filesystem full after many mmap/munmap cycles

[Mark Bucciarelli]

Hi,

I'm running FreeBSD 7.0 RC1 on AMD 64, dual Opteron with h/w raid1 (scsi).

I have an app server that uses mmap a lot.  After running a long batch
(four hours, 5,100+ transactions), I got the message filesystem full
(/usr--ufs, local, soft-updates).  df -i says plenty of space.

I restarted the batch process, and watched app server process
carefully with fstat -p, and it looks to be behaving responsibly.  The
open file list is short, and when I looked up the file names by inum,
they were correct.

Each transaction does a mmap/munap cycle with a (big ?) file (79M),
then copies another smaller file, using mmap to do the copy.  fwrite
failed on the copy operation; FreeBSD said no space.  (When fwrite
failed, I called abort, so I have a core and can see where it
happened.)

pid 29990 (r.fcgi), uid 1001 inumber 359181 on /usr: filesystem full
pid 29990 (r.fcgi), uid 1001: exited on signal 11

Filesystem  1K-blocks    Used   Avail Capacity iused  ifree %iused  Mounted on
/dev/da0s1g   6297070 3380066 2413240    58%  183869 640449   22%   /usr

Is there some file system delay (maybe something related to
softupdates) that could accumulate in some way that would cause the
file system full message?

Please CC me, as I'm not a subscriber.

Thanks,

m