On Mon, 26 Jan 2004, Rick Mallett wrote: > [...] > Now I've got a much bigger problem. The expiry is starting to take > more than 10 minutes and as a result the journal grows to max size and > an opportunistic rebuild kills the lock file and wrecks the expiry > operation. Here is what I observe in the bayes directory area > > -rw------- 1 defang defang 32 Jan 26 12:19 bayes.lock > -rw------- 1 defang defang 2750039 Jan 26 12:27 bayes_journal > -rw------- 1 defang defang 20897792 Jan 26 12:19 bayes_seen > -rw------- 1 defang defang 21733376 Jan 26 12:22 bayes_toks > -rw------- 1 defang defang 9437184 Jan 26 12:22 bayes_toks.expire16781 > -rw------- 1 defang defang 11173888 Jan 26 11:35 bayes_toks.expire23012 > -rw------- 1 defang defang 5341184 Jan 26 10:54 bayes_toks.expire27549 > -rw------- 1 defang defang 11182080 Jan 26 11:59 bayes_toks.expire27570 > -rw------- 1 defang defang 11403264 Jan 26 10:44 bayes_toks.expire4752 > > [...] > and on and on. I was running an expiry every hour, BTW, because I was > worried about the possibility that an expiry might take more than 10 > minutes and get killed by a journal'izing operation but I decided to > return to the default mode of operation and let the expiries happen > automatically. In this case "sa-learn --dump magic" indicates that > the last expiry happened at "Sun Jan 25 22:33:43 2004" and it looks > like my problem happened about 12 hours later so I guess I should > return to running an expiry every hour. Anyone care to speculate?
That looks remarkably like the problem I reported on Jan 20th, under the Subject 'Multiple "bayes_toks.expire$$" files'. Our context is that our SA is driven as subroutines from "MailScanner" (www.mailscanner.info). We later discovered that these abandoned "bayes_toks.expire$$" files correspond to MailScanner (which uses SA's auto-expire) timing out (around 40 seconds) and so somehow these SA files get abandoned. The MailScanner maintainer, Julian Field, is very responsive, and he has already coded up an alternative way of driving SA from MS, so that its use of SA can avoid auto-expire (and thereby avoid the possible multiple simultaneous expires) and instead do a more controlled, occasional and explicit expire. But it does suggest an underlying problem in SA's expiry code. I suspect that the process is being signalled, and probably dies, whereas it should probably try to tidy up the half-complete expire (e.g. some sort of fast abandonment, possibly as simple as unlinking the "bayes_toks.expire$$" file) before dying. Thoughts? -- : David Lee I.T. Service : : Systems Programmer Computer Centre : : University of Durham : : http://www.dur.ac.uk/t.d.lee/ South Road : : Durham : : Phone: +44 191 334 2752 U.K. : ------------------------------------------------------- The SF.Net email is sponsored by EclipseCon 2004 Premiere Conference on Open Tools Development and Integration See the breadth of Eclipse activity. February 3-5 in Anaheim, CA. http://www.eclipsecon.org/osdn _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
