[
https://issues.apache.org/jira/browse/YARN-6405?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15978126#comment-15978126
]
Jian He edited comment on YARN-6405 at 4/21/17 6:14 AM:
--------------------------------------------------------
bq. Instead of having a fixed set of names 1 to N, when a container goes down
and comes back up it will have name N+1, and there will be no container with
the original name.
[~billie.rinaldi], yes, you are right. One question is
If the container fails and re-allocated and started on the same host, I can
understand we should reuse the component name, but for a container allocated
and restarted on a different host, should it also reuse the same component
name? It would lead to below situation, if it does:
1. container1 started on host1, as component1
2. container2 started on host2, as component2
3. container1 failed and restarted on host2, as component1
4. container2 failed and restarted on host1, as component2
In this scenario, initially, we have component1 on host1, component2 on host2.
Then later we end up with component1 on host2 and component2 on host1. And this
can get increasingly messy with a large number of containers failed and
restarted. Is this expected behavior ?
was (Author: jianhe):
bq. Instead of having a fixed set of names 1 to N, when a container goes down
and comes back up it will have name N+1, and there will be no container with
the original name.
[~billie.rinaldi], yes, you are right. One question is
If the container fails and re-allocated and started on the same host, I can
understand we should reuse the component name, but for a container allocated
and restarted on a different host, should it also reuse the same component
name? It would lead to below situation, if it does:
1. container1 started on host1, as component1
2. container2 started on host2, as component2
2. container1 failed and restarted on host2, as component1
3. container2 failed and restarted on host1, as component2
In this scenario, initially, we have component1 on host1, component2 on host2.
Then later we end up with component1 on host2 and component2 on host1. And this
can get increasingly messy with a large number of containers failed and
restarted. Is this expected behavior ?
> Improve configuring services through REST API
> ---------------------------------------------
>
> Key: YARN-6405
> URL: https://issues.apache.org/jira/browse/YARN-6405
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Jian He
> Assignee: Jian He
> Attachments: unique-names-test.patch,
> YARN-6405.yarn-native-services.01.patch,
> YARN-6405.yarn-native-services.02.patch,
> YARN-6405.yarn-native-services.03.patch,
> YARN-6405.yarn-native-services.04.patch,
> YARN-6405.yarn-native-services.05.patch,
> YARN-6405.yarn-native-services.06.patch
>
>
> YARN-4793 defined various ways that user can config services through the
> configuration section of the REST API. But, some semantics are not yet
> supported in the back-end server (AM). YARN-6255 has done some work, this
> jira is to complete this task - support configuring services through REST API
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
