Till Rohrmann created FLINK-3346:
------------------------------------
Summary: Replace Akka death watch with own heartbeats
Key: FLINK-3346
URL: https://issues.apache.org/jira/browse/FLINK-3346
Project: Flink
Issue Type: Improvement
Components: Distributed Runtime
Affects Versions: 1.0.0
Reporter: Till Rohrmann
Priority: Minor
Currently, we're using Akka's death watch to detect failed instances (e.g. the
JM watches the TM and vice versa). Whenever a death watch is triggered, the
corresponding {{ActorSystem}} is quarantined. This will prevent the quarantined
{{ActorSystem}} from reconnecting to the detecting system, in case that it
didn't die. This can happen once in a while given a sub-optimal death watch
configuration.
In order to prevent this from happening, we could implement our own heartbeats
which are periodically sent from the {{TaskManagers}} to the {{JobManager}} and
vice versa, for example. Not receiving a certain number of heartbeats could
then mean that the {{TaskManager}} is marked as dead.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
