use of org.apache.geode.internal.cache.partitioned.rebalance.ParallelBucketOperator in project geode by apache.
the class PartitionedRegionRebalanceOp method execute.
/**
* Do the actual rebalance
*
* @return the details of the rebalance.
*/
public Set<PartitionRebalanceInfo> execute() {
long start = System.nanoTime();
InternalResourceManager resourceManager = InternalResourceManager.getInternalResourceManager(leaderRegion.getCache());
MembershipListener listener = new MembershipChangeListener();
if (isRebalance) {
InternalResourceManager.getResourceObserver().rebalancingStarted(targetRegion);
} else {
InternalResourceManager.getResourceObserver().recoveryStarted(targetRegion);
}
RecoveryLock lock = null;
try {
if (!checkAndSetColocatedRegions()) {
return Collections.emptySet();
}
// have full redundancy.
if (!isRebalanceNecessary()) {
return Collections.emptySet();
}
if (!simulate) {
lock = leaderRegion.getRecoveryLock();
lock.lock();
}
// have fixed it already.
if (!isRebalanceNecessary()) {
return Collections.emptySet();
}
// register a listener to notify us if the new members leave or join.
// When a membership change occurs, we want to restart the rebalancing
// from the beginning.
// TODO rebalance - we should really add a membership listener to ALL of
// the colocated regions.
leaderRegion.getRegionAdvisor().addMembershipListener(listener);
PartitionedRegionLoadModel model = null;
InternalCache cache = leaderRegion.getCache();
Map<PartitionedRegion, InternalPRInfo> detailsMap = fetchDetails(cache);
BucketOperatorWrapper serialOperator = getBucketOperator(detailsMap);
ParallelBucketOperator parallelOperator = new ParallelBucketOperator(MAX_PARALLEL_OPERATIONS, cache.getDistributionManager().getWaitingThreadPool(), serialOperator);
model = buildModel(parallelOperator, detailsMap, resourceManager);
for (PartitionRebalanceDetailsImpl details : serialOperator.getDetailSet()) {
details.setPartitionMemberDetailsBefore(model.getPartitionedMemberDetails(details.getRegionPath()));
}
director.initialize(model);
for (; ; ) {
if (cancelled.get()) {
return Collections.emptySet();
}
if (membershipChange) {
membershipChange = false;
// refetch the partitioned region details after
// a membership change.
debug("Rebalancing {} detected membership changes. Refetching details", leaderRegion);
if (this.stats != null) {
this.stats.incRebalanceMembershipChanges(1);
}
model.waitForOperations();
detailsMap = fetchDetails(cache);
model = buildModel(parallelOperator, detailsMap, resourceManager);
director.membershipChanged(model);
}
leaderRegion.checkClosed();
cache.getCancelCriterion().checkCancelInProgress(null);
if (logger.isDebugEnabled()) {
logger.debug("Rebalancing {} Model:{}\n", leaderRegion, model);
}
if (!director.nextStep()) {
// Stop when the director says we can't rebalance any more.
break;
}
}
debug("Rebalancing {} complete. Model:{}\n", leaderRegion, model);
long end = System.nanoTime();
for (PartitionRebalanceDetailsImpl details : serialOperator.getDetailSet()) {
if (!simulate) {
details.setTime(end - start);
}
details.setPartitionMemberDetailsAfter(model.getPartitionedMemberDetails(details.getRegionPath()));
}
return Collections.<PartitionRebalanceInfo>unmodifiableSet(serialOperator.getDetailSet());
} finally {
if (lock != null) {
try {
lock.unlock();
} catch (CancelException e) {
// lock service has been destroyed
} catch (Exception e) {
logger.error(LocalizedMessage.create(LocalizedStrings.PartitionedRegionRebalanceOp_UNABLE_TO_RELEASE_RECOVERY_LOCK), e);
}
}
try {
if (isRebalance) {
InternalResourceManager.getResourceObserver().rebalancingFinished(targetRegion);
} else {
InternalResourceManager.getResourceObserver().recoveryFinished(targetRegion);
}
} catch (Exception e) {
logger.error(LocalizedMessage.create(LocalizedStrings.PartitionedRegionRebalanceOp_ERROR_IN_RESOURCE_OBSERVER), e);
}
try {
leaderRegion.getRegionAdvisor().removeMembershipListener(listener);
} catch (Exception e) {
logger.error(LocalizedMessage.create(LocalizedStrings.PartitionedRegionRebalanceOp_ERROR_IN_RESOURCE_OBSERVER), e);
}
}
}
Aggregations