Tom Beerbower created AMBARI-15192: -------------------------------------- Summary: Atlas Integration : Atlas Server fails to properly start if Zookeeper isn't started first Key: AMBARI-15192 URL: https://issues.apache.org/jira/browse/AMBARI-15192 Project: Ambari Issue Type: Bug Reporter: Tom Beerbower Assignee: Tom Beerbower
When Atlas Server version 0.6 is started, it creates a Kafka consumer which attempts to connect to Zookeeper. The atlas startup script returns a status of 0 immediately, not waiting for the server to actually start successfully. Because Atlas now has a dependency on Kafka and ZK, this needs to be expressed in role_command_order.json for UI installs. But since we use the same stack definition for both Atlas 0.5 and 0.6 installs and only 0.6 has the Kafka and ZK dependencies we need to ensure that we don't negatively affect 0.5 installs. For blueprint installs, because there is no longer cluster wide ordering for install and start, role_command_order.json won't help as ZK could be on another host. I think that we should add the ordering for UI installs and write an Atlas wrapper startup script in the stack definition that blocks until the web UI is accessible or a timeout occurs. If the server is started successfully the script should return a failure code(or exception ?) so that ambari retry logic would kick in if configured as it is for BP installs. We should also consider modifying the Atlas startup script to block until the server is actually started. -- This message was sent by Atlassian JIRA (v6.3.4#6332)