Prasanth Jayachandran created TEZ-3078:
------------------------------------------
Summary: Provide a mechanism for AM to let Client know about the
reason for failure
Key: TEZ-3078
URL: https://issues.apache.org/jira/browse/TEZ-3078
Project: Apache Tez
Issue Type: Improvement
Affects Versions: 0.8.3
Reporter: Prasanth Jayachandran
When working on HIVE-12959 for LLAP, the requirement is when we submit a query
to LLAP task scheduler and if there are no LLAP daemons we should fail the
query instead of waiting indefinitely for daemons to show up. For this to work,
the task scheduler has to provide a mechanism to let the AM know that the
scheduler service cannot proceed further as there are no daemons running.
Currently there is no way for the task scheduler to let AM know about this
information. The only way right now is to send back exception using
TaskSchedulerContext.onError() API. This will kill the AM but AM will restart
to recover the DAG. It will be better if there a way to let AM know about
daemon status via some status response based on which AM should avoid
restarting. It will be even better if we can provide a way for AM to
communicate this information back to the client (hive CLI or HiveServer2).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
