On Mon, Feb 20, 2017 at 9:54 PM, Ruediger Pluem <rpl...@apache.org> wrote:
>
>
> On 02/20/2017 02:38 PM, yla...@apache.org wrote:
>> Author: ylavic
>> Date: Mon Feb 20 13:38:03 2017
>> New Revision: 1783755
>>
>> URL: http://svn.apache.org/viewvc?rev=1783755&view=rev
>> Log:
>> mpm_event: use a mutex for ptrans' allocator to be safe with concurrent
>> creation and destruction of its subpools, like with mod_http2.
>
> Hm, doesn't that impact performance too much in the HTTP/1.1 case? It
> requires the mutex to be hold not only during
> creation / destruction of subpools, but also for all apr_palloc calls that
> require the allocator to provide memory to
> the pool. I guess there will be no contention in these cases as only one
> thread will be using the pool, but still the
> lock needs to be made.
Yes, quite possibly, I didn't try to measure it actually.
The hard thing is that at the time ptrans is created, we don't know
how it will be used.
Maybe we could differ apr_allocator_mutex_set() at a later time, e.g.
{pre,process}_connection hook?
Regards,
Yann.
