -1 of [] blind
> AbortCurrentTransaction
> NOTICE: AbortTransaction and not in in-progress state
> > Keep plugging away at it though because there's always a chance that
> > your particular machine is uncovering an obscure feature of the
> > code...
> 3 particular machines and all I managed to install software, so
> problem could be in me :=)
> What could you suggest I do to track the problem ?

Well there is something about your installations which is different. The
only one I know about is that at least one of your machines is running
the 2.1.xxx kernels, which I don't have any experience with. If we can
confirm that this is then only difference between your machines and
mine, then it is something which we should understand and report as a
new kernel problem. If that isn't the common element, then I'm at a loss
as to why your machines and mine differ in behavior.

I'll interpret the lack of "hey, I'm having trouble too!" messages from
other people running Linux/iX86 as meaning that others aren't seeing
trouble, but it could just be that others aren't trying it yet. Does
anyone else on Linux see this problem? Does anyone on Linux _not_ see
this problem?

- Tom