Just an update... >>>>> On Mon, 21 Jan 2002 09:55:27 -0600, >>>>> Amos Gouaux <[EMAIL PROTECTED]> (ag) writes:
ag> Previously we would see the occasional 'DBERROR db3: X lockers' for ag> some number of X. This X value would increase, but also decrease. ag> Consequently I never really worried about it. However, since the ag> upgrade I have yet see this number decrease. It now seems to be ag> ever increasing. Before restarting the servers this morning I saw a ag> syslog output of 'DBERROR db3: 4052 lockers'. Could this be ag> symptomatic of something more serious, or just a bug in accounting? I restarted the server yesterday at around noon. While overall it does appear to be increasing, at times it does drop down some. Recent messages: Jan 22 08:13:55 DBERROR db3: 1885 lockers Jan 22 08:15:45 DBERROR db3: 1891 lockers Jan 22 08:17:46 DBERROR db3: 1904 lockers Jan 22 08:18:35 DBERROR db3: 1914 lockers Jan 22 08:20:54 DBERROR db3: 1913 lockers Jan 22 08:21:09 DBERROR db3: 1913 lockers Jan 22 08:28:27 DBERROR db3: 1948 lockers Jan 22 08:30:41 DBERROR db3: 1963 lockers Jan 22 08:31:18 DBERROR db3: 1959 lockers Jan 22 08:31:18 DBERROR db3: 1960 lockers Jan 22 08:31:41 DBERROR db3: 1962 lockers Jan 22 08:36:07 DBERROR db3: 1987 lockers Jan 22 08:36:38 DBERROR db3: 1989 lockers Jan 22 08:39:38 DBERROR db3: 2000 lockers Jan 22 08:39:41 DBERROR db3: 2003 lockers Jan 22 08:43:33 DBERROR db3: 2024 lockers Jan 22 08:44:24 DBERROR db3: 2029 lockers Jan 22 08:44:58 DBERROR db3: 2041 lockers Jan 22 08:45:15 DBERROR db3: 2043 lockers Jan 22 08:46:42 DBERROR db3: 2056 lockers Jan 22 08:48:29 DBERROR db3: 2067 lockers Jan 22 08:51:48 DBERROR db3: 2082 lockers Jan 22 08:51:57 DBERROR db3: 2082 lockers Jan 22 08:52:18 DBERROR db3: 2090 lockers Jan 22 08:55:32 DBERROR db3: 2091 lockers Jan 22 08:58:30 DBERROR db3: 2106 lockers Jan 22 08:59:36 DBERROR db3: 2114 lockers Jan 22 09:02:30 DBERROR db3: 2122 lockers Jan 22 09:02:33 DBERROR db3: 2123 lockers Jan 22 09:02:35 DBERROR db3: 2125 lockers Jan 22 09:07:39 DBERROR db3: 2135 lockers Looks like in the past the most lockers I would see was around 8 or 9. ag> But perhaps more disturbing is that saslauthd appears to be freezing ag> up. I haven't been able to detect a pattern yet, but saslauthd ag> seems to stop handling requests. The daemons are still running, but ag> there is a flood of 'cannot connect to saslauthd server'. I then ag> have to stop saslauthd and restart it. I've had to do this ag> something like 4 times now since the upgrade. Seems odd since we're ag> using sasl in a rather simplistic way, and saslauthd worked just ag> fine prior to 2.1.0. Maybe my problem is that I didn't get THAT ag> from CVS? ;-) I'm not entirely sure yet, but so far it looks like nscd might have been the culprit for the file descriptor leak. If I either don't run nscd or tell it not to cache passwd, saslauthd does not appear to run out of descriptors. I don't know if this is just flakiness on the part of nscd, or just some bad interaction between it and ldap_cachemgr. However, it does not appear to be a problem with the saslauthd code. -- Amos
