[ 
https://issues.apache.org/jira/browse/TS-2559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13894875#comment-13894875
 ] 

ASF subversion and git services commented on TS-2559:
-----------------------------------------------------

Commit e6b9cb533bba28920ad9098b261a6e11db2d4553 in branch refs/heads/master 
from [[email protected]]
[ https://git-wip-us.apache.org/repos/asf?p=trafficserver.git;h=e6b9cb5 ]

TS-2559: Disconnect clients on unrecoverable origin errors

If we receive an unrecoverable error from the origin, then the HTTP
tunnel shuts down the cache write, but leaves clients connected.
This means that when a transaction is aborted due to an error, or
by exceeding the configured active transaction time, the client is
left hanging for a response that will never come.

The solution is to treat errors the same as end-of-stream. In all
cases, if we have the response body we are ok, but otherwise we
have to shut down all the tunnel consumers. They can't recover from
the response body being truncated so there's no real choice.

This can be tested by setting
proxy.config.http.transaction_active_timeout_out to a short value
and taking too long to send the response body. Alternatively, close
the origin socket before sending all the body bytes, pausing to
make sure we procress to the tunnel handler state before closing.


> active timeouts do not disconnect client sessions
> -------------------------------------------------
>
>                 Key: TS-2559
>                 URL: https://issues.apache.org/jira/browse/TS-2559
>             Project: Traffic Server
>          Issue Type: Bug
>          Components: Core
>            Reporter: James Peach
>            Assignee: James Peach
>             Fix For: 5.0.0
>
>
> If you set {{proxy.config.http.transaction_active_timeout_out}}, and a 
> transaction aborts with a timeout when {{traffic_server}} is receiving the 
> response body from the origin, the cache write is shut down but the client is 
> left connected. There's no way for the client to handle this error except by 
> timing our, so the right behavior is to shut down all the consumers.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to