You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by beyond1920 <gi...@git.apache.org> on 2016/09/23 11:31:44 UTC
[GitHub] flink pull request #2543: [FLINK-4606] [cluster management] Add watch mechan...
GitHub user beyond1920 opened a pull request:
https://github.com/apache/flink/pull/2543
[FLINK-4606] [cluster management] Add watch mechanism on current RPC framework
This pr aims at adding watch mechanism on current RPC framework, so RPC gateway could be watched to make sure the rpc server is running just like previous DeathWatch in akka.
There are following main differences:
1. Add watch and unwatch method to RpcEndpoint class
2. Implement WatchOperationHandler in AkkaInvocationHandler, so it could handle watch and unwatch operation
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/alibaba/flink jira-4670
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/2543.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2543
----
commit ed8e1b46b4cb20100fcf161ee9e2b4e4622aaf85
Author: beyond1920 <be...@126.com>
Date: 2016-09-23T00:46:30Z
watch and unwatch mechanism in new rpc framework
Summary: watch and unwatch mechanism in new rpc framework
Test Plan: junit
Reviewers: #blink, kete.yangkt
Differential Revision: http://phabricator.taobao.net/D5849
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---
[GitHub] flink pull request #2543: [FLINK-4670] [cluster management] Add watch mechan...
Posted by beyond1920 <gi...@git.apache.org>.
Github user beyond1920 closed the pull request at:
https://github.com/apache/flink/pull/2543
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---
[GitHub] flink issue #2543: [FLINK-4670] [cluster management] Add watch mechanism on ...
Posted by beyond1920 <gi...@git.apache.org>.
Github user beyond1920 commented on the issue:
https://github.com/apache/flink/pull/2543
Hi, till. Thanks for your review.
First sorry to bind to the wrong jira link, already correct it.
About heartbeatManager, i thought we could separate HeartbeatManager into two part, each part focus only on one function.
1. Death watch to monitor liveness, it could register watch, unregister watch, and notify the dead targets. just like deathwatch in akka. Although now I implement it on Akka's dead watch functionality, but interface is defined. It's fine to give another implementation of it in future.
2. Payload deliver periodically. But because data sync behavior is different in different component, e.g, TM,JM,RM, so I'm not sure it is good to extract the logic into a common logic.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---
[GitHub] flink issue #2543: [FLINK-4606] [cluster management] Add watch mechanism on ...
Posted by tillrohrmann <gi...@git.apache.org>.
Github user tillrohrmann commented on the issue:
https://github.com/apache/flink/pull/2543
The linked JIRA issue seems to be wrong.
I'm not completely sure whether we want to add this functionality to the `RpcService` or not. At the moment it is easy to implement because we can rely on Akka's death watch functionality. But future implementations might work differently. Thus, we would be bound to reimplement the death watch functionality.
Furthermore, once a target actor system has been marked as "dead" by the deathwatch it will be quarantined. If it was marked wrongly, then there is no way for the target actor system to talk to the watching system and vice versa.
What's the benefit over having a separate service for the heartbeat like it is done in #2435?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---