Configuring Oozie Failover (hot/cold)
The functionality described below is supported in CDH 5, but Cloudera recommends that you use the new capabilities introduced in CDH 5 instead.
- Set up your database for High Availability (see the database documentation for details).
Oozie database configuration properties may need special configuration (see the JDBC driver documentation for details).
- Configure Oozie on two or more servers:
- These servers should be configured identically
- Set the OOZIE_HTTP_HOSTNAME variable in oozie-env.sh to the Load Balancer or Virtual IP address (see step 3)
- Only one of the Oozie servers should be started (the hot server).
- Use either a Virtual IP Address or Load Balancer to direct traffic to the hot server.
- Access Oozie via the Virtual IP or Load Balancer address.
Points to note
- The Virtual IP Address or Load Balancer can be used to periodically check the health of the hot server.
- If something is wrong, you can shut down the hot server, start the cold server, and redirect the Virtual IP Address or Load Balancer to the new hot server.
- This can all be automated with a script, but a false positive indicating the hot server is down will cause problems, so test your script carefully.
- There will be no data loss.
- Any running workflows will continue from where they left off.
- It takes only about 15 seconds to start the Oozie server.
See also Configuring Oozie to Use HDFS HA.