[jira] [Commented] (TS-4341) When ATS hits `proxy.config.http.origin_max_connections` it has no limits on the number of requests that can be queued

["ASF GitHub Bot (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TS-4341?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15236640#comment-15236640 ] 

ASF GitHub Bot commented on TS-4341:
------------------------------------

Github user jacksontj commented on the pull request:

    https://github.com/apache/trafficserver/pull/564#issuecomment-208718335
  
    This isn't complete yet, still need to add handling for timeouts etc., but I wanted to get the PR opened to get some eyes on it :)


> When ATS hits `proxy.config.http.origin_max_connections` it has no limits on the number of requests that can be queued
> ----------------------------------------------------------------------------------------------------------------------
>
>                 Key: TS-4341
>                 URL: https://issues.apache.org/jira/browse/TS-4341
>             Project: Traffic Server
>          Issue Type: Bug
>            Reporter: Thomas Jackson
>            Assignee: Thomas Jackson
>
> If you enable `proxy.config.http.origin_max_connections` to a specific origin, and that origin hits the max number of connections-- ATS will just reschedule the connection indefinitely. This is problematic because if the origin is down you can consume all available connections within ATS (UA side) waiting on the origin to finish up some requests.
> I expect the requests past `proxy.config.http.origin_max_connections` to error out (5xx response code), after talking to [~jpeach@apache.org] that might be a bit too agressive-- as this is used by some to limit the number of connections to the origin (not just limit connection usage in ATS).
> My proposed fix is to take a page out of haproxy's book-- and add a `proxy.config.http.origin_max_connections_queue` -- where we will allow a certain number of connections to queue up waiting on the origin, and if we exceed that number we 5xx the request.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)