[
https://issues.apache.org/jira/browse/TS-3439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14361250#comment-14361250
]
Brian Geffon edited comment on TS-3439 at 3/13/15 10:55 PM:
------------------------------------------------------------
[~sudheerv] Yah the case where the server says close was already handled, the
problem in this case was the server omitted the connection close header and
since it was http/1.1 it defaults to keep-alive, so we needed to add the check
that our config allows for connection keep-alive out.
was (Author: briang):
[~sudheerv] Yah the case where the server says close was already handled, the
problem in this case was the server committed the connection close header and
since it was http/1.1 it defaults to keep-alive, so we needed to add the check
that our config allows for connection keep-alive out.
> Chunked responses don't honor keep-alive
> ----------------------------------------
>
> Key: TS-3439
> URL: https://issues.apache.org/jira/browse/TS-3439
> Project: Traffic Server
> Issue Type: Bug
> Reporter: Thomas Jackson
> Assignee: Brian Geffon
> Fix For: 5.3.0
>
>
> If you have ATS configured with keep_alive out disabled, and an origin that
> responds with transfer-encoding chunked ATS puts the connection on the
> keepalive pool after the transfer is finished. Since keep_alive_out is
> disabled the request contains a connection close header. This means that we
> now have a race condition between the origin actually closing the tcp session
> (assuming its well behaved) and ATS re-using that keep-alive session (which
> it shouldn't have kept).
> This means not only are we disobeying the configuration (which specified no
> keep-alive) but we are "breaking" connections-- as they will 502 (since the
> tunnel will be shutdown).
> test case:
> https://github.com/jacksontj/trafficserver/commit/e221e91ad6466ef840f74a1016b8d51c821eb1e9#diff-ed49610150c2617c50f28a047a07c126R130
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
