public void init(Configuration conf) {
LOG.info("QueryMaster init");
try {
this.systemConf = (TajoConf) conf;
this.connPool = RpcConnectionPool.getPool(systemConf);
querySessionTimeout = systemConf.getIntVar(TajoConf.ConfVars.QUERY_SESSION_TIMEOUT);
queryMasterContext = new QueryMasterContext(systemConf);
clock = new SystemClock();
this.dispatcher = new TajoAsyncDispatcher("querymaster_" + System.currentTimeMillis());
addIfService(dispatcher);
this.storageManager = StorageManagerFactory.getStorageManager(systemConf);
globalPlanner = new GlobalPlanner(systemConf, workerContext);
dispatcher.register(QueryStartEvent.EventType.class, new QueryStartEventHandler());
} catch (Throwable t) {
LOG.error(t.getMessage(), t);
throw new RuntimeException(t);
}
super.init(conf);
}
@Override
public void run() {
LOG.info("Start TajoWorkerAllocationThread");
CallFuture<TajoMasterProtocol.WorkerResourceAllocationResponse> callBack =
new CallFuture<TajoMasterProtocol.WorkerResourceAllocationResponse>();
int requiredMemoryMBSlot = 512; // TODO
int requiredDiskSlots = 1; // TODO
TajoMasterProtocol.WorkerResourceAllocationRequest request =
TajoMasterProtocol.WorkerResourceAllocationRequest.newBuilder()
.setMemoryMBSlots(requiredMemoryMBSlot)
.setDiskSlots(requiredDiskSlots)
.setNumWorks(event.getRequiredNum())
.setExecutionBlockId(event.getExecutionBlockId().getProto())
.build();
RpcConnectionPool connPool = RpcConnectionPool.getPool(queryTaskContext.getConf());
NettyClientBase tmClient = null;
try {
tmClient =
connPool.getConnection(
queryTaskContext.getQueryMasterContext().getWorkerContext().getTajoMasterAddress(),
TajoMasterProtocol.class,
true);
TajoMasterProtocol.TajoMasterProtocolService masterClientService = tmClient.getStub();
masterClientService.allocateWorkerResources(null, request, callBack);
} catch (Exception e) {
connPool.closeConnection(tmClient);
tmClient = null;
LOG.error(e.getMessage(), e);
} finally {
connPool.releaseConnection(tmClient);
}
TajoMasterProtocol.WorkerResourceAllocationResponse response = null;
while (!stopped.get()) {
try {
response = callBack.get(3, TimeUnit.SECONDS);
break;
} catch (InterruptedException e) {
if (stopped.get()) {
return;
}
} catch (TimeoutException e) {
LOG.info("No available worker resource for " + event.getExecutionBlockId());
continue;
}
}
int numAllocatedWorkers = 0;
if (response != null) {
List<TajoMasterProtocol.WorkerAllocatedResource> workerHosts =
response.getWorkerAllocatedResourceList();
ExecutionBlockId executionBlockId = event.getExecutionBlockId();
List<Container> containers = new ArrayList<Container>();
for (TajoMasterProtocol.WorkerAllocatedResource eachWorker : workerHosts) {
TajoWorkerContainer container = new TajoWorkerContainer();
NodeIdPBImpl nodeId = new NodeIdPBImpl();
nodeId.setHost(eachWorker.getWorkerHost());
nodeId.setPort(eachWorker.getPeerRpcPort());
TajoWorkerContainerId containerId = new TajoWorkerContainerId();
containerId.setApplicationAttemptId(
ApplicationIdUtils.createApplicationAttemptId(executionBlockId.getQueryId()));
containerId.setId(containerIdSeq.incrementAndGet());
container.setId(containerId);
container.setNodeId(nodeId);
WorkerResource workerResource = new WorkerResource();
workerResource.setAllocatedHost(nodeId.getHost());
workerResource.setPeerRpcPort(nodeId.getPort());
workerResource.setQueryMasterPort(eachWorker.getQueryMasterPort());
workerResource.setPullServerPort(eachWorker.getWorkerPullServerPort());
workerResource.setMemoryMBSlots(requiredMemoryMBSlot);
workerResource.setDiskSlots(requiredDiskSlots);
container.setWorkerResource(workerResource);
containers.add(container);
}
SubQueryState state = queryTaskContext.getSubQuery(executionBlockId).getState();
if (!SubQuery.isRunningState(state)) {
List<WorkerResource> workerResources = new ArrayList<WorkerResource>();
for (Container eachContainer : containers) {
workerResources.add(((TajoWorkerContainer) eachContainer).getWorkerResource());
}
try {
TajoContainerProxy.releaseWorkerResource(
queryTaskContext, executionBlockId, workerResources);
} catch (Exception e) {
LOG.error(e.getMessage(), e);
}
return;
}
if (workerHosts.size() > 0) {
if (LOG.isDebugEnabled()) {
LOG.debug("SubQueryContainerAllocationEvent fire:" + executionBlockId);
}
queryTaskContext
.getEventHandler()
.handle(new SubQueryContainerAllocationEvent(executionBlockId, containers));
}
numAllocatedWorkers += workerHosts.size();
}
if (event.getRequiredNum() > numAllocatedWorkers) {
ContainerAllocationEvent shortRequestEvent =
new ContainerAllocationEvent(
event.getType(),
event.getExecutionBlockId(),
event.getPriority(),
event.getResource(),
event.getRequiredNum() - numAllocatedWorkers,
event.isLeafQuery(),
event.getProgress());
queryTaskContext.getEventHandler().handle(shortRequestEvent);
}
LOG.info("Stop TajoWorkerAllocationThread");
}