You do not need a SecondaryNameNode in HA. Please delete the role to resolve your issue. You can read the HA Architecture at http://archive.cloudera.com/cdh5/cdh/5/hadoop/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.html#Hardware_resources to further understand how HA works. ... View more

If you cannot get the admins to provide your app with a near-infinite ticket lifetime, you need to instead loop the login process via a daemon thread, such as via this thread snippet as an example: private void runLoginAndRenewalThread(final Configuration conf) throws IOException { // Do a blocking login first SecurityUtil.login(conf, KEYTAB_CONF, KEYTAB_USER); // Spawn the relogin thread next Thread reloginThread = new Thread() { @Override public void run() { while (true) { try { SecurityUtil.login(conf, KEYTAB_CONF, KEYTAB_USER); Thread.sleep(ELEVEN_MINUTES); } catch (IOException e) { e.printStackTrace(); interrupt(); } catch (InterruptedException e) { e.printStackTrace(); interrupt(); } } } }; reloginThread.setDaemon(true); reloginThread.start(); } ... View more

Ha, looks like Camus runs local job runner, that is the problem... Need to inform camus that we have yarn here. ... View more

The block mapping of block ID <=> DN location is not stored in the fsimage of HDFS, it is just kept in the running NameNode memory. Since the primary work of the block report is to report statuses of availability of each block replica on the DNs, and the checkpoint's work is to persist the namespace and namespace-associative states (such as snapshots, etc.), the checkpoint can be an operation done without problems in parallel as their two information structures will not conflict with one another. ... View more

This is JIRA HUE-2551 The two options currently available are - kill the coordinator and create a new one - run the workflow manually after the change. ... View more