Andy Jezierski writes:
> Are there any instructions in setting up the Bayes DB using a Redis
> server?
Yes, in release notes (currently also in build/announcements/PROPOSED-3.4.0.txt
in svn). Pretty much exactly as you already have it.
> I've installed the server, took the sample config options and added them
> to local.cf
>
> bayes_store_module Mail::SpamAssassin::BayesStore::Redis
> bayes_store_module_additional Mail::SpamAssassin::Util::TinyRedis
> bayes_sql_dsn server=127.0.0.1:6379;password=spamd;database=2
> bayes_token_ttl 21d
> bayes_seen_ttl 8d
> bayes_auto_expire 1
> use_bayes 1
> bayes_auto_learn 1
>
> Performed a redis-cli -n 2 FLUSHDB
>
> Did a backup of one of my mysql bayes databases and am attempting to do a
> restore to the new system.
Good.
> Looks like the redis server keeps chewing up swap space until it runs out,
> then the redis server terminates.
>
> Running on FreeBSD 9.2 perl 5.18-5.18.2 redis server 2.8.4
> Any ideas?
Depends very much on the number of tokens you have in you SQL database.
Mine (cca 1000 users) keeps hovering at about 1 M tokens (and just keeps
few very recent 'seen' entries), resulting in redis server using under
300 MB of memory.
$ redis-cli -n 2 keys 'w:*' | wc -l
1091475
$ redis-cli -n 2 keys 's:*' | wc -l
1324
May be worthwhile to purge old tokens from SQL first, before
creating a backup. Also, it is safe to ditch the entire 'seen'
set of records, it's not worth transfering them to a new database.
If this still gives unreasonable number of tokens, it may be worth
decimating a set - just preserving a random subset of tokens.
Another option is to just start from an empty database. With a reasonable
set of other rules, network tests and autolearning on, the required
200 samples of ham and 200 of spam can be quickly reached on
a busy server. During initial learning consider decreasing score for
BAYES_00 and BAYES_99 rules.
Note that bayes_token_ttl and bayes_seen_ttl have no effect
on entries loaded from a backup dump, they are all given
a 'current' timestamp (with some random offset so that they
will not expire at exactly the same time). But for a steady-state,
with these *_ttl settings you can control how many items are
kept in a database on the average.
Axb writes:
> what does sa-learn --dump magic say (when using mysql)
Good idea to check this first.
> my Redis
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 5728 root 20 0 5355m 5.1g 1020 S 1.3 37.1 711:19.45 redis-server
>
> sa-learn --dump magic
> 0.000 0 3 0 non-token data: bayes db version
> 0.000 0 16481050 0 non-token data: nspam
> 0.000 0 5690858 0 non-token data: nham
> bayes_token_ttl 864000
> bayes_seen_ttl 2d
A biggie!
Btw, with redis db the number of tokens actually in a database
may not be directly related to the number of learned and reported
tokens bacause of the automatic expiration performed by
redis server (according to bayes_token_ttl) - unlike other bayes
back-ends where purging is done explicitly by SpamAssassin.
Mark
