List of usage examples for org.joda.time Duration Duration
public Duration(ReadableInstant start, ReadableInstant end)
From source file:com.metamx.druid.indexing.coordinator.scaling.SimpleResourceManagementStrategy.java
License:Open Source License
@Override public boolean doTerminate(Collection<RemoteTaskRunnerWorkItem> pendingTasks, Collection<ZkWorker> zkWorkers) { Set<String> workerNodeIds = Sets.newHashSet(autoScalingStrategy .ipToIdLookup(Lists.newArrayList(Iterables.transform(zkWorkers, new Function<ZkWorker, String>() { @Override// w ww . ja va2 s.c om public String apply(ZkWorker input) { return input.getWorker().getIp(); } })))); Set<String> stillExisting = Sets.newHashSet(); for (String s : currentlyTerminating) { if (workerNodeIds.contains(s)) { stillExisting.add(s); } } currentlyTerminating.clear(); currentlyTerminating.addAll(stillExisting); boolean nothingTerminating = currentlyTerminating.isEmpty(); if (nothingTerminating) { final int minNumWorkers = workerSetupdDataRef.get().getMinNumWorkers(); if (zkWorkers.size() <= minNumWorkers) { log.info("Only [%d <= %d] nodes in the cluster, not terminating anything.", zkWorkers.size(), minNumWorkers); return false; } List<ZkWorker> thoseLazyWorkers = Lists .newArrayList(FunctionalIterable.create(zkWorkers).filter(new Predicate<ZkWorker>() { @Override public boolean apply(ZkWorker input) { return input.getRunningTasks().isEmpty() && System.currentTimeMillis() - input.getLastCompletedTaskTime().getMillis() >= config .getMaxWorkerIdleTimeMillisBeforeDeletion(); } })); int maxPossibleNodesTerminated = zkWorkers.size() - minNumWorkers; int numNodesToTerminate = Math.min(maxPossibleNodesTerminated, thoseLazyWorkers.size()); if (numNodesToTerminate <= 0) { log.info("Found no nodes to terminate."); return false; } AutoScalingData terminated = autoScalingStrategy.terminate(Lists .transform(thoseLazyWorkers.subList(0, numNodesToTerminate), new Function<ZkWorker, String>() { @Override public String apply(ZkWorker input) { return input.getWorker().getIp(); } })); if (terminated != null) { currentlyTerminating.addAll(terminated.getNodeIds()); lastTerminateTime = new DateTime(); scalingStats.addTerminateEvent(terminated); return true; } } else { Duration durSinceLastTerminate = new Duration(lastTerminateTime, new DateTime()); log.info("%s still terminating. Wait for all nodes to terminate before trying again.", currentlyTerminating); if (durSinceLastTerminate.isLongerThan(config.getMaxScalingDuration())) { log.makeAlert("Worker node termination taking too long!") .addData("millisSinceLastTerminate", durSinceLastTerminate.getMillis()) .addData("terminatingCount", currentlyTerminating.size()).emit(); currentlyTerminating.clear(); } } return false; }
From source file:com.metamx.druid.indexing.coordinator.scaling.SimpleResourceManagementStrategy.java
License:Open Source License
private boolean hasTaskPendingBeyondThreshold(Collection<RemoteTaskRunnerWorkItem> pendingTasks) { long now = System.currentTimeMillis(); for (TaskRunnerWorkItem pendingTask : pendingTasks) { if (new Duration(pendingTask.getQueueInsertionTime().getMillis(), now) .isEqual(config.getMaxPendingTaskDuration()) || new Duration(pendingTask.getQueueInsertionTime().getMillis(), now) .isLongerThan(config.getMaxPendingTaskDuration())) { return true; }/* w w w .j av a 2 s .c o m*/ } return false; }
From source file:com.metamx.druid.merger.coordinator.RemoteTaskRunner.java
License:Open Source License
@LifecycleStart public void start() { try {// ww w. j a v a 2 s . c o m workerPathCache.getListenable().addListener(new PathChildrenCacheListener() { @Override public void childEvent(CuratorFramework client, final PathChildrenCacheEvent event) throws Exception { if (event.getType().equals(PathChildrenCacheEvent.Type.CHILD_ADDED)) { final Worker worker = jsonMapper.readValue(event.getData().getData(), Worker.class); log.info("New worker[%s] found!", worker.getHost()); addWorker(worker); } else if (event.getType().equals(PathChildrenCacheEvent.Type.CHILD_REMOVED)) { final Worker worker = jsonMapper.readValue(event.getData().getData(), Worker.class); log.info("Worker[%s] removed!", worker.getHost()); removeWorker(worker); } } }); workerPathCache.start(); // Schedule termination of worker nodes periodically Period period = new Period(config.getTerminateResourcesDuration()); PeriodGranularity granularity = new PeriodGranularity(period, config.getTerminateResourcesOriginDateTime(), null); final long startTime = granularity.next(granularity.truncate(new DateTime().getMillis())); ScheduledExecutors.scheduleAtFixedRate(scheduledExec, new Duration(System.currentTimeMillis(), startTime), config.getTerminateResourcesDuration(), new Runnable() { @Override public void run() { if (currentlyTerminating.isEmpty()) { if (zkWorkers.size() <= workerSetupManager.getWorkerSetupData() .getMinNumWorkers()) { return; } int workerCount = 0; List<WorkerWrapper> thoseLazyWorkers = Lists.newArrayList(); for (WorkerWrapper workerWrapper : zkWorkers.values()) { workerCount++; if (workerCount > workerSetupManager.getWorkerSetupData().getMinNumWorkers() && workerWrapper.getRunningTasks().isEmpty() && System.currentTimeMillis() - workerWrapper.getLastCompletedTaskTime().getMillis() > config .getMaxWorkerIdleTimeMillisBeforeDeletion()) { thoseLazyWorkers.add(workerWrapper); } } AutoScalingData terminated = strategy.terminate( Lists.transform(thoseLazyWorkers, new Function<WorkerWrapper, String>() { @Override public String apply(WorkerWrapper input) { return input.getWorker().getIp(); } })); if (terminated != null) { currentlyTerminating.addAll(terminated.getNodeIds()); lastTerminateTime = new DateTime(); } } else { Duration durSinceLastTerminate = new Duration(new DateTime(), lastTerminateTime); if (durSinceLastTerminate.isLongerThan(config.getMaxScalingDuration())) { log.makeAlert("Worker node termination taking too long") .addData("millisSinceLastTerminate", durSinceLastTerminate.getMillis()) .addData("terminatingCount", currentlyTerminating.size()).emit(); } log.info( "%s still terminating. Wait for all nodes to terminate before trying again.", currentlyTerminating); } } }); started = true; } catch (Exception e) { throw Throwables.propagate(e); } }
From source file:com.metamx.druid.merger.coordinator.RemoteTaskRunner.java
License:Open Source License
private WorkerWrapper findWorkerForTask() { try {/*from w w w . j a v a 2s . c om*/ final MinMaxPriorityQueue<WorkerWrapper> workerQueue = MinMaxPriorityQueue .<WorkerWrapper>orderedBy(new Comparator<WorkerWrapper>() { @Override public int compare(WorkerWrapper w1, WorkerWrapper w2) { return -Ints.compare(w1.getRunningTasks().size(), w2.getRunningTasks().size()); } }).create(FunctionalIterable.create(zkWorkers.values()).filter(new Predicate<WorkerWrapper>() { @Override public boolean apply(WorkerWrapper input) { return (!input.isAtCapacity() && input.getWorker().getVersion() .compareTo(workerSetupManager.getWorkerSetupData().getMinVersion()) >= 0); } })); if (workerQueue.isEmpty()) { log.info("Worker nodes do not have capacity to run any more tasks!"); if (currentlyProvisioning.isEmpty()) { AutoScalingData provisioned = strategy.provision(); if (provisioned != null) { currentlyProvisioning.addAll(provisioned.getNodeIds()); lastProvisionTime = new DateTime(); } } else { Duration durSinceLastProvision = new Duration(new DateTime(), lastProvisionTime); if (durSinceLastProvision.isLongerThan(config.getMaxScalingDuration())) { log.makeAlert("Worker node provisioning taking too long") .addData("millisSinceLastProvision", durSinceLastProvision.getMillis()) .addData("provisioningCount", currentlyProvisioning.size()).emit(); } log.info( "%s still provisioning. Wait for all provisioned nodes to complete before requesting new worker.", currentlyProvisioning); } return null; } return workerQueue.peek(); } catch (Exception e) { throw Throwables.propagate(e); } }
From source file:com.metamx.druid.merger.coordinator.scaling.SimpleResourceManagementStrategy.java
License:Open Source License
@Override public boolean doProvision(Collection<TaskRunnerWorkItem> pendingTasks, Collection<ZkWorker> zkWorkers) { if (zkWorkers.size() >= workerSetupdDataRef.get().getMaxNumWorkers()) { log.info("Cannot scale anymore. Num workers = %d, Max num workers = %d", zkWorkers.size(), workerSetupdDataRef.get().getMaxNumWorkers()); return false; }//from ww w. j a v a 2 s . c o m List<String> workerNodeIds = autoScalingStrategy .ipToIdLookup(Lists.newArrayList(Iterables.transform(zkWorkers, new Function<ZkWorker, String>() { @Override public String apply(ZkWorker input) { return input.getWorker().getIp(); } }))); currentlyProvisioning.removeAll(workerNodeIds); boolean nothingProvisioning = currentlyProvisioning.isEmpty(); if (nothingProvisioning) { if (hasTaskPendingBeyondThreshold(pendingTasks)) { AutoScalingData provisioned = autoScalingStrategy.provision(); if (provisioned != null) { currentlyProvisioning.addAll(provisioned.getNodeIds()); lastProvisionTime = new DateTime(); scalingStats.addProvisionEvent(provisioned); return true; } } } else { Duration durSinceLastProvision = new Duration(lastProvisionTime, new DateTime()); log.info( "%s still provisioning. Wait for all provisioned nodes to complete before requesting new worker. Current wait time: %s", currentlyProvisioning, durSinceLastProvision); if (durSinceLastProvision.isLongerThan(config.getMaxScalingDuration())) { log.makeAlert("Worker node provisioning taking too long!") .addData("millisSinceLastProvision", durSinceLastProvision.getMillis()) .addData("provisioningCount", currentlyProvisioning.size()).emit(); currentlyProvisioning.clear(); } } return false; }
From source file:com.metamx.druid.merger.coordinator.scaling.SimpleResourceManagementStrategy.java
License:Open Source License
@Override public boolean doTerminate(Collection<TaskRunnerWorkItem> pendingTasks, Collection<ZkWorker> zkWorkers) { Set<String> workerNodeIds = Sets.newHashSet(autoScalingStrategy .ipToIdLookup(Lists.newArrayList(Iterables.transform(zkWorkers, new Function<ZkWorker, String>() { @Override//from w w w. ja va 2 s . c o m public String apply(ZkWorker input) { return input.getWorker().getIp(); } })))); Set<String> stillExisting = Sets.newHashSet(); for (String s : currentlyTerminating) { if (workerNodeIds.contains(s)) { stillExisting.add(s); } } currentlyTerminating.clear(); currentlyTerminating.addAll(stillExisting); boolean nothingTerminating = currentlyTerminating.isEmpty(); if (nothingTerminating) { final int minNumWorkers = workerSetupdDataRef.get().getMinNumWorkers(); if (zkWorkers.size() <= minNumWorkers) { log.info("Only [%d <= %d] nodes in the cluster, not terminating anything.", zkWorkers.size(), minNumWorkers); return false; } List<ZkWorker> thoseLazyWorkers = Lists .newArrayList(FunctionalIterable.create(zkWorkers).filter(new Predicate<ZkWorker>() { @Override public boolean apply(ZkWorker input) { return input.getRunningTasks().isEmpty() && System.currentTimeMillis() - input.getLastCompletedTaskTime().getMillis() >= config .getMaxWorkerIdleTimeMillisBeforeDeletion(); } })); int maxPossibleNodesTerminated = zkWorkers.size() - minNumWorkers; int numNodesToTerminate = Math.min(maxPossibleNodesTerminated, thoseLazyWorkers.size()); if (numNodesToTerminate <= 0) { log.info("Found no nodes to terminate."); return false; } AutoScalingData terminated = autoScalingStrategy.terminate(Lists .transform(thoseLazyWorkers.subList(0, numNodesToTerminate), new Function<ZkWorker, String>() { @Override public String apply(ZkWorker input) { return input.getWorker().getIp(); } })); if (terminated != null) { currentlyTerminating.addAll(terminated.getNodeIds()); lastTerminateTime = new DateTime(); scalingStats.addTerminateEvent(terminated); return true; } } else { Duration durSinceLastTerminate = new Duration(lastTerminateTime, new DateTime()); log.info("%s still terminating. Wait for all nodes to terminate before trying again.", currentlyTerminating); if (durSinceLastTerminate.isLongerThan(config.getMaxScalingDuration())) { log.makeAlert("Worker node termination taking too long!") .addData("millisSinceLastTerminate", durSinceLastTerminate.getMillis()) .addData("terminatingCount", currentlyTerminating.size()).emit(); currentlyTerminating.clear(); } } return false; }
From source file:com.metamx.druid.merger.coordinator.scaling.SimpleResourceManagementStrategy.java
License:Open Source License
private boolean hasTaskPendingBeyondThreshold(Collection<TaskRunnerWorkItem> pendingTasks) { long now = System.currentTimeMillis(); for (TaskRunnerWorkItem pendingTask : pendingTasks) { if (new Duration(pendingTask.getQueueInsertionTime().getMillis(), now) .isEqual(config.getMaxPendingTaskDuration()) || new Duration(pendingTask.getQueueInsertionTime().getMillis(), now) .isLongerThan(config.getMaxPendingTaskDuration())) { return true; }/*from ww w. j a va 2 s .c o m*/ } return false; }
From source file:com.metamx.druid.realtime.firehose.GracefulShutdownFirehose.java
License:Open Source License
public void shutdown() throws IOException { final long truncatedNow = segmentGranularity.truncate(new DateTime()).getMillis(); final long end = segmentGranularity.increment(truncatedNow) + windowMillis; final Duration timeUntilShutdown = new Duration(System.currentTimeMillis(), end); log.info("Shutdown at approx. %s (in %s)", new DateTime(end), timeUntilShutdown); ScheduledExecutors.scheduleWithFixedDelay(scheduledExecutor, timeUntilShutdown, new Callable<ScheduledExecutors.Signal>() { @Override/* ww w . ja v a 2 s . c o m*/ public ScheduledExecutors.Signal call() throws Exception { try { valveOn.set(false); } catch (Exception e) { throw Throwables.propagate(e); } return ScheduledExecutors.Signal.STOP; } }); beginRejectionPolicy = true; }
From source file:com.metamx.druid.realtime.plumber.RealtimePlumberSchool.java
License:Open Source License
@Override public Plumber findPlumber(final Schema schema, final FireDepartmentMetrics metrics) { verifyState();// w w w . j a va2 s . c o m final RejectionPolicy rejectionPolicy = rejectionPolicyFactory.create(windowPeriod); log.info("Creating plumber using rejectionPolicy[%s]", rejectionPolicy); return new Plumber() { private volatile boolean stopped = false; private volatile ExecutorService persistExecutor = null; private volatile ScheduledExecutorService scheduledExecutor = null; private final Map<Long, Sink> sinks = Maps.newConcurrentMap(); private final VersionedIntervalTimeline<String, Sink> sinkTimeline = new VersionedIntervalTimeline<String, Sink>( String.CASE_INSENSITIVE_ORDER); @Override public void startJob() { computeBaseDir(schema).mkdirs(); initializeExecutors(); bootstrapSinksFromDisk(); registerServerViewCallback(); startPersistThread(); } @Override public Sink getSink(long timestamp) { if (!rejectionPolicy.accept(timestamp)) { return null; } final long truncatedTime = segmentGranularity.truncate(timestamp); Sink retVal = sinks.get(truncatedTime); if (retVal == null) { final Interval sinkInterval = new Interval(new DateTime(truncatedTime), segmentGranularity.increment(new DateTime(truncatedTime))); retVal = new Sink(sinkInterval, schema, versioningPolicy.getVersion(sinkInterval)); try { segmentAnnouncer.announceSegment(retVal.getSegment()); sinks.put(truncatedTime, retVal); sinkTimeline.add(retVal.getInterval(), retVal.getVersion(), new SingleElementPartitionChunk<Sink>(retVal)); } catch (IOException e) { log.makeAlert(e, "Failed to announce new segment[%s]", schema.getDataSource()) .addData("interval", retVal.getInterval()).emit(); } } return retVal; } @Override public <T> QueryRunner<T> getQueryRunner(final Query<T> query) { final QueryRunnerFactory<T, Query<T>> factory = conglomerate.findFactory(query); final QueryToolChest<T, Query<T>> toolchest = factory.getToolchest(); final Function<Query<T>, ServiceMetricEvent.Builder> builderFn = new Function<Query<T>, ServiceMetricEvent.Builder>() { @Override public ServiceMetricEvent.Builder apply(@Nullable Query<T> input) { return toolchest.makeMetricBuilder(query); } }; List<TimelineObjectHolder<String, Sink>> querySinks = Lists.newArrayList(); for (Interval interval : query.getIntervals()) { querySinks.addAll(sinkTimeline.lookup(interval)); } return toolchest.mergeResults(factory.mergeRunners(EXEC, FunctionalIterable.create(querySinks) .transform(new Function<TimelineObjectHolder<String, Sink>, QueryRunner<T>>() { @Override public QueryRunner<T> apply(TimelineObjectHolder<String, Sink> holder) { final Sink theSink = holder.getObject().getChunk(0).getObject(); return new SpecificSegmentQueryRunner<T>(new MetricsEmittingQueryRunner<T>(emitter, builderFn, factory.mergeRunners(EXEC, Iterables.transform(theSink, new Function<FireHydrant, QueryRunner<T>>() { @Override public QueryRunner<T> apply(FireHydrant input) { return factory.createRunner(input.getSegment()); } }))), new SpecificSegmentSpec(new SegmentDescriptor(holder.getInterval(), theSink.getSegment().getVersion(), theSink.getSegment().getShardSpec().getPartitionNum()))); } }))); } @Override public void persist(final Runnable commitRunnable) { final List<Pair<FireHydrant, Interval>> indexesToPersist = Lists.newArrayList(); for (Sink sink : sinks.values()) { if (sink.swappable()) { indexesToPersist.add(Pair.of(sink.swap(), sink.getInterval())); } } log.info("Submitting persist runnable for dataSource[%s]", schema.getDataSource()); persistExecutor.execute(new ThreadRenamingRunnable( String.format("%s-incremental-persist", schema.getDataSource())) { @Override public void doRun() { for (Pair<FireHydrant, Interval> pair : indexesToPersist) { metrics.incrementRowOutputCount(persistHydrant(pair.lhs, schema, pair.rhs)); } commitRunnable.run(); } }); } // Submits persist-n-merge task for a Sink to the persistExecutor private void persistAndMerge(final long truncatedTime, final Sink sink) { final String threadName = String.format("%s-%s-persist-n-merge", schema.getDataSource(), new DateTime(truncatedTime)); persistExecutor.execute(new ThreadRenamingRunnable(threadName) { @Override public void doRun() { final Interval interval = sink.getInterval(); for (FireHydrant hydrant : sink) { if (!hydrant.hasSwapped()) { log.info("Hydrant[%s] hasn't swapped yet, swapping. Sink[%s]", hydrant, sink); final int rowCount = persistHydrant(hydrant, schema, interval); metrics.incrementRowOutputCount(rowCount); } } final File mergedTarget = new File(computePersistDir(schema, interval), "merged"); if (mergedTarget.exists()) { log.info("Skipping already-merged sink: %s", sink); return; } File mergedFile = null; try { List<QueryableIndex> indexes = Lists.newArrayList(); for (FireHydrant fireHydrant : sink) { Segment segment = fireHydrant.getSegment(); final QueryableIndex queryableIndex = segment.asQueryableIndex(); log.info("Adding hydrant[%s]", fireHydrant); indexes.add(queryableIndex); } mergedFile = IndexMerger.mergeQueryableIndex(indexes, schema.getAggregators(), mergedTarget); QueryableIndex index = IndexIO.loadIndex(mergedFile); DataSegment segment = dataSegmentPusher.push(mergedFile, sink.getSegment() .withDimensions(Lists.newArrayList(index.getAvailableDimensions()))); segmentPublisher.publishSegment(segment); } catch (IOException e) { log.makeAlert(e, "Failed to persist merged index[%s]", schema.getDataSource()) .addData("interval", interval).emit(); } if (mergedFile != null) { try { if (mergedFile != null) { log.info("Deleting Index File[%s]", mergedFile); FileUtils.deleteDirectory(mergedFile); } } catch (IOException e) { log.warn(e, "Error deleting directory[%s]", mergedFile); } } } }); } @Override public void finishJob() { log.info("Shutting down..."); for (final Map.Entry<Long, Sink> entry : sinks.entrySet()) { persistAndMerge(entry.getKey(), entry.getValue()); } while (!sinks.isEmpty()) { try { log.info("Cannot shut down yet! Sinks remaining: %s", Joiner.on(", ") .join(Iterables.transform(sinks.values(), new Function<Sink, String>() { @Override public String apply(Sink input) { return input.getSegment().getIdentifier(); } }))); synchronized (handoffCondition) { while (!sinks.isEmpty()) { handoffCondition.wait(); } } } catch (InterruptedException e) { throw Throwables.propagate(e); } } // scheduledExecutor is shutdown here, but persistExecutor is shutdown when the // ServerView sends it a new segment callback if (scheduledExecutor != null) { scheduledExecutor.shutdown(); } stopped = true; } private void initializeExecutors() { if (persistExecutor == null) { persistExecutor = Executors.newFixedThreadPool(1, new ThreadFactoryBuilder().setDaemon(true).setNameFormat("plumber_persist_%d").build()); } if (scheduledExecutor == null) { scheduledExecutor = Executors.newScheduledThreadPool(1, new ThreadFactoryBuilder() .setDaemon(true).setNameFormat("plumber_scheduled_%d").build()); } } private void bootstrapSinksFromDisk() { for (File sinkDir : computeBaseDir(schema).listFiles()) { Interval sinkInterval = new Interval(sinkDir.getName().replace("_", "/")); //final File[] sinkFiles = sinkDir.listFiles(); // To avoid reading and listing of "merged" dir final File[] sinkFiles = sinkDir.listFiles(new FilenameFilter() { @Override public boolean accept(File dir, String fileName) { return !(Ints.tryParse(fileName) == null); } }); Arrays.sort(sinkFiles, new Comparator<File>() { @Override public int compare(File o1, File o2) { try { return Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName())); } catch (NumberFormatException e) { log.error(e, "Couldn't compare as numbers? [%s][%s]", o1, o2); return o1.compareTo(o2); } } }); try { List<FireHydrant> hydrants = Lists.newArrayList(); for (File segmentDir : sinkFiles) { log.info("Loading previously persisted segment at [%s]", segmentDir); // Although this has been tackled at start of this method. // Just a doubly-check added to skip "merged" dir. from being added to hydrants // If 100% sure that this is not needed, this check can be removed. if (Ints.tryParse(segmentDir.getName()) == null) { continue; } hydrants.add( new FireHydrant(new QueryableIndexSegment(null, IndexIO.loadIndex(segmentDir)), Integer.parseInt(segmentDir.getName()))); } Sink currSink = new Sink(sinkInterval, schema, versioningPolicy.getVersion(sinkInterval), hydrants); sinks.put(sinkInterval.getStartMillis(), currSink); sinkTimeline.add(currSink.getInterval(), currSink.getVersion(), new SingleElementPartitionChunk<Sink>(currSink)); segmentAnnouncer.announceSegment(currSink.getSegment()); } catch (IOException e) { log.makeAlert(e, "Problem loading sink[%s] from disk.", schema.getDataSource()) .addData("interval", sinkInterval).emit(); } } } private void registerServerViewCallback() { serverView.registerSegmentCallback(persistExecutor, new ServerView.BaseSegmentCallback() { @Override public ServerView.CallbackAction segmentAdded(DruidServer server, DataSegment segment) { if (stopped) { log.info("Unregistering ServerViewCallback"); persistExecutor.shutdown(); return ServerView.CallbackAction.UNREGISTER; } if ("realtime".equals(server.getType())) { return ServerView.CallbackAction.CONTINUE; } log.debug("Checking segment[%s] on server[%s]", segment, server); if (schema.getDataSource().equals(segment.getDataSource())) { final Interval interval = segment.getInterval(); for (Map.Entry<Long, Sink> entry : sinks.entrySet()) { final Long sinkKey = entry.getKey(); if (interval.contains(sinkKey)) { final Sink sink = entry.getValue(); log.info("Segment[%s] matches sink[%s] on server[%s]", segment, sink, server); final String segmentVersion = segment.getVersion(); final String sinkVersion = sink.getSegment().getVersion(); if (segmentVersion.compareTo(sinkVersion) >= 0) { log.info("Segment version[%s] >= sink version[%s]", segmentVersion, sinkVersion); try { segmentAnnouncer.unannounceSegment(sink.getSegment()); FileUtils .deleteDirectory(computePersistDir(schema, sink.getInterval())); log.info("Removing sinkKey %d for segment %s", sinkKey, sink.getSegment().getIdentifier()); sinks.remove(sinkKey); sinkTimeline.remove(sink.getInterval(), sink.getVersion(), new SingleElementPartitionChunk<Sink>(sink)); synchronized (handoffCondition) { handoffCondition.notifyAll(); } } catch (IOException e) { log.makeAlert(e, "Unable to delete old segment for dataSource[%s].", schema.getDataSource()).addData("interval", sink.getInterval()) .emit(); } } } } } return ServerView.CallbackAction.CONTINUE; } }); } private void startPersistThread() { final long truncatedNow = segmentGranularity.truncate(new DateTime()).getMillis(); final long windowMillis = windowPeriod.toStandardDuration().getMillis(); log.info("Expect to run at [%s]", new DateTime().plus(new Duration(System.currentTimeMillis(), segmentGranularity.increment(truncatedNow) + windowMillis))); ScheduledExecutors.scheduleAtFixedRate(scheduledExecutor, new Duration(System.currentTimeMillis(), segmentGranularity.increment(truncatedNow) + windowMillis), new Duration(truncatedNow, segmentGranularity.increment(truncatedNow)), new ThreadRenamingCallable<ScheduledExecutors.Signal>(String.format("%s-overseer-%d", schema.getDataSource(), schema.getShardSpec().getPartitionNum())) { @Override public ScheduledExecutors.Signal doCall() { if (stopped) { log.info("Stopping merge-n-push overseer thread"); return ScheduledExecutors.Signal.STOP; } log.info("Starting merge and push."); long minTimestamp = segmentGranularity .truncate(rejectionPolicy.getCurrMaxTime().minus(windowMillis)).getMillis(); List<Map.Entry<Long, Sink>> sinksToPush = Lists.newArrayList(); for (Map.Entry<Long, Sink> entry : sinks.entrySet()) { final Long intervalStart = entry.getKey(); if (intervalStart < minTimestamp) { log.info("Adding entry[%s] for merge and push.", entry); sinksToPush.add(entry); } } for (final Map.Entry<Long, Sink> entry : sinksToPush) { persistAndMerge(entry.getKey(), entry.getValue()); } if (stopped) { log.info("Stopping merge-n-push overseer thread"); return ScheduledExecutors.Signal.STOP; } else { return ScheduledExecutors.Signal.REPEAT; } } }); } }; }
From source file:com.metamx.druid.realtime.RealtimePlumberSchool.java
License:Open Source License
@Override public Plumber findPlumber(final Schema schema, final FireDepartmentMetrics metrics) { verifyState();// w w w . jav a2 s . c o m initializeExecutors(); computeBaseDir(schema).mkdirs(); final Map<Long, Sink> sinks = Maps.newConcurrentMap(); for (File sinkDir : computeBaseDir(schema).listFiles()) { Interval sinkInterval = new Interval(sinkDir.getName().replace("_", "/")); final File[] sinkFiles = sinkDir.listFiles(); Arrays.sort(sinkFiles, new Comparator<File>() { @Override public int compare(File o1, File o2) { try { return Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName())); } catch (NumberFormatException e) { log.error(e, "Couldn't compare as numbers? [%s][%s]", o1, o2); return o1.compareTo(o2); } } }); try { List<FireHydrant> hydrants = Lists.newArrayList(); for (File segmentDir : sinkFiles) { log.info("Loading previously persisted segment at [%s]", segmentDir); hydrants.add(new FireHydrant(new QueryableIndexSegment(null, IndexIO.loadIndex(segmentDir)), Integer.parseInt(segmentDir.getName()))); } Sink currSink = new Sink(sinkInterval, schema, hydrants); sinks.put(sinkInterval.getStartMillis(), currSink); metadataUpdater.announceSegment(currSink.getSegment()); } catch (IOException e) { log.makeAlert(e, "Problem loading sink[%s] from disk.", schema.getDataSource()) .addData("interval", sinkInterval).emit(); } } serverView.registerSegmentCallback(persistExecutor, new ServerView.BaseSegmentCallback() { @Override public ServerView.CallbackAction segmentAdded(DruidServer server, DataSegment segment) { if ("realtime".equals(server.getType())) { return ServerView.CallbackAction.CONTINUE; } log.debug("Checking segment[%s] on server[%s]", segment, server); if (schema.getDataSource().equals(segment.getDataSource())) { final Interval interval = segment.getInterval(); for (Map.Entry<Long, Sink> entry : sinks.entrySet()) { final Long sinkKey = entry.getKey(); if (interval.contains(sinkKey)) { final Sink sink = entry.getValue(); log.info("Segment matches sink[%s]", sink); if (segment.getVersion().compareTo(sink.getSegment().getVersion()) >= 0) { try { metadataUpdater.unannounceSegment(sink.getSegment()); FileUtils.deleteDirectory(computePersistDir(schema, sink.getInterval())); sinks.remove(sinkKey); } catch (IOException e) { log.makeAlert(e, "Unable to delete old segment for dataSource[%s].", schema.getDataSource()).addData("interval", sink.getInterval()).emit(); } } } } } return ServerView.CallbackAction.CONTINUE; } }); final long truncatedNow = segmentGranularity.truncate(new DateTime()).getMillis(); final long windowMillis = windowPeriod.toStandardDuration().getMillis(); final RejectionPolicy rejectionPolicy = rejectionPolicyFactory.create(windowPeriod); log.info("Creating plumber using rejectionPolicy[%s]", rejectionPolicy); log.info("Expect to run at [%s]", new DateTime().plus(new Duration(System.currentTimeMillis(), segmentGranularity.increment(truncatedNow) + windowMillis))); ScheduledExecutors.scheduleAtFixedRate(scheduledExecutor, new Duration(System.currentTimeMillis(), segmentGranularity.increment(truncatedNow) + windowMillis), new Duration(truncatedNow, segmentGranularity.increment(truncatedNow)), new ThreadRenamingRunnable(String.format("%s-overseer", schema.getDataSource())) { @Override public void doRun() { log.info("Starting merge and push."); long minTimestamp = segmentGranularity.truncate(rejectionPolicy.getCurrMaxTime()) .getMillis() - windowMillis; List<Map.Entry<Long, Sink>> sinksToPush = Lists.newArrayList(); for (Map.Entry<Long, Sink> entry : sinks.entrySet()) { final Long intervalStart = entry.getKey(); if (intervalStart < minTimestamp) { log.info("Adding entry[%s] for merge and push.", entry); sinksToPush.add(entry); } } for (final Map.Entry<Long, Sink> entry : sinksToPush) { final Sink sink = entry.getValue(); final String threadName = String.format("%s-%s-persist-n-merge", schema.getDataSource(), new DateTime(entry.getKey())); persistExecutor.execute(new ThreadRenamingRunnable(threadName) { @Override public void doRun() { final Interval interval = sink.getInterval(); for (FireHydrant hydrant : sink) { if (!hydrant.hasSwapped()) { log.info("Hydrant[%s] hasn't swapped yet, swapping. Sink[%s]", hydrant, sink); final int rowCount = persistHydrant(hydrant, schema, interval); metrics.incrementRowOutputCount(rowCount); } } final File mergedFile; try { List<QueryableIndex> indexes = Lists.newArrayList(); for (FireHydrant fireHydrant : sink) { Segment segment = fireHydrant.getSegment(); final QueryableIndex queryableIndex = segment.asQueryableIndex(); log.info("Adding hydrant[%s]", fireHydrant); indexes.add(queryableIndex); } mergedFile = IndexMerger.mergeQueryableIndex(indexes, schema.getAggregators(), new File(computePersistDir(schema, interval), "merged")); QueryableIndex index = IndexIO.loadIndex(mergedFile); DataSegment segment = segmentPusher.push(mergedFile, sink.getSegment().withDimensions( Lists.newArrayList(index.getAvailableDimensions()))); metadataUpdater.publishSegment(segment); } catch (IOException e) { log.makeAlert(e, "Failed to persist merged index[%s]", schema.getDataSource()).addData("interval", interval).emit(); } } }); } } }); return new Plumber() { @Override public Sink getSink(long timestamp) { if (!rejectionPolicy.accept(timestamp)) { return null; } final long truncatedTime = segmentGranularity.truncate(timestamp); Sink retVal = sinks.get(truncatedTime); if (retVal == null) { retVal = new Sink(new Interval(new DateTime(truncatedTime), segmentGranularity.increment(new DateTime(truncatedTime))), schema); try { metadataUpdater.announceSegment(retVal.getSegment()); sinks.put(truncatedTime, retVal); } catch (IOException e) { log.makeAlert(e, "Failed to announce new segment[%s]", schema.getDataSource()) .addData("interval", retVal.getInterval()).emit(); } } return retVal; } @Override public <T> QueryRunner<T> getQueryRunner(final Query<T> query) { final QueryRunnerFactory<T, Query<T>> factory = conglomerate.findFactory(query); final Function<Query<T>, ServiceMetricEvent.Builder> builderFn = new Function<Query<T>, ServiceMetricEvent.Builder>() { private final QueryToolChest<T, Query<T>> toolchest = factory.getToolchest(); @Override public ServiceMetricEvent.Builder apply(@Nullable Query<T> input) { return toolchest.makeMetricBuilder(query); } }; return factory.mergeRunners(EXEC, FunctionalIterable.create(sinks.values()).transform(new Function<Sink, QueryRunner<T>>() { @Override public QueryRunner<T> apply(@Nullable Sink input) { return new MetricsEmittingQueryRunner<T>(emitter, builderFn, factory.mergeRunners( EXEC, Iterables.transform(input, new Function<FireHydrant, QueryRunner<T>>() { @Override public QueryRunner<T> apply(@Nullable FireHydrant input) { return factory.createRunner(input.getSegment()); } }))); } })); } @Override public void persist(final Runnable commitRunnable) { final List<Pair<FireHydrant, Interval>> indexesToPersist = Lists.newArrayList(); for (Sink sink : sinks.values()) { if (sink.swappable()) { indexesToPersist.add(Pair.of(sink.swap(), sink.getInterval())); } } log.info("Submitting persist runnable for dataSource[%s]", schema.getDataSource()); persistExecutor.execute(new ThreadRenamingRunnable( String.format("%s-incremental-persist", schema.getDataSource())) { @Override public void doRun() { for (Pair<FireHydrant, Interval> pair : indexesToPersist) { metrics.incrementRowOutputCount(persistHydrant(pair.lhs, schema, pair.rhs)); } commitRunnable.run(); } }); } @Override public void finishJob() { throw new UnsupportedOperationException(); } }; }