On Wed, 11 Jun 2008 12:44:46 +0900 (JST)
yamamoto@valinux.co.jp (YAMAMOTO Takashi) wrote:

> > I'm now considering following logic. How do you think ?
> > 
> > Assume: move TASK from group:CURR to group:DEST.
> > 
> > == move_task(TASK, CURR, DEST)
> > 
> > if (DEST's limit is unlimited)
> > 	moving TASK
> > 	return success.
> > 
> > usage = check_usage_of_task(TASK).
> > 
> > /* try to reserve enough room in destionation */
> > if (try_to_reserve_enough_room(DEST, usage)) {
> > 	move TASK to DEST and move pages AMAP.
> > 	/* usage_of_task(TASK) can be changed while we do this.
> > 	   Then, we move AMAP. */
> > 	return success;
> > }
> > return failure.
> > ==
> 
> AMAP means that you might leave some random charges in CURR?
> 
yes. but we can reduce bad case by some way 
 - reserve more than necessary.
 or
 - read_lock mm->sem while move.

> i think that you can redirect new charges in TASK to DEST
> so that usage_of_task(TASK) will not grow.
> 

Hmm, to do that, we have to handle complicated cgroup's attach ops.

at this moving, memcg is pointed by
 - TASK->cgroup->memcg(CURR)
after move
 - TASK->another_cgroup->memcg(DEST)

This move happens before cgroup is replaced by another_cgroup.


Thanks,
-Kame

_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers