[Rspamd-Users] redis OOM

[Alexander Moisseev]

On 28.06.2019 10:42, Wiethoff, Helge wrote:
> used_memory:2577962576
> used_memory_human:2.40G

It is below maxmemory limit. Do you have enough free physical memory?

> maxmemory:3221225472
> maxmemory_human:3.00G

> 
> The rspamd log "looks kind of normal" beside this repeating message:
> 2019-06-28 09:33:40 #3067(rspamd_proxy) <9cb4a3>; lua; neural.lua:487: cannot learn ANN tRFANNCA81FF5E85529498260: too many ham samples: 320
It is normal behavior.

> 
> Beside that: I was trying to limit the Bayes entries but local.d/statistic.conf conflicts with local.d/classifier-bayes.conf:
> new_schema = true;
> expire = 8640000;
> 
I don't see anything wrong. How exactly it conflicts?

> Maybe it's a good idea to create a redis instance for bayes entries?
> 
Generally, yes. Especially when you want to apply a maxmemory policy. For instance, you don't want to evict greylisting keys just because their TTLs are lower than statistic keys TTLs.
But anyway you need to figure out why are you getting OOM when memory usage is 600M below maxmemory threshold.