expireover core (perhaps an ov3search bug)

[Russ Allbery]

Fletcher Mattox <fletcher at cs.utexas.edu> writes:

> This is essentially the same stack trace I reported on Feb 29.  My
> overview file was definitely corrupted.  Actually it's fairly
> reproducible.  Just send expireover a SIGTERM and wait for news.daily to
> run.  Expireover segfaulted every night until I manually unlinked the
> corrupt overview file.

What I'm not understanding is how it's getting corrupted in the first
place.  The index entry is clearly wrong, and it's being read directly out
of the file.  But the only places that write index entries are OV3addrec
and OV3packgroup, and when expire is running the index and overview files
are being written to new files that only replace the old files after
they've been successfully written.

I don't see how this could possibly create the corrupted index entry with
a massive offset like was reported.  It seems like something must be
writing out junk to that file, but the file is unliked and created anew
during expire.

So I'm very confused....

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>