6.0-release hanging without a clue (gmirror related?)

[Ville Lundberg]

Hi,

I have a 6.0-release-p4 system that is hanging constantly after about a
week of uptime. Nothing is printed to the logs - it just hangs, the HD
light is stuck on. I don't know if anything is printed out on the console,
as the system is at client's premises.

The system is a Epox 4PDA3I mb (Intel ICH5 disc controller), Pentium 4
2,6GHz, 1Gb ram, 2 x WD Raptor 36,7Gb SATA harddrives on gmirror. It is
very lightly stressed, as it's used for one database application only.

Actually, I don't know if the system freezes completely, as it is used
only thru Apache - these crashes are noticed by client when the app no
longer responses. After a cold reboot, gmirror loses one of the hds
(component broken, skipping).

I have two theories: 1) gmirror (or fbsd sata stuff) is the cause for
crash. The HD light thing is what makes me suspect this (hd action when
freezing). And, when rebuilding the mirror, it failed with WRITE_DMA
timeouts. I cleaned the first and last blocks of the failing hd, and then
I was able to add it back to the mirror. Manufacturer disk diagnostics did
not report any errors on either hds - so the cold reboot is the cause of
dropping the hd from gmirror.

2) Apache 1.3.34, MySQL 4.1.16, mod_perl 1.29, Perl (don't remember exact
version, but pkg_add -r perl from 6.0-release), is the fault. The system
was upgraded recently when the new hds were installed, from FB 4.10-rel,
Apache 1.3.19 (no mod_perl), MySQL 4.0.18 to the above mentioned. With
4.10, it was rock solid with nice uptimes like 176 days until maintenance
had to reboot it...

Anyone have ideas how to get to the bottom of the problem - to know why it
freezes in the first place? Or know if any of the software versions
mentioned above have some issues? I can provide dmesg and such if wanted.
  --Ville