Mailing-List: contact new-httpd-help@apache.org; run by ezmlm
Precedence: bulk
Reply-To: new-httpd@apache.org
Date: Wed, 08 Aug 2001 13:01:07 -0700
From: Brian Pane <bpane@pacbell.net>
Subject: graceful restart  Re: Currently known issues with 2.0.23
To: new-httpd@apache.org
Message-id: <3B719A83.6080701@pacbell.net>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii; format=flowed
Content-transfer-encoding: 7BIT
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.3) Gecko/20010801
References: <Pine.GSO.4.32.0108081110410.7199-100000@cobra.cs.Virginia.EDU>
 <01080808291306.03684@koj.rkbloom.net>
Status: O

Ryan Bloom wrote:

>>>>2) Unix: The threaded MPM might take longer than expected to restart or
>>>>   shutdown on very-low-traffic (near idle) servers.  This is due to
>>>>the
>>>>
>>>I dislike this.  It's the "near idle" that bothers me.  We rely on the
>>>OS to never starve a thread to get any restart or shutdown.  The
>>>server could easily be getting hit moderately, and not shutdown
>>>correctly.  I don't want to see us down-play the problem, and have
>>>people think that the server is working when it isn't.  There is a
>>>real chance that an admin can do a graceful restart, and still be
>>>serving requests off the old config for a very long time.
>>>
>>How about "servers getting very few requests" instead of
>>"very-low-traffic (near idle) servers"?
>>
>>I threw the "near idle" bit in there to emphasize that (a) it effects
>>people running test servers that have nobody connecting to them at all,
>>and (b) that our definition here of "very low traffic" is not 100 conn/sec
>>as opposed to 1000 conn/sec, but more on the order of just a handful of
>>conn/sec, since 100 conn/sec might be seen as very low traffic by some
>>administrators.  ;-)
>>
>
>a)  is just not true.  This doesn't affect only test servers, it affects anybody who
>isn't seeing a log of hits at a time.  And we haven't got a clue what the threshold
>is.  Again, we rely on the OS to make sure that every single thread gets
>the it's share of the connections.
>
Yep.  And I think the 'server not getting many hits' case should be expected
even on high-traffic sites.  A cautious operations team might reasonably do
something like this when making config changes:
  - Configure the load balancer to stop sending new connections to one 
of n web servers
  - Install a new httpd.conf on that server
  - Graceful restart of apache (to give any open connections a chance to 
complete)
  - Verify that the server handles requests as expected
  - Configure the load balancer to start sending traffic to that server 
again

--Brian