Client connection becomes stale after long periods of time.

[warroyo]

I have a long running background job and I am making calls to ES at different points throughout the job to update fields in ES. After not calling ES for about 8 minutes and trying to make a call though the client again it will just hang. I do not receive any exceptions or timeout. it just hangs. is there a reason this could be happening? Does the client connection persist the whole time my job is running from the initial instantiation? is there a way to reconnect?

UPDATE:
I have found this to be a problem with my load balancer that ES is behind, it kills tcp connections after 5 minutes. the problem is that my persistent connection does not register the connection being closed so it just hangs. Is there a way to reset the connection? or have the client ping every 60 seconds?

[polyfractal]

Hmm, interesting. If the socket doesn't drop cleanly (e.g. half-open socket), it's possible the curl handles will just hang...although they should timeout eventually. I'm surprised you don't see a timeout at all. I'll try to setup some tests to see why this is the case.

As mentioned in #226, you can tell curl to explicitly close connections after each use by using CURLOPT_FORBID_REUSE. Pinging every 60 seconds would also be a viable option to keep the socket open and avoid the cost of tearing it down after every API call.

[warroyo]

I attempted to set the the following options but it did not timeout. I ended up having my code create a new instance of the Client for every connection( I am not making that many requests so it is not a big deal, only about 10 requests over a 20 minute period). I just added the CURLOPT_FORBID_REUSE, and now the connections will close immediately which is a big help.

I would still be interested int he results form your tests though.

'guzzleOptions' => array(
        'curl.options'=> array(CURLOPT_CONNECTTIMEOUT=> 2.0)
      )

[polyfractal]

Agreed, it's quite odd. I wonder if it is a limitation of libcurl perhaps? Not sure, will need to dig into it next week with proper testing. I'll update this thread when I have more news :)

[polyfractal]

Closing, this is likely no longer relevant since we're not on that version of guzzle anymore :)

[asanovr]

How i can tell curl use CURLOPT_FORBID_REUSE in latest client version ? I can't find it in docs

[polyfractal]

Arbitrary curl options go inside ['client']['curl'] on the request. This is an indexing request example, but it works for any kind of request:

    $params = [
        'index' => 'posts',
        'type' => 'tags',
        'id' => 24,
        'client' => [
            'curl' => [
                CURLOPT_FORBID_REUSE => true // <-- here
            ]
        ]
    ];