Examples with Function com.google.common.base.Function used on opensource projects

Example 21 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class KafkaSupervisor method discoverTasks.

private void discoverTasks() throws ExecutionException, InterruptedException, TimeoutException {
    int taskCount = 0;
    List<String> futureTaskIds = Lists.newArrayList();
    List<ListenableFuture<Boolean>> futures = Lists.newArrayList();
    List<Task> tasks = taskStorage.getActiveTasks();
    for (Task task : tasks) {
        if (!(task instanceof KafkaIndexTask) || !dataSource.equals(task.getDataSource())) {
            continue;
        }
        taskCount++;
        final KafkaIndexTask kafkaTask = (KafkaIndexTask) task;
        final String taskId = task.getId();
        // Determine which task group this task belongs to based on one of the partitions handled by this task. If we
        // later determine that this task is actively reading, we will make sure that it matches our current partition
        // allocation (getTaskGroupIdForPartition(partition) should return the same value for every partition being read
        // by this task) and kill it if it is not compatible. If the task is instead found to be in the publishing
        // state, we will permit it to complete even if it doesn't match our current partition allocation to support
        // seamless schema migration.
        Iterator<Integer> it = kafkaTask.getIOConfig().getStartPartitions().getPartitionOffsetMap().keySet().iterator();
        final Integer taskGroupId = (it.hasNext() ? getTaskGroupIdForPartition(it.next()) : null);
        if (taskGroupId != null) {
            // check to see if we already know about this task, either in [taskGroups] or in [pendingCompletionTaskGroups]
            // and if not add it to taskGroups or pendingCompletionTaskGroups (if status = PUBLISHING)
            TaskGroup taskGroup = taskGroups.get(taskGroupId);
            if (!isTaskInPendingCompletionGroups(taskId) && (taskGroup == null || !taskGroup.tasks.containsKey(taskId))) {
                futureTaskIds.add(taskId);
                futures.add(Futures.transform(taskClient.getStatusAsync(taskId), new Function<KafkaIndexTask.Status, Boolean>() {

                    @Override
                    public Boolean apply(KafkaIndexTask.Status status) {
                        if (status == KafkaIndexTask.Status.PUBLISHING) {
                            addDiscoveredTaskToPendingCompletionTaskGroups(taskGroupId, taskId, kafkaTask.getIOConfig().getStartPartitions().getPartitionOffsetMap());
                            // update partitionGroups with the publishing task's offsets (if they are greater than what is
                            // existing) so that the next tasks will start reading from where this task left off
                            Map<Integer, Long> publishingTaskCurrentOffsets = taskClient.getCurrentOffsets(taskId, true);
                            for (Map.Entry<Integer, Long> entry : publishingTaskCurrentOffsets.entrySet()) {
                                Integer partition = entry.getKey();
                                Long offset = entry.getValue();
                                ConcurrentHashMap<Integer, Long> partitionOffsets = partitionGroups.get(getTaskGroupIdForPartition(partition));
                                boolean succeeded;
                                do {
                                    succeeded = true;
                                    Long previousOffset = partitionOffsets.putIfAbsent(partition, offset);
                                    if (previousOffset != null && previousOffset < offset) {
                                        succeeded = partitionOffsets.replace(partition, previousOffset, offset);
                                    }
                                } while (!succeeded);
                            }
                        } else {
                            for (Integer partition : kafkaTask.getIOConfig().getStartPartitions().getPartitionOffsetMap().keySet()) {
                                if (!taskGroupId.equals(getTaskGroupIdForPartition(partition))) {
                                    log.warn("Stopping task [%s] which does not match the expected partition allocation", taskId);
                                    try {
                                        stopTask(taskId, false).get(futureTimeoutInSeconds, TimeUnit.SECONDS);
                                    } catch (InterruptedException | ExecutionException | TimeoutException e) {
                                        log.warn(e, "Exception while stopping task");
                                    }
                                    return false;
                                }
                            }
                            if (taskGroups.putIfAbsent(taskGroupId, new TaskGroup(ImmutableMap.copyOf(kafkaTask.getIOConfig().getStartPartitions().getPartitionOffsetMap()), kafkaTask.getIOConfig().getMinimumMessageTime())) == null) {
                                log.debug("Created new task group [%d]", taskGroupId);
                            }
                            if (!isTaskCurrent(taskGroupId, taskId)) {
                                log.info("Stopping task [%s] which does not match the expected parameters and ingestion spec", taskId);
                                try {
                                    stopTask(taskId, false).get(futureTimeoutInSeconds, TimeUnit.SECONDS);
                                } catch (InterruptedException | ExecutionException | TimeoutException e) {
                                    log.warn(e, "Exception while stopping task");
                                }
                                return false;
                            } else {
                                taskGroups.get(taskGroupId).tasks.putIfAbsent(taskId, new TaskData());
                            }
                        }
                        return true;
                    }
                }, workerExec));
            }
        }
    }
    List<Boolean> results = Futures.successfulAsList(futures).get(futureTimeoutInSeconds, TimeUnit.SECONDS);
    for (int i = 0; i < results.size(); i++) {
        if (results.get(i) == null) {
            String taskId = futureTaskIds.get(i);
            log.warn("Task [%s] failed to return status, killing task", taskId);
            killTask(taskId);
        }
    }
    log.debug("Found [%d] Kafka indexing tasks for dataSource [%s]", taskCount, dataSource);
}

Example 22 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class IngestSegmentFirehoseFactory method connect.

@Override
public Firehose connect(InputRowParser inputRowParser) throws IOException, ParseException {
    log.info("Connecting firehose: dataSource[%s], interval[%s]", dataSource, interval);
    if (taskToolbox == null) {
        // Noop Task is just used to create the toolbox and list segments.
        taskToolbox = injector.getInstance(TaskToolboxFactory.class).build(new NoopTask("reingest", 0, 0, null, null, null));
    }
    try {
        final List<DataSegment> usedSegments = taskToolbox.getTaskActionClient().submit(new SegmentListUsedAction(dataSource, interval, null));
        final Map<DataSegment, File> segmentFileMap = taskToolbox.fetchSegments(usedSegments);
        VersionedIntervalTimeline<String, DataSegment> timeline = new VersionedIntervalTimeline<>(Ordering.<String>natural().nullsFirst());
        for (DataSegment segment : usedSegments) {
            timeline.add(segment.getInterval(), segment.getVersion(), segment.getShardSpec().createChunk(segment));
        }
        final List<TimelineObjectHolder<String, DataSegment>> timeLineSegments = timeline.lookup(interval);
        final List<String> dims;
        if (dimensions != null) {
            dims = dimensions;
        } else if (inputRowParser.getParseSpec().getDimensionsSpec().hasCustomDimensions()) {
            dims = inputRowParser.getParseSpec().getDimensionsSpec().getDimensionNames();
        } else {
            Set<String> dimSet = Sets.newHashSet(Iterables.concat(Iterables.transform(timeLineSegments, new Function<TimelineObjectHolder<String, DataSegment>, Iterable<String>>() {

                @Override
                public Iterable<String> apply(TimelineObjectHolder<String, DataSegment> timelineObjectHolder) {
                    return Iterables.concat(Iterables.transform(timelineObjectHolder.getObject(), new Function<PartitionChunk<DataSegment>, Iterable<String>>() {

                        @Override
                        public Iterable<String> apply(PartitionChunk<DataSegment> input) {
                            return input.getObject().getDimensions();
                        }
                    }));
                }
            })));
            dims = Lists.newArrayList(Sets.difference(dimSet, inputRowParser.getParseSpec().getDimensionsSpec().getDimensionExclusions()));
        }
        final List<String> metricsList;
        if (metrics != null) {
            metricsList = metrics;
        } else {
            Set<String> metricsSet = Sets.newHashSet(Iterables.concat(Iterables.transform(timeLineSegments, new Function<TimelineObjectHolder<String, DataSegment>, Iterable<String>>() {

                @Override
                public Iterable<String> apply(TimelineObjectHolder<String, DataSegment> input) {
                    return Iterables.concat(Iterables.transform(input.getObject(), new Function<PartitionChunk<DataSegment>, Iterable<String>>() {

                        @Override
                        public Iterable<String> apply(PartitionChunk<DataSegment> input) {
                            return input.getObject().getMetrics();
                        }
                    }));
                }
            })));
            metricsList = Lists.newArrayList(metricsSet);
        }
        final List<WindowedStorageAdapter> adapters = Lists.newArrayList(Iterables.concat(Iterables.transform(timeLineSegments, new Function<TimelineObjectHolder<String, DataSegment>, Iterable<WindowedStorageAdapter>>() {

            @Override
            public Iterable<WindowedStorageAdapter> apply(final TimelineObjectHolder<String, DataSegment> holder) {
                return Iterables.transform(holder.getObject(), new Function<PartitionChunk<DataSegment>, WindowedStorageAdapter>() {

                    @Override
                    public WindowedStorageAdapter apply(final PartitionChunk<DataSegment> input) {
                        final DataSegment segment = input.getObject();
                        try {
                            return new WindowedStorageAdapter(new QueryableIndexStorageAdapter(indexIO.loadIndex(Preconditions.checkNotNull(segmentFileMap.get(segment), "File for segment %s", segment.getIdentifier()))), holder.getInterval());
                        } catch (IOException e) {
                            throw Throwables.propagate(e);
                        }
                    }
                });
            }
        })));
        return new IngestSegmentFirehose(adapters, dims, metricsList, dimFilter, Granularities.NONE);
    } catch (IOException e) {
        throw Throwables.propagate(e);
    } catch (SegmentLoadingException e) {
        throw Throwables.propagate(e);
    }
}

Example 23 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class TaskLockbox method findLockPossesForInterval.

/**
   * Return all locks that overlap some search interval.
   */
private List<TaskLockPosse> findLockPossesForInterval(final String dataSource, final Interval interval) {
    giant.lock();
    try {
        final NavigableMap<Interval, TaskLockPosse> dsRunning = running.get(dataSource);
        if (dsRunning == null) {
            // No locks at all
            return Collections.emptyList();
        } else {
            // Tasks are indexed by locked interval, which are sorted by interval start. Intervals are non-overlapping, so:
            final NavigableSet<Interval> dsLockbox = dsRunning.navigableKeySet();
            final Iterable<Interval> searchIntervals = Iterables.concat(// Single interval that starts at or before ours
            Collections.singletonList(dsLockbox.floor(new Interval(interval.getStart(), new DateTime(JodaUtils.MAX_INSTANT)))), // All intervals that start somewhere between our start instant (exclusive) and end instant (exclusive)
            dsLockbox.subSet(new Interval(interval.getStart(), new DateTime(JodaUtils.MAX_INSTANT)), false, new Interval(interval.getEnd(), interval.getEnd()), false));
            return Lists.newArrayList(FunctionalIterable.create(searchIntervals).filter(new Predicate<Interval>() {

                @Override
                public boolean apply(@Nullable Interval searchInterval) {
                    return searchInterval != null && searchInterval.overlaps(interval);
                }
            }).transform(new Function<Interval, TaskLockPosse>() {

                @Override
                public TaskLockPosse apply(Interval interval) {
                    return dsRunning.get(interval);
                }
            }));
        }
    } finally {
        giant.unlock();
    }
}

Example 24 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class PendingTaskBasedWorkerResourceManagementStrategy method doTerminate.

@Override
public boolean doTerminate(WorkerTaskRunner runner) {
    Collection<ImmutableWorkerInfo> zkWorkers = runner.getWorkers();
    synchronized (lock) {
        final WorkerBehaviorConfig workerConfig = workerConfigRef.get();
        if (workerConfig == null) {
            log.warn("No workerConfig available, cannot terminate workers.");
            return false;
        }
        if (!currentlyProvisioning.isEmpty()) {
            log.debug("Already provisioning nodes, Not Terminating any nodes.");
            return false;
        }
        boolean didTerminate = false;
        final Collection<String> workerNodeIds = getWorkerNodeIDs(runner.getLazyWorkers(), workerConfig);
        final Set<String> stillExisting = Sets.newHashSet();
        for (String s : currentlyTerminating) {
            if (workerNodeIds.contains(s)) {
                stillExisting.add(s);
            }
        }
        currentlyTerminating.clear();
        currentlyTerminating.addAll(stillExisting);
        if (currentlyTerminating.isEmpty()) {
            final int maxWorkersToTerminate = maxWorkersToTerminate(zkWorkers, workerConfig);
            final Predicate<ImmutableWorkerInfo> isLazyWorker = ResourceManagementUtil.createLazyWorkerPredicate(config);
            final List<String> laziestWorkerIps = Lists.newArrayList(Collections2.transform(runner.markWorkersLazy(isLazyWorker, maxWorkersToTerminate), new Function<Worker, String>() {

                @Override
                public String apply(Worker zkWorker) {
                    return zkWorker.getIp();
                }
            }));
            if (laziestWorkerIps.isEmpty()) {
                log.debug("Found no lazy workers");
            } else {
                log.info("Terminating %,d lazy workers: %s", laziestWorkerIps.size(), Joiner.on(", ").join(laziestWorkerIps));
                final AutoScalingData terminated = workerConfig.getAutoScaler().terminate(laziestWorkerIps);
                if (terminated != null) {
                    currentlyTerminating.addAll(terminated.getNodeIds());
                    lastTerminateTime = new DateTime();
                    scalingStats.addTerminateEvent(terminated);
                    didTerminate = true;
                }
            }
        } else {
            Duration durSinceLastTerminate = new Duration(lastTerminateTime, new DateTime());
            log.info("%s terminating. Current wait time: %s", currentlyTerminating, durSinceLastTerminate);
            if (durSinceLastTerminate.isLongerThan(config.getMaxScalingDuration().toStandardDuration())) {
                log.makeAlert("Worker node termination taking too long!").addData("millisSinceLastTerminate", durSinceLastTerminate.getMillis()).addData("terminatingCount", currentlyTerminating.size()).emit();
                currentlyTerminating.clear();
            }
        }
        return didTerminate;
    }
}

Example 25 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class EC2AutoScaler method terminate.

@Override
public AutoScalingData terminate(List<String> ips) {
    if (ips.isEmpty()) {
        return new AutoScalingData(Lists.<String>newArrayList());
    }
    DescribeInstancesResult result = amazonEC2Client.describeInstances(new DescribeInstancesRequest().withFilters(new Filter("private-ip-address", ips)));
    List<Instance> instances = Lists.newArrayList();
    for (Reservation reservation : result.getReservations()) {
        instances.addAll(reservation.getInstances());
    }
    try {
        return terminateWithIds(Lists.transform(instances, new Function<Instance, String>() {

            @Override
            public String apply(Instance input) {
                return input.getInstanceId();
            }
        }));
    } catch (Exception e) {
        log.error(e, "Unable to terminate any instances.");
    }
    return null;
}