David B Funk wrote:
On Thu, 22 Feb 2007, Daryl C. W. O'Shea wrote:
David B Funk wrote:
On Thu, 22 Feb 2007, Daryl C. W. O'Shea wrote:
Yes. Learning is slow. If two spamd children try to learn at the same
time only one will get a lock to write to the database. The child who
doesn't get the lock (tie fails) will report that autolearn is unavailable.
But doesn't bayes_learn_to_journal help this?
If he enabled it, sure, it'd help reduce the frequency that this occurs.
There's still only a single journal, though, so it's still possible
that a spamd child process won't get a tie on it during busy periods.
Hmm, as the journal file is an append-only write I had assumed that
it didn't need locking, thus theoretically "simultaneous" updates
could be possible.
Yeah... I had (hopefully temporarily) envisioned a pretty retarded
journaling method. Like Theo pointed out, there's no locking the
journal. I plead sleep deprivation.
Daryl
