[Dovecot] Dovecot hangs randomly

[Jacob Elder]

I have been run into a weird problem with dovecot. We have 4  
identical servers running dovecot, serving both IMAP and POP3  
clients. Every few hours, clients start timing out after the PASS  
command. Here is what the process tree looks like at that time:

  6718 ?        Ss     0:00 /usr/sbin/dovecot
6719 ?        S      0:00  \_ dovecot-auth
6881 ?        S      0:00  |   \_ dovecot-auth
6883 ?        S      0:00  |   \_ dovecot-auth
6890 ?        S      0:00  |   \_ dovecot-auth
6928 ?        S      0:00  |   \_ dovecot-auth
6929 ?        S      0:00  |   \_ dovecot-auth
6934 ?        S      0:00  |   \_ dovecot-auth
6936 ?        S      0:00  |   \_ dovecot-auth
6939 ?        S      0:00  |   \_ dovecot-auth
6720 ?        S      0:00  \_ imap-login
6721 ?        S      0:00  \_ imap-login
6722 ?        S      0:00  \_ imap-login
6782 ?        S      0:00  \_ pop3-login
6808 ?        S      0:00  \_ pop3-login
6884 ?        S      0:00  \_ pop3-login
6885 ?        S      0:00  \_ pop3-login
6886 ?        S      0:00  \_ pop3-login
6887 ?        S      0:00  \_ pop3-login
6888 ?        S      0:00  \_ pop3-login
6889 ?        S      0:00  \_ pop3-login
6931 ?        S      0:00  \_ pop3-login
6937 ?        S      0:00  \_ pop3-login
6940 ?        S      0:00  \_ pop3-login

After running invoke-rc.d dovecot stop, the process tree looks like  
this:

6782 ?        S      0:00 pop3-login
6808 ?        S      0:00 pop3-login
6881 ?        S      0:00 dovecot-auth
6883 ?        S      0:00 dovecot-auth
6884 ?        S      0:00 pop3-login
6885 ?        S      0:00 pop3-login
6886 ?        S      0:00 pop3-login
6887 ?        S      0:00 pop3-login
6889 ?        S      0:00 pop3-login
6890 ?        S      0:00 dovecot-auth
6928 ?        S      0:00 dovecot-auth
6929 ?        S      0:00 dovecot-auth
6931 ?        S      0:00 pop3-login
6934 ?        S      0:00 dovecot-auth
6936 ?        S      0:00 dovecot-auth
6939 ?        S      0:00 dovecot-auth

If I run "killall -9 dovecot-auth pop3-login; invoke-rc.d dovecot  
start", service resumes for a few hours. This happens both on 2.6.8  
and 2.4.27. This is Debian testing, 512 MB RAM, about 20 users per  
server. We use libnss-ldap and libpam-ldap for all users other than  
root. Samba, SSH, saslauthd (for Postfix), login, etc all work as  
expected. There is a 5th server that is identical to the other 4  
except is NOT using libpam-ldap, and dovecot does not hang on this  
machine. Any ideas?

--
Jacob Elder