MySQL Bugs: #12582: using sleep() can make the thread unkillable

Bug #12582	using sleep() can make the thread unkillable
Submitted:	15 Aug 2005 12:46	Modified:	18 Aug 2005 18:47
Reporter:	Michael Widenius	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server	Severity:	S2 (Serious)
Version:	5.0	OS:	Any (all)
Assigned to:	Jim Winstead	CPU Architecture:	Any

Description:
One can make a thread unkillable by executing sleep(#) with a high argument
Another problem is that my_usleep() is not guranteed to sleep the requested number of microseconds as in a threaded environment it's not uncommon that threads wake up from
select() or sleep() much earlier

How to repeat:
- Create two connections to mysql and let one do a sleep(10000) and in the other send a kill to the first connection.
- Test on all os (especially Linux with linuxthreads and freebsd) to verify that during a heavy multi-thread test that sleep() doesn't wake us to early.

Suggested fix:
Change call of my_sleep() to use pthread_cond_timedwait() and make it breakable by 'kill #mysql_connection_id'

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/28345

i'm working on testing this on freebsd and linux with linuxthreads, but it seems that any failures under load would be a problem shared with GET_LOCK().

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/28357

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/28429

Fixed in 5.0.12. (Doesn't need documentation, since SLEEP() hasn't been in a release yet.)