Hello,

I experience some troubles about bayes tokens expiration.

I use a per-user MYSQL database to store bayes tokens.
I set `bayes_auto_expire 0` and `bayes_expiry_max_db_size 150000`.
I run a cron to force bayes expiry for each users like this:
----8<----
sa-learn --username=example@example.com --force-expire
---->8----

The problem is that bayes tokens doesn't seem to expire and the
`bayes_token` table recently exceeds 80 million records. For example one
of the user seem to have 288278 tokens:
----8<----
$ sa-learn --username=user1@example.com --dump magic
0.000          0          3          0  non-token data: bayes db version
0.000          0        261          0  non-token data: nspam
0.000          0       6339          0  non-token data: nham
0.000          0     288278          0  non-token data: ntokens
0.000          0 1277480932          0  non-token data: oldest atime
0.000          0 1381762106          0  non-token data: newest atime
0.000          0          0          0  non-token data: last journal
sync atime
0.000          0 1381754818          0  non-token data: last expiry
atime0.000          0          0          0  non-token data: last expire
atime delta
0.000          0          0          0  non-token data: last expire
reduction count
---->8----

So I try to manually run an expire for this user and I got this debug logs:
----8<----
$ sa-learn --username=user1@example.com --force-expire -D
dbg: bayes: bayes journal sync starting
dbg: bayes: bayes journal sync completed
dbg: plugin: Mail::SpamAssassin::Plugin::Bayes=HASH(0x22da088)
implements 'learner_expire_old_training', priority 0
bayes: expiry starting
dbg: bayes: expiry check keep size, 0.75 * max: 112500
dbg: bayes: token count: 288103, final goal reduction size: 175603
dbg: bayes: first pass? current: 1381754809, Last: 1381716636, atime: 0,
count: 0, newdelta: 0, ratio: 0, period: 43200
dbg: bayes: can't use estimation method for expiry, unexpected result,
calculating optimal atime delta (first pass)
dbg: bayes: expiry max exponent: 9
Odbg: bayes: atime token reduction
dbg: bayes: ======== ===============
dbg: bayes: 43200 287217
dbg: bayes: 86400 287217
dbg: bayes: 172800 287189
dbg: bayes: 345600 285030
dbg: bayes: 691200 281466
dbg: bayes: 1382400 276177
dbg: bayes: 2764800 266007
dbg: bayes: 5529600 256971
dbg: bayes: 11059200 237786
dbg: bayes: 22118400 201142
dbg: bayes: couldn't find a good delta atime, need more token
difference, skipping expire
---->8----
It seems that sa-learn try to reduce the tokens numbers but finally skip
the expire process because it "couldn't find a good delta atime".

I try to reduce or increase the 'bayes_expiry_max_db_size' but it
doesn't fix the issue and the tokens numbers continue to grow out of the
150000 limit.

Is it an issue or maybe I misunderstood something?
Should I mannualy purge old tokens using something like `DELETE FROM
bayes_token WHERE bayes_token.atime <= ...`?

Thanks in advance for your help.

Best regards,

Thomas Pierson