The short story:
We sometimes end up with nulls in users' mailboxes. Delivering a large
(250k) message nearly guarantees that nulls will show up. This is probably
a locking problem. Any ideas how to fix it?
The long story:
We run DCE/DFS on a cluster of AIX 4.1.5 machines. Mail is delivered into
users' home directories in DFS via procmail. Any machine in the cluster
(15 machines) may deliver mail into any user's mailbox at any time. We
have 4 machines that do most (95+%) of the mail delivery.
Users read mail via Pine and UW's imapd.
According to 'docs/locking' in the IMAP4 distribution, Pine and imapd use
flock and dot-locking.
When we compile procmail it says "I will use every type of locking *except*
flock". It says this even if we tell it to just test a DFS directory.
If I understand procmail's src/lock.log then even if we change config.h so
that *only* flock will be tested, all of the tests fails (lots of X's in
(Has anybody seperated procmail's locking tests from autoconf so they can
be run by hand and give results like "flock always fails"?)
I'm sure there's a way out but it looks like we are in this sad situation:
- procmail thinks flock won't work in DFS.
- Pine and imapd think flock is the only way to go.
- procmail, pine, and imapd all use dot-locking but dot-locking is not
Any insight into this problem will be greatly appreciated. (One scary idea
is "make pine and imapd use something besides flock".)
Mark Plaksin http://www.arches.uga.edu/~happy/