private boolean isHostAlreadyRunningTask(Protos.Offer offer) {
Boolean result = false;
List<Protos.TaskInfo> stateList = clusterState.getTaskList();
for (Protos.TaskInfo t : stateList) {
if (t.getSlaveId().equals(offer.getSlaveId())) {
result = true;
}
}
return result;
}
@Override
public void resourceOffers(SchedulerDriver schedulerDriver, List<Protos.Offer> list) {
double CPUS_PER_TASK = 1;
double MEM_PER_TASK = 128;
for (Protos.Offer offer : list) {
List<Protos.TaskInfo> taskInfoList = new ArrayList<Protos.TaskInfo>();
double offerCpus = 0;
double offerMem = 0;
for (Protos.Resource resource : offer.getResourcesList()) {
if (resource.getName().equals("cpus")) {
offerCpus += resource.getScalar().getValue();
} else if (resource.getName().equals("mem")) {
offerMem += resource.getScalar().getValue();
}
}
LOGGER.info(
"Received Offer : "
+ offer.getId().getValue()
+ " with cpus = "
+ offerCpus
+ " and mem ="
+ offerMem);
double remainingCpus = offerCpus;
double remainingMem = offerMem;
if (launchedTasks < totalTasks
&& remainingCpus >= CPUS_PER_TASK
&& remainingMem >= MEM_PER_TASK) {
Protos.TaskID taskID =
Protos.TaskID.newBuilder().setValue(Integer.toString(launchedTasks++)).build();
LOGGER.info(
"Launching task :"
+ taskID.getValue()
+ " using the offer : "
+ offer.getId().getValue());
Protos.TaskInfo piTaskInfo =
Protos.TaskInfo.newBuilder()
.setName("task " + taskID.getValue())
.setTaskId(taskID)
.setSlaveId(offer.getSlaveId())
.addResources(
Protos.Resource.newBuilder()
.setName("cpus")
.setType(Protos.Value.Type.SCALAR)
.setScalar(Protos.Value.Scalar.newBuilder().setValue(CPUS_PER_TASK)))
.addResources(
Protos.Resource.newBuilder()
.setName("mem")
.setType(Protos.Value.Type.SCALAR)
.setScalar(Protos.Value.Scalar.newBuilder().setValue(MEM_PER_TASK)))
.setExecutor(Protos.ExecutorInfo.newBuilder(piExecutor))
.build();
taskID = Protos.TaskID.newBuilder().setValue(Integer.toString(launchedTasks++)).build();
LOGGER.info(
"Launching task :"
+ taskID.getValue()
+ " using the offer : "
+ offer.getId().getValue());
taskInfoList.add(piTaskInfo);
}
schedulerDriver.launchTasks(offer.getId(), taskInfoList);
}
}
public SlaveMatchState doesOfferMatch(
Protos.Offer offer,
SingularityTaskRequest taskRequest,
SingularitySchedulerStateCache stateCache) {
final String host = offer.getHostname();
final String rackId = slaveAndRackHelper.getRackIdOrDefault(offer);
final String slaveId = offer.getSlaveId().getValue();
final MachineState currentSlaveState =
stateCache.getSlave(slaveId).get().getCurrentState().getState();
if (currentSlaveState == MachineState.FROZEN) {
return SlaveMatchState.SLAVE_FROZEN;
}
if (currentSlaveState.isDecommissioning()) {
return SlaveMatchState.SLAVE_DECOMMISSIONING;
}
final MachineState currentRackState =
stateCache.getRack(rackId).get().getCurrentState().getState();
if (currentRackState == MachineState.FROZEN) {
return SlaveMatchState.RACK_FROZEN;
}
if (currentRackState.isDecommissioning()) {
return SlaveMatchState.RACK_DECOMMISSIONING;
}
if (!taskRequest.getRequest().getRackAffinity().or(Collections.<String>emptyList()).isEmpty()) {
if (!taskRequest.getRequest().getRackAffinity().get().contains(rackId)) {
LOG.trace(
"Task {} requires a rack in {} (current rack {})",
taskRequest.getPendingTask().getPendingTaskId(),
taskRequest.getRequest().getRackAffinity().get(),
rackId);
return SlaveMatchState.RACK_AFFINITY_NOT_MATCHING;
}
}
final SlavePlacement slavePlacement =
taskRequest.getRequest().getSlavePlacement().or(configuration.getDefaultSlavePlacement());
if (!taskRequest.getRequest().isRackSensitive() && slavePlacement == SlavePlacement.GREEDY) {
return SlaveMatchState.NOT_RACK_OR_SLAVE_PARTICULAR;
}
final int numDesiredInstances = taskRequest.getRequest().getInstancesSafe();
double numOnRack = 0;
double numOnSlave = 0;
double numCleaningOnSlave = 0;
double numOtherDeploysOnSlave = 0;
final String sanitizedHost = JavaUtils.getReplaceHyphensWithUnderscores(host);
final String sanitizedRackId = JavaUtils.getReplaceHyphensWithUnderscores(rackId);
Collection<SingularityTaskId> cleaningTasks = stateCache.getCleaningTasks();
for (SingularityTaskId taskId :
SingularityTaskId.matchingAndNotIn(
stateCache.getActiveTaskIds(),
taskRequest.getRequest().getId(),
Collections.<SingularityTaskId>emptyList())) {
// TODO consider using executorIds
if (taskId.getSanitizedHost().equals(sanitizedHost)) {
if (taskRequest.getDeploy().getId().equals(taskId.getDeployId())) {
if (cleaningTasks.contains(taskId)) {
numCleaningOnSlave++;
} else {
numOnSlave++;
}
} else {
numOtherDeploysOnSlave++;
}
}
if (taskId.getSanitizedRackId().equals(sanitizedRackId)
&& !cleaningTasks.contains(taskId)
&& taskRequest.getDeploy().getId().equals(taskId.getDeployId())) {
numOnRack++;
}
}
if (taskRequest.getRequest().isRackSensitive()) {
final double numPerRack = numDesiredInstances / (double) stateCache.getNumActiveRacks();
final boolean isRackOk = numOnRack < numPerRack;
if (!isRackOk) {
LOG.trace(
"Rejecting RackSensitive task {} from slave {} ({}) due to numOnRack {} and cleaningOnSlave {}",
taskRequest.getRequest().getId(),
slaveId,
host,
numOnRack,
numCleaningOnSlave);
return SlaveMatchState.RACK_SATURATED;
}
}
switch (slavePlacement) {
case SEPARATE:
case SEPARATE_BY_DEPLOY:
if (numOnSlave > 0 || numCleaningOnSlave > 0) {
LOG.trace(
"Rejecting SEPARATE task {} from slave {} ({}) due to numOnSlave {} numCleaningOnSlave {}",
taskRequest.getRequest().getId(),
slaveId,
host,
numOnSlave,
numCleaningOnSlave);
return SlaveMatchState.SLAVE_SATURATED;
}
break;
case SEPARATE_BY_REQUEST:
if (numOnSlave > 0 || numCleaningOnSlave > 0 || numOtherDeploysOnSlave > 0) {
LOG.trace(
"Rejecting SEPARATE task {} from slave {} ({}) due to numOnSlave {} numCleaningOnSlave {} numOtherDeploysOnSlave {}",
taskRequest.getRequest().getId(),
slaveId,
host,
numOnSlave,
numCleaningOnSlave,
numOtherDeploysOnSlave);
return SlaveMatchState.SLAVE_SATURATED;
}
break;
case OPTIMISTIC:
final double numPerSlave = numDesiredInstances / (double) stateCache.getNumActiveSlaves();
final boolean isSlaveOk = numOnSlave < numPerSlave;
if (!isSlaveOk) {
LOG.trace(
"Rejecting OPTIMISTIC task {} from slave {} ({}) due to numOnSlave {}",
taskRequest.getRequest().getId(),
slaveId,
host,
numOnSlave);
return SlaveMatchState.SLAVE_SATURATED;
}
break;
case GREEDY:
}
return SlaveMatchState.OK;
}
@Override
public void resourceOffers(SchedulerDriver driver, List<Protos.Offer> offers) {
long masterCpu = Configuration.getInt(PropertyKey.INTEGRATION_MASTER_RESOURCE_CPU);
long masterMem =
Configuration.getBytes(PropertyKey.INTEGRATION_MASTER_RESOURCE_MEM) / Constants.MB;
long workerCpu = Configuration.getInt(PropertyKey.INTEGRATION_WORKER_RESOURCE_CPU);
long workerMem =
Configuration.getBytes(PropertyKey.INTEGRATION_WORKER_RESOURCE_MEM) / Constants.MB;
LOG.info(
"Master launched {}, master count {}, "
+ "requested master cpu {} mem {} MB and required master hostname {}",
mMasterLaunched,
mMasterCount,
masterCpu,
masterMem,
mRequiredMasterHostname);
for (Protos.Offer offer : offers) {
Protos.Offer.Operation.Launch.Builder launch = Protos.Offer.Operation.Launch.newBuilder();
double offerCpu = 0;
double offerMem = 0;
for (Protos.Resource resource : offer.getResourcesList()) {
if (resource.getName().equals(Constants.MESOS_RESOURCE_CPUS)) {
offerCpu += resource.getScalar().getValue();
} else if (resource.getName().equals(Constants.MESOS_RESOURCE_MEM)) {
offerMem += resource.getScalar().getValue();
} else {
// Other resources are currently ignored.
}
}
LOG.info(
"Received offer {} on host {} with cpus {} and mem {} MB and hasMasterPorts {}",
offer.getId().getValue(),
offer.getHostname(),
offerCpu,
offerMem,
OfferUtils.hasAvailableMasterPorts(offer));
Protos.ExecutorInfo.Builder executorBuilder = Protos.ExecutorInfo.newBuilder();
List<Protos.Resource> resources;
if (!mMasterLaunched
&& offerCpu >= masterCpu
&& offerMem >= masterMem
&& mMasterCount
< Configuration.getInt(PropertyKey.INTEGRATION_MESOS_ALLUXIO_MASTER_NODE_COUNT)
&& OfferUtils.hasAvailableMasterPorts(offer)
&& (mRequiredMasterHostname == null
|| mRequiredMasterHostname.equals(offer.getHostname()))) {
LOG.debug("Creating Alluxio Master executor");
executorBuilder
.setName("Alluxio Master Executor")
.setSource("master")
.setExecutorId(Protos.ExecutorID.newBuilder().setValue("master"))
.addAllResources(getExecutorResources())
.setCommand(
Protos.CommandInfo.newBuilder()
.setValue(createStartAlluxioCommand("alluxio-master-mesos.sh"))
.addAllUris(getExecutorDependencyURIList())
.setEnvironment(
Protos.Environment.newBuilder()
.addVariables(
Protos.Environment.Variable.newBuilder()
.setName("ALLUXIO_UNDERFS_ADDRESS")
.setValue(Configuration.get(PropertyKey.UNDERFS_ADDRESS))
.build())
.build()));
// pre-build resource list here, then use it to build Protos.Task later.
resources = getMasterRequiredResources(masterCpu, masterMem);
mMasterHostname = offer.getHostname();
mTaskName = Configuration.get(PropertyKey.INTEGRATION_MESOS_ALLUXIO_MASTER_NAME);
mMasterCount++;
mMasterTaskId = mLaunchedTasks;
} else if (mMasterLaunched
&& !mWorkers.contains(offer.getHostname())
&& offerCpu >= workerCpu
&& offerMem >= workerMem
&& OfferUtils.hasAvailableWorkerPorts(offer)) {
LOG.debug("Creating Alluxio Worker executor");
final String memSize = FormatUtils.getSizeFromBytes((long) workerMem * Constants.MB);
executorBuilder
.setName("Alluxio Worker Executor")
.setSource("worker")
.setExecutorId(Protos.ExecutorID.newBuilder().setValue("worker"))
.addAllResources(getExecutorResources())
.setCommand(
Protos.CommandInfo.newBuilder()
.setValue(createStartAlluxioCommand("alluxio-worker-mesos.sh"))
.addAllUris(getExecutorDependencyURIList())
.setEnvironment(
Protos.Environment.newBuilder()
.addVariables(
Protos.Environment.Variable.newBuilder()
.setName("ALLUXIO_MASTER_HOSTNAME")
.setValue(mMasterHostname)
.build())
.addVariables(
Protos.Environment.Variable.newBuilder()
.setName("ALLUXIO_WORKER_MEMORY_SIZE")
.setValue(memSize)
.build())
.addVariables(
Protos.Environment.Variable.newBuilder()
.setName("ALLUXIO_UNDERFS_ADDRESS")
.setValue(Configuration.get(PropertyKey.UNDERFS_ADDRESS))
.build())
.build()));
// pre-build resource list here, then use it to build Protos.Task later.
resources = getWorkerRequiredResources(workerCpu, workerMem);
mWorkers.add(offer.getHostname());
mTaskName = Configuration.get(PropertyKey.INTEGRATION_MESOS_ALLUXIO_WORKER_NAME);
} else {
// The resource offer cannot be used to start either master or a worker.
LOG.info("Declining offer {}", offer.getId().getValue());
driver.declineOffer(offer.getId());
continue;
}
Protos.TaskID taskId =
Protos.TaskID.newBuilder().setValue(String.valueOf(mLaunchedTasks)).build();
LOG.info("Launching task {} using offer {}", taskId.getValue(), offer.getId().getValue());
Protos.TaskInfo task =
Protos.TaskInfo.newBuilder()
.setName(mTaskName)
.setTaskId(taskId)
.setSlaveId(offer.getSlaveId())
.addAllResources(resources)
.setExecutor(executorBuilder)
.build();
launch.addTaskInfos(Protos.TaskInfo.newBuilder(task));
mLaunchedTasks++;
// NOTE: We use the new API `acceptOffers` here to launch tasks.
// The 'launchTasks' API will be deprecated.
List<Protos.OfferID> offerIds = new ArrayList<Protos.OfferID>();
offerIds.add(offer.getId());
List<Protos.Offer.Operation> operations = new ArrayList<Protos.Offer.Operation>();
Protos.Offer.Operation operation =
Protos.Offer.Operation.newBuilder()
.setType(Protos.Offer.Operation.Type.LAUNCH)
.setLaunch(launch)
.build();
operations.add(operation);
Protos.Filters filters = Protos.Filters.newBuilder().setRefuseSeconds(1).build();
driver.acceptOffers(offerIds, operations, filters);
}
}