On Tue, Jul 26, 2011 at 06:13:54PM -0400, Zardosht Kasheff wrote:
> For the last case, where we have:
> > handlers[0]->index_XXX();
> > handlers[1]->index_XXX();
> > handlers[2]->index_XXX();
> > handlers[0]->index_XXX();
> > handlers[0]->index_XXX();
> > handlers[1]->index_XXX();
>
> I assume each of the handlers share the same thd object? So if I check
> the thd object for each of the handlers, I will get the same one?
Yes. handlers[i]->ha_thd() == handlers[j]->ha_thd() for any i and j.
> I guess here is another question. Is there any way for a handler with
> some thd object be blocked from executing until some operation with
> ANOTHER thd finishes. For instance, in the above example, handlers[1]
> cannot execute until handlers[0] performs its index_XXX function, but
> (I hope) they have the same thd.
Every query is executed by some particular thread, and all ha_xxx objects
used by a thread must have their thd equal to thread's thd. This means that
the blockings you're interested in cannot be caused by a single query.
As for multiple queries, I can only think of locking:
- If your storage engine's ha_xxx::store_lock() produces lock objects that are
exclusive for one another, then the blocking may occur
- I suspect something similar can happen with DDL operations. if there is a
DDL operation running, other threads won't be able to open the table until
the DDL op has finished.
BR
Sergey
--
Sergey Petrunia, Software Developer
Monty Program AB, http://askmonty.org
Blog: http://s.petrunia.net/blog
