org.apache.drill.yarn.zk

Class AMRegistry

java.lang.Object

org.apache.drill.yarn.zk.AMRegistry

public class AMRegistry
extends Object

Register this App Master in ZK to prevent duplicates.

Possible enhancement is to put the registry in some well-known location, such as /drill-am,

Constructor Summary

Constructors

Constructor and Description
AMRegistry(ZKClusterCoordinator zkCoord)

Method Summary

Modifier and Type	Method and Description
void	register(String amHost, int amPort, String amAppId) Register this AM as an ephemeral znode in ZK.
void	useLocalRegistry(String zkRoot, String clusterId)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

AMRegistry

public AMRegistry(ZKClusterCoordinator zkCoord)

Method Detail

useLocalRegistry

public void useLocalRegistry(String zkRoot,
                             String clusterId)

register

public void register(String amHost,
                     int amPort,
                     String amAppId)
              throws ZKRuntimeException

Register this AM as an ephemeral znode in ZK. The structure of ZK is as follows:

 /drill
 . <cluster-id>
 . . <Drillbit GUID> (Value is Proto-encoded drillbit info)
 . drill-on-yarn
 . . <cluster-id> (value: amHost:port)
 

The structure acknowledges that the cluster-id znode may be renamed, and there may be multiple cluster IDs for a single drill root node. (Odd, but supported.) To address this, we put the AM registrations in their own (persistent) znode: drill-on-yarn. Each is keyed by the cluster ID (so we can find it), and holds the host name, HTTP port and Application ID of the AM.

When the AM starts, it atomically checks and sets the AM registration. If another AM already is running, then this AM will fail, displaying a log error message with the host, port and (most importantly) app ID so the user can locate the problem.

Throws:

ZKRuntimeException