Rentao Wu created OOZIE-3624: -------------------------------- Summary: Oozie scheduled workflows fail when yarn/hdfs cluster changes Key: OOZIE-3624 URL: https://issues.apache.org/jira/browse/OOZIE-3624 Project: Oozie Issue Type: Improvement Components: coordinator, workflow Affects Versions: 5.2.0 Reporter: Rentao Wu
When the yarn cluster which is used by a Oozie scheduled workflow gets recreated with a new cluster, future runs of the scheduled workflow will break as they depend on the workflow/ job.properties files which was deployed on hdfs. The yarn jobtracker will also no longer work due to: {noformat} Caused by: org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.security.token.SecretManager$InvalidToken): appattempt_1622844783178_0004_000002 not found in AMRMTokenSecretManager. {noformat} It seem there are some tokens store in yarn and when the yarn cluster gets terminated and replaced with a new yarn cluster. The oozie launcher will hit this error message. The yarn cluster getting recreated is a common case in cloud, I'm wondering is there a way for oozie to be resilient to the underlying yarn cluster being ephemeral? is it supported for workflow/coordinator/ job.properties files to be deployed on s3 instead of hdfs? -- This message was sent by Atlassian Jira (v8.3.4#803005)