Errors during expire

[Alex Kiernan]

Joe St Sauver <JOE at OREGON.UOREGON.EDU> writes:

> >We've a transit box which persists in panicing every few days (the
> >vendor believes hardware...), but every time we end up having to fix
> >the history file as it ends up with truncated lines or a spurious
> >Hole, or...
> 
> I'm actually seeing pretty bizarre crap getting written to history, e.g.,
> doing:
> 
> perl -ne 'print "$.: $_" if length ($_) > 200' history
> 
> on a box that recently developed history related problems showed a variety
> of lines that contained non-printable characters (copy of 21 such entries 
> available off the list on request if you're interested, not included here
> due to the non-printable nature of the lines); undoubtably there are other
> bad (shorter) lines also in there. The presence of those spurious lines
> makes me think that some buffer is getting stepped on, maybe? 
> 

If the garbage starts aligned with the start of a filesystem block,
I'd guess its more likely you had an unclean shutdown and had fsck
wire in the wrong pages when it came back up, or if you've got it on a
journalled file system the meta data replay wired in pages where the
data hadn't made it to the disk (unless you've got it logging data as
well as meta data, but WAFL's the only one I know off the top of my
head - but it looks like VxFS will do it too, didn't know that!).

With -CURRENT I've not seen any problems on both our transit and
writer boxes in the history code, other than after panics.

-- 
Alex Kiernan, Principal Engineer, Development, THUS plc