Re: bzr commit into mysql-trunk branch (kristofer.pettersson:3197) Bug#29071

[Kristofer Pettersson]

Sergey Vojtovich skrev 2010-10-01 21:57:
> Kristofer,
> 
> attached please find mtr test case. It is supposed to give empty
> resultset.
> 
> With non-patched server I get empty resultset.
> With patched server I observe three problems:
> - myisampack cannot join 64 tables;
> - mysqltest cannot cat_file when there are 64 connections open;
> - libmysqlclient cannot "LOAD DATA LOCAL INFILE" when
>   there are 64 connections open.
> 
It seems all of these issues are resolved when MY_NFILE is used as a
parameter in a preceding my_set_max_open_files(MY_NFILE)

The patch should be fixed to support MY_NFILE as the default number of
max open files as you've already pointed out in earlier emails.

> Unfortunately it is yet another good proof that this patch is risky.
I don't see any such proof. You also don't seem to account for the
bigger risk of not doing anything. In fact I think we're doing just
fine: I present a way for the future, and you review it with facts of
the present. The result can only be the bridge which connects the future
and present.

> Please let me know if that's not enough - I can analyze more applications.
> 
I leave it to you to determine when you think the patch has reached high
quality. You're doing a great job!

>>>
>>> My statement was: "my_file_limit cannot be lower than MY_NFILE".
>>>
This should be fixed by asserting that the default value for max open
files is correct without a mandatory call being made to
my_set_max_open_files().