[
https://issues.apache.org/jira/browse/TS-4723?focusedWorklogId=33757&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-33757
]
ASF GitHub Bot logged work on TS-4723:
--------------------------------------
Author: ASF GitHub Bot
Created on: 16/Dec/16 17:22
Start Date: 16/Dec/16 17:22
Worklog Time Spent: 10m
Work Description: Github user bryancall closed the pull request at:
https://github.com/apache/trafficserver/pull/843
Issue Time Tracking
-------------------
Worklog Id: (was: 33757)
Time Spent: 8h 20m (was: 8h 10m)
> ATS CARP Plugin
> ---------------
>
> Key: TS-4723
> URL: https://issues.apache.org/jira/browse/TS-4723
> Project: Traffic Server
> Issue Type: New Feature
> Components: Plugins
> Reporter: Eric Schwartz
> Assignee: Eric Schwartz
> Fix For: 7.1.0
>
> Time Spent: 8h 20m
> Remaining Estimate: 0h
>
> Open sourcing this plugin we use internally within Yahoo in place of
> hierarchical caching.
> CARP is a plugin that allows you to group a bunch of ATS hosts into a cluster
> and share cache space across the entire group. This is done with consistent
> hashing on the object URL to generate an "owner" node in the cluster.
> Requests to any other node in the cluster will be forwarded on to the
> corresponding owner. More info in the README.
> Difference from internal version of note:
> I've ripped out some code we weren't entirely sure we could open source
> because of a hash function. If it turns out that we can open source this,
> I'll do so. The CarpHashAlgorithm class is meant to be extensible, so any
> consistent hash function can replace it. The function included here is pretty
> straightforward but not what we use in production, so just wanted to use that
> caveat.
> One last caveat:
> You'll see some code and documentation in here for object replication. This
> is something I added recently to CARP that allows you to specify an object be
> replicated a certain number of times in the cluster. This is useful if you
> have a network partition or if you're performing some sort of update. When an
> object's primary owner is unreachable, a node in the cluster can go to the
> secondary owner if it's available rather than having to fall all the way back
> to origin. While I've done some initial testing on this with my own cluster
> of hosts, it's not been tested in production so use at your own risk for now.
> I'll be sure to keep the open source community informed on the progress of
> our tests with this feature.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
