[Rspamd-Users] Redis learned_ids key has huge size.

[Gabriele Nencioni]

On 1/27/20 11:19 AM, Alexander Moisseev wrote:
>> maxmemory_policy:allkeys-lru
> 
> LRU is not good for Bayesian statistics as it leaves a lot of
> meaningless tokens. "used" != "useful".
> And it can be harmful for other data if you store everything in the same
> Redis instance.
> I would suggest to use smarter Bayes expiry module instead of LRU
> policy. Or, better yet, expiry module in conjunction with a separate
> Redis instance and volatile-ttl policy for statistics.


Hi,
thank you for your suggestion.

I have an rspamd version before 2.x, so I guess to configure the lazy
expiration mode:
https://rspamd.com/doc/modules/bayes_expiry.html

am I right?

Is there a way to migrate the actual Bayesian statistics from old schema
to the new one withou loose all of them?

Regards,
-- 
Gabriele Nencioni