[ 
https://issues.apache.org/jira/browse/TS-4041?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15037919#comment-15037919
 ] 

Susan Hinrichs commented on TS-4041:
------------------------------------

Unfortunately, it looks like squid removed all client/server connection error 
related codes.  If I'm reading things right, the ERR_ codes got mapped to the 
status codes which are http://www.comfsm.fm/computing/squid/FAQ-6.html#ss6.8

It is valuable to have information about how connections failed.  Do we add our 
own as a separate status?  In these cases, the transaction failed before we get 
to the point of a HTTP status code.

> Refine squid codes to distinguish connection shutdown causes
> ------------------------------------------------------------
>
>                 Key: TS-4041
>                 URL: https://issues.apache.org/jira/browse/TS-4041
>             Project: Traffic Server
>          Issue Type: Improvement
>          Components: Logging
>            Reporter: Susan Hinrichs
>             Fix For: 6.1.0
>
>
> SQUID_LOG_ERR_CLIENT_ABORT is used in multiple scenarios.  The ambiguity 
> between these scenarios has caused us to take some false turns in debugging 
> failures.  Based on code inspection, CLIENT_ABORT will show up in the 
> following cases.
> * The client sends an EOS unexpectedy. CLIENT_ABORT seems completely 
> appropriate.
> * The client VC times out (either inactive or active timeout). CLIENT_ABORT 
> doesn't seem right here. Perhaps SQUID_LOG_ERR_READ_TIMEOUT would be better. 
> That seems to be currently used for server side timeouts. Might be nice to 
> have a differentiation on which side timed out.
> * The client VC has a read failure. This would include the SSL_read failure 
> case. Could use SQUID_LOG_ERR_READ_ERROR. Currently only used if the server 
> side sends EOS before end of expected data. Again differentiating server side 
> and client side error would be beneficial.
> How fixed are the squid error codes?  Could we add more codes?  If so, we 
> could distinguish between the client-side failures and server-side failures 
> by adding the following codes.
>     * Use CLIENT_ABORT only for for client sending EOS too early.
>     * Use CLIENT_READ_TIMEOUT for inactive/active timeouts on client side.
>     * Use CLIENT_READ_ERROR for read failures on client side.



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

Reply via email to