[Pan-users] Re: Weird cache problem

[Jim Henderson]

On Sat, 29 Aug 2009 04:38:20 +0000, Duncan wrote:

> Jim Henderson posted on Sat, 29 Aug 2009 03:54:04 +0000 as excerpted:
> 
>> Come to think, there is a third oddity that I just realized was new -
>> occasionally when hitting groups on forums.opensuse.org, a group's
>> headers will completely redownload, and old messages (that haven't
>> expired) will show up as unread.
> 
> That one's almost always a server oddity -- due to xref: renumbering.
> However, in your case, given the other issues, it sounds like you may
> have more complex problems.

I think in this case it's not a server issue; the server is the same 
server (and I've actually administered the server in the past) - the 
group didn't do that with the old name/address, just the new one.  But 
it's the same physical hardware.

> You don't happen to have ECC backed memory, do you?  If for some reason
> your memory (or for that matter, your disk, or maybe the bus connecting
> the disk to the machine) is going bad, you may be getting bit-flips that
> are turning one thing into something entirely different.  ECC backed
> memory is designed to try to detect and correct such issues with memory,
> and there are disk based solutions as well, but neither one is
> particularly common yet, and ecc memory is much more expensive as it
> requires additional chips on the memory modules (and because they /are/
> more expensive, they're not as common, which makes them even /more/
> expensive, ecc memory is thus normally considered a server option and
> rarely available in desktops/laptops).

Yeah, I don't have ECC memory, but if I was having memory problems, I'd 
expect other weirdness in the system as well.  The behaviour of pan is 
the only odd thing I'm seeing, and just since I switched that server 
setting.

> I'd try something like memtest.  Just because it says you're clean
> doesn't necessarily mean you are (it doesn't stress the memory bus, just
> the cells, so if it's memory that's barely within tolerance at the rated
> clockspeed, it may not show up, I had that problem with some memory at
> one point), but if it says your memory is dodgy, better replace it
> before you start having worse issues.
> 
> And check the cable seating to your drives, etc, as well.
> 
> Whatever's happening, it does seem strange, that's for sure!  FWIW, one
> of the best indicators I had a problem with my memory (the memory
> clocking issue mentioned above, it went away when I declocked the memory
> a notch) was that bzip2 would often complain trying to decompress
> something, as it's well checksummed.  Big compiles would often error out
> as well, but big compiles aren't so common on binary distributions.  But
> try bzip2-ing a bunch of files, then bunzip2ing them, and see if bunzip2
> complains about any of them.  If it does, you're a sure candidate for
> physical system issues, memory, disk, otherwise, but somewhere!

I'd normally agree, but the system is pretty heavily used - video 
encoding going on in the background at the moment, lots of compression/
decompression going on, and encrypted filesystems in use.  A fair amount 
of things that would reflect a problem quickly if it were hardware-
related.

On the weird cached articles issue, I just duped it and the cache files 
are messed up.  I grepped the text and found both cache files.  Hadn't 
noticed before, but the message text that *should* be there is appended 
to the end of the correct file.  Threading looks like a likely culprit - 
I'll reduce the number of concurrent tasks across all servers to 1 each 
and see if that helps.

Jim

-- 
 Jim Henderson
 Please keep on-topic replies on the list so everyone benefits