Hi Shrinand,
In HTTP1.0, client is required to add an additional header to the request
so that
the server kept the connection reused by multiple HTTP requests.
Connection: Keep-Alive
In HTTP1.1, all connections are considered persistent unless declared
otherwise.
The keepalive message is not used separately. I believe Swift is using the
HTTP 1.1
protocol. However, the HTTP server has default time out settings.
In Swift proxy server, the default connection timeout is controlled in
config file as
below. You can try to look deeper inside of code how it is used.
# node_timeout = 10
# client_timeout = 60
# conn_timeout = 0.5
# post_quorum_timeout = 0.5
I'm also developing a trace tool for Swift here. You can try it and see if
it can provide
you more insight of this question. After you enable it, you can record the
info
into the logs that interests you such as remote client ip and port in each
request.
(currently this info is not recorded by default. you need to add it in the
code. Or you
can tell me what you need so that I update this patch.) Aggregate those
logs and see
how many ports on each ip/node are used at specific time slots.
This is what I know. Hope this information is useful for you.
-Edward Zhang
Shrinand
Javadekar
<shrinand@maginat To
ics.com> "openstack@lists.openstack.org"
<openstack@lists.openstack.org>
2014-07-10 上午 cc
01:37
Subject
Re: [Openstack] [Swift] Running out
of ports or fds?
Any ideas folks?
On Tue, Jul 8, 2014 at 4:26 PM, Shrinand Javadekar
<shrin...@maginatics.com> wrote:
> Hi,
>
> I have a question about the http connections made between the various
> swift server processes. Particularly between the swift proxy server
> and the swift object server.
>
> I see that these servers do not use a persistent http connection
> between them. So every blob get/put/delete request will create a new
> connection, use it and tear it down. In a highly concurrent
> environment with thousands of such operations happening per second,
> there could be two problems:
>
> i) Time required for creating the new connections could hamper
performance.
> ii) After the requests are complete, several connections will be in
> the TIME_WAIT state and it might be possible that the proxy server and
> object server node will run out of ports or fds.
>
> If the proxy and object servers are on the same machine, the problem
> is exacerbated. I have one such instance and at one point saw ~30K
> sockets in the TIME_WAIT state. Though, this would've included
> connections with the account server and container server also.
>
> Does this analysis make sense? If yes, are there ways to do something
> about it (other than asking clients to slow down :P)?
>
> Thanks in advance.
> -Shri
_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to : openstack@lists.openstack.org
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to : openstack@lists.openstack.org
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
