On Wed, 2009-06-10 at 18:09 +0800, He Zhenxing wrote:
> I think max_error_count only limits the errors stored in memory, it
> does
> not limit the errors write into logs. And I think it is good to have a
> limit if the information is stored in memory because usually memory
> resources are very limited and vital for applications, OTOH, there is
> no
> such need to limit the number of errors/warnings written to log file.
Hi Andrei, Zhenxing,
I have reconsidered and I think I agree with Zhenxing. I think a
boolean should be fine. Why? Here are the reasons:
1. If the user is using this server option, then he expects to
suppress the warnings, most likely, completely. Note that the
original motivation for introducing this additional changeset to
the bug fix was that these warnings were printed frequently in
some workloads. As such, this flag will only matter for these
cases, and the user will be aware that he his suppressing *all*
warnings (I suspect that the user would be setting the limit always
to 0 in these cases). All other cases will be fine to have some
warnings, perhaps, even unnoticed wrt to the error log size (I
suspect that, in this case, users would not even bother setting the
suppress flag, let alone a limit for the number of warnings);
2. As Andrei said, support has also agreed with boolean;
3. The patch with boolean, should turn on/off 5.0 behavior wrt to this
warning, which is basically what users in have requested in the bug
report (Yes, it could still be done with "int", but their request
was more like: Please, stop printing the warnings in the error
log, they're hurting me and my sysadmins/DBAs/storage!;
4. Not really a reason :), but the patch for boolean already
exists (eheh).
If we agree, then please review/approve the patch. If not, we can
escalate this to Lars so he can have the final word on this.
Regards,
Luís
