[users@httpd] Apache breaks connections after requests

[Yuri <yu...@rawbw.com>]

I have a question about the situation that happens with Apache with 
Wikimedia app.
In short, Apache begins to break connections when there are many 
concurrent requests.
(Here is the bug report that I created for wikipedia 
https://phabricator.wikimedia.org/T85191).
Connections are broken after http request is complete, but 'Connection: 
keep-alive' has been set by the server in the last response.

I view this a violation of RFC2616 because server breaks connection 
without letting the client know.

My questions are:
What could be misconfigured in Apache? Or maybe this is considered 
normal that it breaks connections?
Is there an option that Apache slows down clients without disconnecting 
in case of congestion?

Yuri

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org

Re: [users@httpd] Apache breaks connections after requests

[Yuri <yu...@rawbw.com>]

On 12/29/2014 17:19, Eric Covener wrote:
> >I have a question about the situation that happens with Apache with
> >Wikimedia app.
> >In short, Apache begins to break connections when there are many concurrent
> >requests.
> >(Here is the bug report that I created for wikipedia
> >https://phabricator.wikimedia.org/T85191).
> >Connections are broken after http request is complete, but 'Connection:
> >keep-alive' has been set by the server in the last response.
> >
> The event MPM does this as a specific process runs out of threads, in
> the hope that there are other processes in better shape.  If it
> didn't, and the keepalive connections became ready, they would queue.

I run only 30 requests concurrently, and every ~15-25 sec an event 
occurs when many connections are dropped. 30 must be a drop in a bucket 
for the powerful wikipedia server, which probably serves thousands of 
requests concurrently.

So it is not clear why 30 concurrent users can cause it run out of 
threads, and why connections are broken in bulk, many of them at the 
same time. And why congestion is triggered later, after 15 seconds of 
successful work (assuming the problem is congestion that triggers some 
logic in apache server)?

Since 30 must be very little compared to how many requests it serves in 
general, does this mean that it stops every 15 seconds and just sheds 
connections randomly?

Yuri

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org

Re: [users@httpd] Apache breaks connections after requests

[Eric Covener <co...@gmail.com>]

On Mon, Dec 29, 2014 at 7:16 PM, Yuri <yu...@rawbw.com> wrote:
> I have a question about the situation that happens with Apache with
> Wikimedia app.
> In short, Apache begins to break connections when there are many concurrent
> requests.
> (Here is the bug report that I created for wikipedia
> https://phabricator.wikimedia.org/T85191).
> Connections are broken after http request is complete, but 'Connection:
> keep-alive' has been set by the server in the last response.
>

The event MPM does this as a specific process runs out of threads, in
the hope that there are other processes in better shape.  If it
didn't, and the keepalive connections became ready, they would queue.

> I view this a violation of RFC2616 because server breaks connection without
> letting the client know.

A client, server, or proxy MAY close the transport connection at any
   time. For example, a client might have started to send a new request
   at the same time that the server has decided to close the "idle"
   connection. From the server's point of view, the connection is being
   closed while it was idle, but from the client's point of view, a
   request is in progress.

   This means that clients, servers, and proxies MUST be able to recover
   from asynchronous close events.

> My questions are:
> What could be misconfigured in Apache? Or maybe this is considered normal
> that it breaks connections?

Seems normal, albeit relatively new in event in 2.4?

> Is there an option that Apache slows down clients without disconnecting in
> case of congestion?

http://httpd.apache.org/docs/2.4/mod/event.html#asyncrequestworkerfactor

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org