liuxun created ZEPPELIN-3612:
--------------------------------
Summary: Cluster High availability module design
Key: ZEPPELIN-3612
URL: https://issues.apache.org/jira/browse/ZEPPELIN-3612
Project: Zeppelin
Issue Type: Sub-task
Components: zeppelin-server
Affects Versions: 0.9.0
Reporter: liuxun
Fix For: 0.9.0
h3. In the case of a partial Zeppelin-Server service process or server
exception, the service can continue to be served; the Zeppelin-Server service
can sense the availability of all Interpreter processes in the server cluster;
# *Raft protocol*
The Raft protocol ensures that only N/2+1 servers in the cluster need to be in
a normal state without affecting the service.
# *Interpreter process monitoring*
The Interpreter process creates a process heartbeat thread through the
ClusterMonitor class, and periodically sends the Interpreter process heartbeat
information and the IP and port information of the Thrift interface to the
cluster.
When the Interpreter process is closed, the process of deleting the process
metadata is sent to the cluster.
The process health check thread is created in the Zeppelin-Server through the
ClusterMonitor class. The heartbeat of all Interpreter processes in the Cluster
MetaData is periodically checked. If the timeout expires, the process metadata
is deleted to prevent the Interpreter process from being abnormal.
# *Interpreter process rebuild*
When the interpreter process is created, the Zeppelin-Server detects the
session information of the Interpreter process. First, it checks whether the
process is valid. If it is not available, the corresponding session is cleared
and the Interpreter process is re-created. Preventing an Interpreter process
from being abnormal or a server exception on the process causes Interpreter to
be unavailable.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
