[Rspamd-Users] constant redis/lua errors in rspamd logs [bug]

[Vsevolod Stakhov]

On 12/12/2024 18:00, Vsevolod Stakhov wrote:
> On 12/12/2024 17:33, Gerald Galster wrote:
>>>>> Do you require expand_keys and/or prefix?
>>>>>
>>>>> If not, try to remove them from config. Stop rspamd and redis, move 
>>>>> the *.rdb file elsewhere.
>>>>> Start redis and rspamd and learn a bunch of spam-/ham mails and see 
>>>>> if the error persists.
>>>>>
>>>>> You could also try redis_cli monitor to see what's going on.
>>>>>
>>>> Hmm, after restarting rspamd everything went back to normal - with 
>>>> everything across the board related to redis - as far as logs go at 
>>>> least.
>>>> It turned out restarting redis caused the issues - it looks like 
>>>> rspamd doesn't not attempt to reconnect and just throws errors in 
>>>> the logs. And such state persists indefinitely until rspamd itself 
>>>> is restarted.
>>>> Is there any way to enforce reconnection ? I see no obvious options 
>>>> anywhere ...
>>>
>>> One more to add - judging from connections rspamd indeed reconnects 
>>> to redis, but it doesn't really change the functional side - parts of 
>>> rspamd are not aware that reconnection happened.
>>>
>>> Maybe this should be reported reported as a bug.
>>
>> This seems like a bug and is reproducible with rspamd 3.10.2 and 
>> valkey on RHEL derivatives.
>>
>> Open a bug report here: https://github.com/rspamd/rspamd/issues
>>
>>
>> # systemctl restart valkey
>>
>> rspamd[2907245]: <>; ; lua_redis_push_error: call to callback failed:
>> /usr/share/rspamd/lualib/lua_redis.lua:1465: bad format string type: 
>> nil; trace:
>>   [1]:{[C]:-1 - infox [C]};
>>   [2]:{/usr/share/rspamd/lualib/lua_redis.lua:1465 - callback [Lua]};
>>   [3]:{/usr/share/rspamd/lualib/lua_redis.lua:1063 - <unknown> [Lua]};
>>
>> rspamd[2907244]: <4Y8Jvk>; lua; lua_redis.lua:1465: redis script from 
>> file: bayes_classify.lua
>>   is not loaded (NOSCRIPT returned), scheduling reload
>>
>> rspamd[2907244]: <4Y8Jvk>; lua; lua_redis.lua:1465: redis script with 
>> id: 24
>>   is not loaded (NOSCRIPT returned), scheduling reload
>>
>>
>> Best regards,
>> Gerald
>>
> 
> It's a different issue: https://github.com/rspamd/rspamd/ 
> commit/887b84d1fa6235d4582d7468a662873e5edefe91
> 
> The original report is slightly different but it's a similar one I suppose.

Yes, your guess was correct: it is related to `expand_keys` and 
https://github.com/rspamd/rspamd/pull/1838, so we might need to get some 
feedback from Andrew Lewis about it, as I don't have enough context there.