Created 11-04-2015 12:53 AM
I was always wondering what value I should use for "jobtTacker" in my job.properites for Resource Manager HA.
Now a customer asked same question, so I thought this might be a good opportunity to find out.
Does anyone know which string we should use to utilize YARN Resource Manager HA?
According to Google, Cloudera uses "logicaljt" but I don't see this string in HDP code so far.
Created 11-04-2015 12:58 AM
Useful information. Link
When there are multiple RMs, the configuration (yarn-site.xml) used by clients and nodes is expected to list all the RMs. Clients, ApplicationMasters (AMs) and NodeManagers (NMs) try connecting to the RMs in a round-robin fashion until they hit the Active RM. If the Active goes down, they resume the round-robin polling until they hit the “new” Active. This default retry logic is implemented as org.apache.hadoop.yarn.client.ConfiguredRMFailoverProxyProvider. You can override the logic by implementing org.apache.hadoop.yarn.client.RMFailoverProxyProvider and setting the value of yarn.client.failover-proxy-provider to the class name.
Editing based on the discussion.
Created 11-04-2015 12:58 AM
Useful information. Link
When there are multiple RMs, the configuration (yarn-site.xml) used by clients and nodes is expected to list all the RMs. Clients, ApplicationMasters (AMs) and NodeManagers (NMs) try connecting to the RMs in a round-robin fashion until they hit the Active RM. If the Active goes down, they resume the round-robin polling until they hit the “new” Active. This default retry logic is implemented as org.apache.hadoop.yarn.client.ConfiguredRMFailoverProxyProvider. You can override the logic by implementing org.apache.hadoop.yarn.client.RMFailoverProxyProvider and setting the value of yarn.client.failover-proxy-provider to the class name.
Editing based on the discussion.
Created 11-04-2015 01:00 AM
But how about "jobTracker" in job.properties? I need to type "jobTrackers=hostname:8032"
Created 11-04-2015 01:03 AM
@Hajime What version of HDP are you using?
Created 11-04-2015 01:06 AM
i'm using 2.3.2 (but customer is 2.2.4.2)
Created 11-04-2015 01:39 AM
@jeff@hortonworks.com This is something we need to look into.
Created 11-04-2015 04:39 PM
I believe you can specify one of the nodes, if it's not the active RM, API will detect the proper RM and redirect.
Created 11-04-2015 01:32 PM
Based on this discussion with Hajime "this is what we have in our repo. Credit to Hajime for finding that link"
Created 02-03-2016 01:57 AM
@Hajime usually you pick one of the nodes for RM as far as Oozie is concerned. If you issue to a standby node, it will round-robin to the active RM.
Created 05-04-2016 01:17 AM
This is the hortonworks docs that gives reference to it.