MySQL Bugs: #12588: deadlock detector

Bug #12588	deadlock detector - out of stack
Submitted:	15 Aug 2005 17:50	Modified:	13 Sep 2005 15:50
Reporter:	Vadim Tkachenko	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: InnoDB storage engine	Severity:	S3 (Non-critical)
Version:		OS:
Assigned to:	Osku Salerma	CPU Architecture:	Any

Description:
I tested mysql-5.0 on several high-end systems (2 x Opteron 2.6GHz, 2x
XEON 3.6GHz EMT64, 2 x dual-core Opteron 2.6GHz).

OS: RedHat AS 4 (Update 1).

Mysqld repeatable crashed under certain workload with 1024 threads.
Mysqld did not store any info in error log, but kernel did:

kernel: mysqld[15717]: segfault at 00000000455fbff8 rip 0000000000723a28
rsp 00000000455fc000 error 6

gdb backtrace see in attached file.

From this backtrace I made the conclusion that mysql exceeds stack size (I had thread_stack = 128K, and manual says 64K is enough if I don't use UDP).

So I increased thread_stack to 256K and the problem was resolved.

How to repeat:
see Description

gdb backtrace

Attachment: bt.txt (text/plain), 10.01 KiB.

Assigning this to Osku.

Since the stack consumption is about 100 bytes per level, and InnoDB should normally run with a 32 kB stack, it is good to restrict the depth to 200. Look how the function handles the case *cost > 1000000.

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/internals/28375

Fixed in 5.0.12.

Noted in 5.0.12 changelog.

Note that due to stupidity on my part, this change was actually pushed to 5.0.13, not 5.0.12.

Osku or Paul,

please correct the docs. The bug fix is in 5.0.13, not 5.0.12.

Heikki

I've moved the changelog entry to 5.0.13.