Canbin Zheng created FLINK-15822:
------------------------------------
Summary: Rethink the necessity of the internal Service
Key: FLINK-15822
URL: https://issues.apache.org/jira/browse/FLINK-15822
Project: Flink
Issue Type: Sub-task
Components: Deployment / Kubernetes
Affects Versions: 1.10.0
Reporter: Canbin Zheng
Fix For: 1.11.0, 1.10.1
The current design introduces two Kubernetes Services when deploying a new
session cluster. The rest Service serves external communication while the
internal Service mainly serves two purposes:
# A discovery service directs the communication from TaskManagers to the
JobManager Pod that has labels containing the internal Service’s selector in
the non-HA mode, so that the TM Pods could re-register to the new JM Pod once a
JM Pod failover occurs, while in the HA mode, there could be one active and
multiple standby JM Pods, so we use the Pod IP of the active one for internal
communication instead of using the internal Service .
# The OwnerReference of all other Kubernetes Resources, including the rest
Service, the ConfigMap and the JobManager Deployment.
Is it possible that we just create one single Service instead of two? I think
things could work quite well with only the rest Service, meanwhile the design
and code could be more succinct.
This ticket proposes to remove the internal Service, the core changes including
# In the non-HA mode, we use the rest Service as the JobManager Pod discovery
service.
# Set the JobManager Deployment as the OwnerReference of all the other
Kubernetes Resources, including the rest Service and the ConfigMap.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
