You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@mesos.apache.org by Gastón Kleiman <ga...@mesosphere.io> on 2018/09/05 03:11:26 UTC
Re: Review Request 68553: Set 'Connection: close' in the master's
streaming API responses.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68553/#review208340
-----------------------------------------------------------
Ship it!
We should do the same in the executor API handler.
- Gastón Kleiman
On Aug. 28, 2018, 7:03 p.m., Benjamin Mahler wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68553/
> -----------------------------------------------------------
>
> (Updated Aug. 28, 2018, 7:03 p.m.)
>
>
> Review request for mesos, Gastón Kleiman and Vinod Kone.
>
>
> Bugs: MESOS-9189
> https://issues.apache.org/jira/browse/MESOS-9189
>
>
> Repository: mesos
>
>
> Description
> -------
>
> We've seen some HTTP intermediaries (e.g. ELB) decide to re-use
> connections to mesos as an optimization to avoid re-connection
> overhead. As a result, when the end-client of the streaming API
> disconnects from the intermediary, the intermediary leaves the
> connection to Mesos open in an attempt to re-use the connection
> for another request once the response completes. Mesos then thinks
> that the subscriber never disconnected and the intermediary happily
> continues to read the streaming events even though there's no
> end-client.
>
> To help indicate to intermediaries that the connection SHOULD NOT
> be re-used, we can set the 'Connection: close' header for streaming
> API responses. It may not be respected (since the language seems to
> be SHOULD NOT), but some intermediaries may respect it and close the
> connection if the end-client disconnects.
>
> Note that libprocess' http server currently doesn't close the the
> connection based on a handler setting this header, but it doesn't
> matter here since the master's operator / scheduler and agent's
> executor streaming API responses are infinite.
>
>
> Diffs
> -----
>
> src/master/http.cpp e074a93666d82944774e7b0c3fa32d7021d93c97
>
>
> Diff: https://reviews.apache.org/r/68553/diff/1/
>
>
> Testing
> -------
>
> Manually verified headers make it through:
>
> ```
> $ telnet 10.0.49.2 5050
> Trying 10.0.49.2...
> Connected to 10.0.49.2.
> Escape character is '^]'.
> POST /master/api/v1 HTTP/1.1
> Content-Type: application/json
> Accept: application/json
> Content-Length: 20
>
> {"type":"SUBSCRIBE"}
> HTTP/1.1 200 OK
> Transfer-Encoding: chunked
> Content-Type: application/json
> Date: Wed, 29 Aug 2018 01:58:34 GMT
> Connection: close
>
> 9e
> 154
> {"type":"SUBSCRIBED","subscribed":{"get_state":{"get_tasks":{},"get_executors":{},"get_frameworks":{},"get_agents":{}},"heartbeat_interval_seconds":15.0}}
> 17
> 20
> {"type":"HEARTBEAT"}
> ```
>
>
> Thanks,
>
> Benjamin Mahler
>
>
Re: Review Request 68553: Set 'Connection: close' in the master's
streaming API responses.
Posted by Benjamin Mahler <bm...@apache.org>.
> On Sept. 5, 2018, 3:11 a.m., Gastón Kleiman wrote:
> > We should do the same in the executor API handler.
Yeah, originally I did that in this patch, but it ended up being more complicated than I anticipated. In any case, I commented in MESOS-9189 that I would leave the ticket open (although, probably I need to pull out another ticket for the sake of the CHANGELOG and backporting).
- Benjamin
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68553/#review208340
-----------------------------------------------------------
On Aug. 29, 2018, 2:03 a.m., Benjamin Mahler wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68553/
> -----------------------------------------------------------
>
> (Updated Aug. 29, 2018, 2:03 a.m.)
>
>
> Review request for mesos, Gastón Kleiman and Vinod Kone.
>
>
> Bugs: MESOS-9189
> https://issues.apache.org/jira/browse/MESOS-9189
>
>
> Repository: mesos
>
>
> Description
> -------
>
> We've seen some HTTP intermediaries (e.g. ELB) decide to re-use
> connections to mesos as an optimization to avoid re-connection
> overhead. As a result, when the end-client of the streaming API
> disconnects from the intermediary, the intermediary leaves the
> connection to Mesos open in an attempt to re-use the connection
> for another request once the response completes. Mesos then thinks
> that the subscriber never disconnected and the intermediary happily
> continues to read the streaming events even though there's no
> end-client.
>
> To help indicate to intermediaries that the connection SHOULD NOT
> be re-used, we can set the 'Connection: close' header for streaming
> API responses. It may not be respected (since the language seems to
> be SHOULD NOT), but some intermediaries may respect it and close the
> connection if the end-client disconnects.
>
> Note that libprocess' http server currently doesn't close the the
> connection based on a handler setting this header, but it doesn't
> matter here since the master's operator / scheduler and agent's
> executor streaming API responses are infinite.
>
>
> Diffs
> -----
>
> src/master/http.cpp e074a93666d82944774e7b0c3fa32d7021d93c97
>
>
> Diff: https://reviews.apache.org/r/68553/diff/1/
>
>
> Testing
> -------
>
> Manually verified headers make it through:
>
> ```
> $ telnet 10.0.49.2 5050
> Trying 10.0.49.2...
> Connected to 10.0.49.2.
> Escape character is '^]'.
> POST /master/api/v1 HTTP/1.1
> Content-Type: application/json
> Accept: application/json
> Content-Length: 20
>
> {"type":"SUBSCRIBE"}
> HTTP/1.1 200 OK
> Transfer-Encoding: chunked
> Content-Type: application/json
> Date: Wed, 29 Aug 2018 01:58:34 GMT
> Connection: close
>
> 9e
> 154
> {"type":"SUBSCRIBED","subscribed":{"get_state":{"get_tasks":{},"get_executors":{},"get_frameworks":{},"get_agents":{}},"heartbeat_interval_seconds":15.0}}
> 17
> 20
> {"type":"HEARTBEAT"}
> ```
>
>
> Thanks,
>
> Benjamin Mahler
>
>