Re: Re[2]: Bug ? - Sasha Pachev - com.mysql.lists.bugs

6 messages in com.mysql.lists.bugsRe: Re[2]: Bug ?

From	Sent On	Attachments
Peter Zaitsev	10 Jul 2001 00:12
Sasha Pachev	10 Jul 2001 12:40
Peter Zaitsev	11 Jul 2001 04:14
Sasha Pachev	11 Jul 2001 07:23
Peter Zaitsev	12 Jul 2001 00:20
Sasha Pachev	12 Jul 2001 03:44

Subject:	Re: Re[2]: Bug ?
From:	Sasha Pachev (sas...@mysql.com)
Date:	07/11/2001 07:23:56 AM
List:	com.mysql.lists.bugs

On Wednesday 11 July 2001 05:14, Peter Zaitsev wrote:

Hello Sasha,

Tuesday, July 10, 2001, 11:40:24 PM, you wrote:

No. I used my own complied version, whith the all parameters I have been used for a more then a year. This is just becase I started to play with RAID/BDB/INNODB before mysql-max was released.

So I do not really think this is a binary bug, but I'll try to use you binary if I'll still have the errors.

Then I have a pretty good explanation of what happened. Your STACK_SIZE is probably 2 MB, unless you have patched glibc. This means that each new thread will be allocating the stack 2 MB higher than the previous thread. If you have a large key buffer + some extra memory allocations, such as big heap tables, or some InnoDB internal buffers, at a certain, very reasonable ( 200-300) number of connections the thread stacks start running into the heap. What is going to happen, contrary to what you would expect ( error from mmap() in pthread_create()), is that the page is simply taken from the heap and is zeroed out. With our binary, things are a little better - STACK_SIZE is only 256 K.