I’ve searched around for this and see other questions, but no answers. I’m hoping someone can give some insight to what appears to be a pretty basic problem.
We are using Ambari for a little cluster of ZK/Kafka/HDFS/MR/YARN. When a machine gets rebooted, we’d like the services on that machine to automatically restart. I found the init.d script for the Ganglia components and can start that. However, through the host starts and Ambari can see it, any of the services (like ZK and Kafka, for example) do not start until I log in to the Ambari console and click “start”.
Note that, if you start the services manually on the console or the hosting server, Ambari apparently has no idea that they are running. You must start them via Ambari for the monitoring to be correct.
We thought about writing an init.d script on each host that would send an HTTP request to Ambari telling it to start the relevant services on that host. Our concern is that Ambari now becomes a single point of failure in what is otherwise a fully redundant system.
Can someone either:
a) tell me that it is a spectacularly bad idea to autostart services such as ZK and Kafka, or
b) point me in the right direction to manage this with Ambari
Much appreciated.