SA Bugzilla – Bug 3726
sa leaves 5MB expire files in /var/spool/spamassassin/nobody/
Last modified: 2005-05-10 12:03:25 UTC
Forwarded from Debian BTS: It's probable that I've mucked up something in the configuration, but... Having just ran out of disk space I was astonished to find that /var/spool/spamassassin/nobody/ was taking up 42GB (sic). This consisted of thousands of bayes_toks.expire files, one for every single email received, for all users, going back months: diz $ ls -l total 47576 -rw-rw-rw- 1 root root 86016 2003-10-25 23:16 auto-whitelist -rw------- 2 root root 12288 2003-10-26 01:03 auto-whitelist.dir -rw------- 2 root root 12288 2003-10-26 01:03 auto-whitelist.pag -rw-rw-rw- 1 calum calum 4155 2004-08-23 17:26 bayes_journal -rw-rw-rw- 1 nobody nogroup 2584576 2004-08-23 17:26 bayes_seen -rw-rw-rw- 1 root root 10244096 2004-08-23 17:26 bayes_toks -rw-rw-rw- 1 brian brian 5255168 2004-08-23 15:58 bayes_toks.expire13091 -rw-rw-rw- 1 calum calum 5255168 2004-08-23 16:08 bayes_toks.expire13925 -rw-rw-rw- 1 calum calum 5255168 2004-08-23 16:14 bayes_toks.expire14454 -rw-rw-rw- 1 calum calum 5255168 2004-08-23 16:29 bayes_toks.expire14922 -rw-rw-rw- 1 calum calum 5255168 2004-08-23 16:36 bayes_toks.expire15160 etc, etc. I don't have spamd enabled, and I have these options: OPTIONS="-a -m 10 --helper-home-dir /var/spool/spamassassin/nobody" and: > bayes_path /var/spool/spamassassin/nobody/bayes > bayes_file_mode 0777 in local.cf. have I mucked up here?
How is SA being called? spamc or something like MailScanner?
move bug to Future milestone (previously set to Future -- I hope)
btw, this got figured out: > I didn't get any replies, but for the archives: I fixed this by removing > the sticky bit (mode 01000) from the /var/spool/spamassassin/nobody > directory. >> any idea why that had that effect? > yup, SA, running as the user who was getting mail, was unable to move > the existing journal file out of the way, since it didn't own it. The > sticky bit means that you can only unlink (e.g. move) files that you > own, despite having write perm on the directory. we can't do anything about that -- it's a setup issue.