[Mimedefang] MD unable to create spool directories

Clayton, Nik [IT] nik.clayton at citigroup.com
Thu Jan 23 05:42:01 EST 2003


Caveat -- still using 2.16, until I find the time to package and test a
more recent version.  So this issue may have gone away by now, although
by reading the code and the changelogs I don't see how.

We're seeing an odd issue with MD, and, annoyingly, there doesn't seem
to be much pattern to it.

The environment is Solaris 2.8, with /var/spool/MIMEDefang being a 
symlink to a tmpfs filesystem for efficiency reasons.

After running for between 10 to 15 days MD starts complaining that it
can't create files in the /var/spool/MIMEDefang/mdefang-* directories.

e.g.,

    Could not create /var/spool/MIMEDefang/mdefang-3E2DD335-1: File exists

This goes on for a period of time, sometimes interspersed with

    Could not create /var/spool/MIMEDefang/mdefang-3E2DD33E-0: Error 0

which isn't terribly informative.  Eventually, things get worse, logging

    Could not create mdefang directory in /var/spool/MIMEDefang: File exists

Which, from looking at the code, is one of those things that just
shouldn't happen (8,192 different directory names should have been 
attempted before we see that error).

When this happens, checking out the box in question shows that 
/var/spool/MIMEDefang typically has only a handful of directories in it,
is nowhere near being full, and has plenty of inodes available.  Stopping
and starting sendmail+mimedefang is enough to cure the problem.

I'm beginning to wonder whether perhaps mkdir(2) isn't atomic on Solaris
tmpfs filesystems, or that perhaps we're running in to some other subtle
race condition.  When I deploy 2.29 I'll add additional log info to see 
if the code's bailing out early, but for the time being, this is all I've
got.

Does this problem ring any bells with anyone out there?

N
-- 
1        1         2         3         4         5         6         7    7
         0         0         0         0         0         0         0    5
  -- The 75 column-ometer
                                             Contributing to the heat death
Global Messaging, 120 Cheapside, x83331      of the universe since 1973.



More information about the MIMEDefang mailing list