[
https://issues.apache.org/jira/browse/TS-3999?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15202074#comment-15202074
]
John Rushford commented on TS-3999:
-----------------------------------
I agree with and like the idea of using a list labeled "peers" rather than
"parent" as I think that makes the intent clear as was pointed out. So when
the list is labeled "peers" one of the hostnames is self and if on startup ATS
cannot identify itself from the list, the entire rule is rejected and the error
logged.
Now I added that secondary_parent list with the secondary hash ring feature,
this is entirely optional when using consistent_hash. Do you think that this
secondary_parent list should be allowed as part of a "peers" config? In case
the "peers" are unreachable then the secondary_parents list could be used as
alternate "peers"?
> Add option to go direct for specific parent entry
> -------------------------------------------------
>
> Key: TS-3999
> URL: https://issues.apache.org/jira/browse/TS-3999
> Project: Traffic Server
> Issue Type: Improvement
> Components: Parent Proxy
> Reporter: Alan M. Carroll
> Assignee: Alan M. Carroll
> Labels: A
> Fix For: 7.0.0
>
>
> We want to use parenty proxying in a peer relationship so that a host can be
> both a parent and child in general but only for a specific URL. Currently
> this can be done using the CARP plugin but that can create some difficulties.
> Being able to specify that one parent in the list is actually direct to
> origin would make that much easier.
> The current suggestion is to overload the port specifier as a indicator of
> direct. For example, if you had three hosts in a pod, {{rikku}}, {{tidus}},
> and {{yuna}}, then you would configure the parent proxying on {{tidus}} as
> {code}
> "tidus:@direct, yuna:8080, rikku:8080"
> {code}
> while the configuration on {{yuna}} would be
> {code}
> "tidus:8080, yuna:@direct, rikku:8080"
> {code}.
> Similarly for {{rikku}} the port would be changed to "@direct" just for the
> "rikku" parent. I discussed several configuration options for this with
> [~dcarlin] and putting the override directly in the parent list was his
> preferred mechanism. Note that in order to have consistency between hosts in
> a pod, the list of names *must* be exactly the same across all the peers. In
> this case, it must be "tidus, yuna, rikku" for all three or loops will occur
> because of different hash seeds.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
