[
https://issues.apache.org/jira/browse/TS-4601?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15355387#comment-15355387
]
Thomas Jackson commented on TS-4601:
------------------------------------
+1
Definitely don't want to bother retrying since we are explicitly throttling,
and it is probably nice to have the separate counter-- also I assume this
changes the error string you'd see in the logs from ERR_CONNECT_FAIL to
something like ERR_THROTTLE if so-- that would be ideal so users could
differentiate connection failures from throttles (as you'd presumably want to
alert on those separately).
> Connection error from origin_max_connection with origin_max_connections_queue
> set to 0 should not retry
> -------------------------------------------------------------------------------------------------------
>
> Key: TS-4601
> URL: https://issues.apache.org/jira/browse/TS-4601
> Project: Traffic Server
> Issue Type: Bug
> Components: Core
> Reporter: Susan Hinrichs
> Assignee: Susan Hinrichs
>
> While adding a metric to track the number of times a connection to origin is
> dropped due to being over the origin_max_connections limit, I noticed that
> the count was increments three times as fast as I expected from my
> experiment. My connection retry count was 3. I changed the logic to set the
> current attempts to max to avoid the retries in that case.
> [~jacksontj] avoiding the retries makes sense in this case, right?
> I also propose adding a proxy.process.http.origin_connections_throttled
> metric to track how many connections to origin are being error'ed out the
> origin_max_connection limit. The metric is only incremented when the queue
> is 0 or we are over the delay queued limit.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
