Trying to Understand Upstream Keepalive

[Maxim Dounin]

Hello!

On Thu, May 08, 2014 at 03:12:44AM -0400, abstein2 wrote:

> I'm trying to better wrap my head around the keepalive functionality in the
> upstream module as when enabling keepalive, I'm seeing little to no
> performance benefits using the FOSS version of nginx.
> 
> My upstream block is:
> 
> upstream upstream_test_1 { server 1.1.1.1 max_fails=0; keepalive 50; }
> 
> With a proxy block of:
> 
> proxy_set_header X-Forwarded-For $IP;
> proxy_set_header Host $http_host;
> proxy_http_version 1.1;
> proxy_set_header Connection "";
> proxy_pass http://upstream_test_1;
> 
> 1) How can I tell whether there are any connections currently in the
> keepalive pool for the upstream block? My origin server has keepalive
> enabled and I see that there are some connections in a keepalive state,
> however not the 50 defined and all seem to close much quicker than the
> keepalive timeout for the backend server. (I am using the Apache server
> status module to view this which is likely part of the problem)

As long as load is even enough, don't expect to see many keepalive 
connections on the backend - new connections will be only open if 
there are no idle connections in the cache of a worker process.

> 2) Are upstream blocks shared across workers? So in this situation, would
> all 4 workers I have shared the same upstream keepalive pool or would each
> worker have it's own block of 50?

It's per worker, see http://nginx.org/r/keepalive.

> 3) How is the length of the keepalive determined? The origin server's
> keepalive settings? Do the origin server's keepalive settings factor in at
> all?

Connections are kept in the cache till the origin server closes them.

> 4) If no traffic comes across this upstream for an extended period of time,
> will the connections be closed automatically or will they stay open
> infinitely?

See above.

> 5) Are the connections in keepalive shared across visitors to the proxy? For
> example, if I have three visitors to the proxy one after the other, would
> the expectation be that they use the same connection via keepalive or would
> a new connection be opened for each of them?

Connections in the cache are shared for all uses of the upstream.  
As long as a connection is idle (and hence in the cache), it can 
by used for any request by any visitor.

> 6) Is there any common level of performance benefit I should be seeing from
> enabling keepalive compared to just performing a proxy_pass directly to the
> origin server with no upstream block?

No.

There are two basic cases when keeping connections alive is 
really beneficial:

- Fast backends, which produce responses is a very short time, 
  comparable to a TCP handshake.

- Distant backends, when a TCP handshake takes a long time, 
  comparable to a backend response time.

There are also some bonus side effects (reducing number of sockets 
in TIME-WAIT state, less work for OS to establish new connections, 
less packets on a network), but these are unlikely to result in 
measurable performance benefits in a typical setup.

-- 
Maxim Dounin
http://nginx.org/