[Bug 3080] New: maybe locking problems - not sure

[bu...@bugzilla.spamassassin.org]

http://bugzilla.spamassassin.org/show_bug.cgi?id=3080

           Summary: maybe locking problems - not sure
           Product: Spamassassin
           Version: 2.63
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P5
         Component: spamassassin
        AssignedTo: spamassassin-dev@incubator.apache.org
        ReportedBy: boldi@mail2003.etl.hu


Background:
spamassassin protected mailbox (by procmailrc), debian, compiled by hand. 
upgraded from various versions. No daemon, only user stuff (no central bayes 
stuff)

Situation:
Tried to learn some bayes stuff. Spamassassin already had working bayesian 
databases. then I added a 
sa-learn --spam --dir Maildir/something/cur --show-dots.
It contained some 1000s of messages so it ran for a while.

Inbetween I recognized that some parameters might useful to  make larger:
bayes_expiry_max_db_size 350000
bayes_journal_max_size 1024000
addded to /etc/spamassassin/local.cf. (done while sa-learn was running) (might 
be not important)

Before the process, --dump magic containt about 26000 spam and 10000 ham or 
something.

During the run of sa-learn the mailbox received several mails, maybe spams too.
Finally, after finishing the whole process, sa-learn --dump magic has shown 5 
or 7 spams and no hams...

Trying to run sa-learn resulted in "i already have this message and that 
and ...."  and the size of the database was also o.k. , sa-learn --rebuild did 
not helped the problem.

I finally decided to clear the whole database (initially created by previous 
versions (db ver 2) ), and make a new one.

I successfully added 26000 spams and 3000 hams but when I added another 3000 
hams, it said "3000 hams added (3010 total)" or something, but the number of 
hams did not increased.

---
During sa-learn phases I found that sa-learn initially creates some bayes.lock 
file, but it disappears after a while. (maybe the spamassassin-procmail 
deletes it somehow?). At some point , the procmail spamassassin also creates 
something bayes.computername.lock.

Is it possible that during the lock file query/creation the single bayes.lock 
is not checked but after the run (delete phase) it is unlinked by 
spamassassnin?

On the other hand, is it possible to revitalize the database? it seems it 
contains the right amount of data (like the md's of the messages entered) but 
the magic is wrong. (when nhams/spams "disappeared" spamassassin also said 
that "there is too few messages to run bayes" something, although the file 
size was not decreased)

It was a bit long and complicated, both I hope that helps your work.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.