Re: bzr commit into mysql-5.1-bugteam branch (Dao-Gang.Qu:3192) Bug#42851

[Davi Arnaut]

On 11/24/09 10:54 AM, Andrei Elkin wrote:
> Davi, hello.
>
>     >>  1. Rating for the error log should be done per session
>     >>  2. Rating should target at duplicate messages
>
> Just to be sure, notice, that I mean 1&  2. I.e rating duplicate messages
> generated by a session. And that covers the immediate issue of the bug.
>
>     >
>     >  IMHO, we should do either one, not the two. 1. should be enough.
>
>     Actually, 1 has problems too. For example, in cases where there is a
>     exceptional condition and a storage engine wants to dump a lot of
>     information to the error log in a short period of time. This might
>     inadvertently trigger rate limiting.
>
> I think we would rather to preserve all the volume in this case.

Yeah, but how are we going to achieve it?

>
>     2 problems of its own too if we
>     want to detect duplicate messages within a time interval.
>
> I am not sure what you mean here. I have been thinking on rating always
> implying the time interval context.

I mean that it would increase the complexity. We would need some kind of 
way to track duplicate messages within a interval and this is complicate 
to implement (we don't want a huge memory leak.. we would need periodic 
garbage collection too.. i guess).

>     This is proving to be quite controversial for something that should be
>     kept simple. Perhaps we should stick with what syslog does: eliminate
>     subsequent duplicated messages. We track those by using a checksum on
>     the message format. What do you guys thing?
>
> "Eliminate subsequent duplicated messages" *of a session*. We should not discard the
> 2nd session's
> possible only message.

Fine by me.

> About checksum as the hash key - now I see it is better than my earlier numberic id
> suggestion simply
> because of two errors with the common num code can be unrelated.
>

Yep.

Regards,

--
Davi Arnaut