[
https://issues.apache.org/jira/browse/TS-658?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14266278#comment-14266278
]
Alan M. Carroll commented on TS-658:
------------------------------------
My current approach to this is to make the write lock have more temporal
locality. To make partial object caching work I have to allow multiple
simultaneous writers to an alternate of a cached object. After that is done,
fixing this is simple presuming it doesn't just disappear as a problem.
> HTTP SM holds the cache write lock for too long
> -----------------------------------------------
>
> Key: TS-658
> URL: https://issues.apache.org/jira/browse/TS-658
> Project: Traffic Server
> Issue Type: Bug
> Components: HTTP
> Reporter: Leif Hedstrom
> Assignee: Alan M. Carroll
> Labels: A
> Fix For: 5.3.0
>
>
> It seems we open the cache for write very early on in the HTTP SM, which can
> have very bad effect on performance if the object is not cacheable. It's not
> totally clear as to why this is done this way, but we should examine this for
> v3.1, and try to minimize how long we hold the lock. It's possible this is
> related to read_while_writer, but then it should be modified IMO.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
