I've been using bogofilter through many versions with
no problems. However, I've encountered a serious
problem when trying to upgrade from 0.92.8 to 0.93.1.
Running `make check' will freeze roughly one time in
two, and what's more will bring down the entire system
so that pressing the reset button is the only solution.
It does not happen at the same place in the check
procedure every time. For example, in one case the last
message was "PASS: t.lock1". Next time, the problems
began after "PASS: t.abort".
The machine in question is a Red Hat 7.3 system; kernel
2.4.20-37.7.legacy, db version 3.3.11. The directory in
which I was building bogofilter is NFS mounted. lockd
is running and HAVE_FCNTL_H is defined in config.h.
What seems to happen is that bogoutil goes into
uninterruptible sleep at some point during the test
procedure. The load on the machine rises to crazy
levels (more than 30); problems develop with NFS and
locking; and after a while the system freezes up. The
only thing I can see in /var/log/messages that looks
relevant are that things like "kernel: lockd: cannot
[un]monitor xxx.xxx.xx.xx" start to appear; but this is
probably a symptom rather than a cause.
I'm aware this report is somewhat vague. Unfortunately,
the only machines I can test this on are the main and
backup servers at work (which is where bogofilter runs
when invoked from procmail). Understandably, I won't be
too popular if I keep crashing these! I can probably
crash the backup server once more though if there is a
useful test that can be run.
If of any use, I can provide the system logs from the
time of the problem; and I also have the bogofilter
build tree in the state it was left during two separate
crashes (checks.xxxx.yyyy directories left around).
The same procedure on a Fedora Core 2 system seems to
work fine. However, I did notice that the test
"t.lock3" failed on one occasion (out of about 10).