Examples with IdealState org.apache.helix.model.IdealState used on opensource projects

Example 26 with IdealState

use of org.apache.helix.model.IdealState in project pinot by linkedin.

the class PinotHelixResourceManagerTest method testRebuildBrokerResourceFromHelixTags.

@Test
public void testRebuildBrokerResourceFromHelixTags() throws Exception {
    AbstractTableConfig tableConfig = AbstractTableConfig.init(ControllerRequestBuilderUtil.buildCreateOfflineTableJSON("faketable", "serverTenant", "brokerTenant", 3).toString());
    Tenant tenant = new Tenant();
    tenant.setTenantName("brokerTenant");
    tenant.setTenantRole("BROKER");
    tenant.setNumberOfInstances(3);
    pinotHelixResourceManager.createBrokerTenant(tenant);
    pinotHelixResourceManager.addTable(tableConfig);
    // Check that the broker ideal state has 3 brokers assigned to it for faketable_OFFLINE
    IdealState idealState = pinotHelixResourceManager.getHelixAdmin().getResourceIdealState(HELIX_CLUSTER_NAME, CommonConstants.Helix.BROKER_RESOURCE_INSTANCE);
    Assert.assertEquals(idealState.getInstanceStateMap("faketable_OFFLINE").size(), 3);
    // Retag all instances current assigned to brokerTenant to be unassigned
    Set<String> brokerInstances = pinotHelixResourceManager.getAllInstancesForBrokerTenant("brokerTenant");
    for (String brokerInstance : brokerInstances) {
        pinotHelixResourceManager.getHelixAdmin().removeInstanceTag(HELIX_CLUSTER_NAME, brokerInstance, "brokerTenant_BROKER");
        pinotHelixResourceManager.getHelixAdmin().addInstanceTag(HELIX_CLUSTER_NAME, brokerInstance, CommonConstants.Helix.UNTAGGED_BROKER_INSTANCE);
    }
    // Rebuilding the broker tenant should update the ideal state size
    pinotHelixResourceManager.rebuildBrokerResourceFromHelixTags("faketable_OFFLINE");
    idealState = pinotHelixResourceManager.getHelixAdmin().getResourceIdealState(HELIX_CLUSTER_NAME, CommonConstants.Helix.BROKER_RESOURCE_INSTANCE);
    Assert.assertEquals(idealState.getInstanceStateMap("faketable_OFFLINE").size(), 0);
    // Tag five instances
    int instancesRemainingToTag = 5;
    List<String> instances = pinotHelixResourceManager.getAllInstanceNames();
    for (String instance : instances) {
        if (instance.startsWith(CommonConstants.Helix.PREFIX_OF_BROKER_INSTANCE)) {
            pinotHelixResourceManager.getHelixAdmin().removeInstanceTag(HELIX_CLUSTER_NAME, instance, CommonConstants.Helix.UNTAGGED_BROKER_INSTANCE);
            pinotHelixResourceManager.getHelixAdmin().addInstanceTag(HELIX_CLUSTER_NAME, instance, "brokerTenant_BROKER");
            instancesRemainingToTag--;
            if (instancesRemainingToTag == 0) {
                break;
            }
        }
    }
    // Rebuilding the broker tenant should update the ideal state size
    pinotHelixResourceManager.rebuildBrokerResourceFromHelixTags("faketable_OFFLINE");
    idealState = pinotHelixResourceManager.getHelixAdmin().getResourceIdealState(HELIX_CLUSTER_NAME, CommonConstants.Helix.BROKER_RESOURCE_INSTANCE);
    Assert.assertEquals(idealState.getInstanceStateMap("faketable_OFFLINE").size(), 5);
    // Untag all instances for other tests
    for (String instance : instances) {
        if (instance.startsWith(CommonConstants.Helix.PREFIX_OF_BROKER_INSTANCE)) {
            pinotHelixResourceManager.getHelixAdmin().removeInstanceTag(HELIX_CLUSTER_NAME, instance, "brokerTenant_BROKER");
            pinotHelixResourceManager.getHelixAdmin().addInstanceTag(HELIX_CLUSTER_NAME, instance, CommonConstants.Helix.UNTAGGED_BROKER_INSTANCE);
        }
    }
    // Delete table
    pinotHelixResourceManager.deleteOfflineTable("faketable");
}

Example 27 with IdealState

use of org.apache.helix.model.IdealState in project pinot by linkedin.

the class PinotLLCRealtimeSegmentManagerTest method testAutoReplaceConsumingSegment.

public void testAutoReplaceConsumingSegment(final String tableConfigStartOffset) throws Exception {
    FakePinotLLCRealtimeSegmentManager segmentManager = new FakePinotLLCRealtimeSegmentManager(true, null);
    final int nPartitions = 8;
    final int nInstances = 3;
    final int nReplicas = 2;
    final String topic = "someTopic";
    final String rtTableName = "table_REALTIME";
    List<String> instances = getInstanceList(nInstances);
    final String startOffset = KAFKA_OFFSET;
    IdealState idealState = PinotTableIdealStateBuilder.buildEmptyKafkaConsumerRealtimeIdealStateFor(rtTableName, nReplicas);
    segmentManager.setupHelixEntries(topic, rtTableName, nPartitions, instances, nReplicas, startOffset, DUMMY_HOST, idealState, false, 10000);
    // Add another segment for each partition
    long now = System.currentTimeMillis();
    List<String> existingSegments = new ArrayList<>(segmentManager._idealStateEntries.keySet());
    final int partitionToBeFixed = 3;
    final int partitionWithHigherOffset = 4;
    final int emptyPartition = 5;
    final long smallestPartitionOffset = 0x259080984568L;
    final long largestPartitionOffset = smallestPartitionOffset + 100000;
    final long higherOffset = smallestPartitionOffset + 100;
    for (String segmentNameStr : existingSegments) {
        LLCSegmentName segmentName = new LLCSegmentName(segmentNameStr);
        switch(segmentName.getPartitionId()) {
            case partitionToBeFixed:
                // Do nothing, we will test adding a new segment for this partition when there is only one segment in there.
                break;
            case emptyPartition:
                // Remove existing segment, so we can test adding a new segment for this partition when none exists
                segmentManager._idealStateEntries.remove(segmentNameStr);
                break;
            case partitionWithHigherOffset:
                // Set segment metadata for this segment such that its offset is higher than startOffset we get from kafka.
                // In that case, we should choose the new segment offset as this one rather than the one kafka hands us.
                LLCRealtimeSegmentZKMetadata metadata = new LLCRealtimeSegmentZKMetadata();
                metadata.setSegmentName(segmentName.getSegmentName());
                metadata.setEndOffset(higherOffset);
                metadata.setStatus(CommonConstants.Segment.Realtime.Status.DONE);
                segmentManager._metadataMap.put(segmentName.getSegmentName(), metadata);
                break;
            default:
                // Add a second segment for this partition. It will not be repaired.
                LLCSegmentName newSegmentName = new LLCSegmentName(segmentName.getTableName(), segmentName.getPartitionId(), segmentName.getSequenceNumber() + 1, now);
                List<String> hosts = segmentManager._idealStateEntries.get(segmentNameStr);
                segmentManager._idealStateEntries.put(newSegmentName.getSegmentName(), hosts);
                break;
        }
    }
    Map<String, String> streamPropMap = new HashMap<>(1);
    streamPropMap.put(StringUtil.join(".", CommonConstants.Helix.DataSource.STREAM_PREFIX, CommonConstants.Helix.DataSource.Realtime.Kafka.CONSUMER_TYPE), "simple");
    streamPropMap.put(StringUtil.join(".", CommonConstants.Helix.DataSource.STREAM_PREFIX, CommonConstants.Helix.DataSource.Realtime.Kafka.KAFKA_CONSUMER_PROPS_PREFIX, CommonConstants.Helix.DataSource.Realtime.Kafka.AUTO_OFFSET_RESET), tableConfigStartOffset);
    KafkaStreamMetadata kafkaStreamMetadata = new KafkaStreamMetadata(streamPropMap);
    AbstractTableConfig tableConfig = mock(AbstractTableConfig.class);
    IndexingConfig indexingConfig = mock(IndexingConfig.class);
    when(indexingConfig.getStreamConfigs()).thenReturn(streamPropMap);
    when(tableConfig.getIndexingConfig()).thenReturn(indexingConfig);
    Set<Integer> nonConsumingPartitions = new HashSet<>(1);
    nonConsumingPartitions.add(partitionToBeFixed);
    nonConsumingPartitions.add(partitionWithHigherOffset);
    nonConsumingPartitions.add(emptyPartition);
    segmentManager._kafkaSmallestOffsetToReturn = smallestPartitionOffset;
    segmentManager._kafkaLargestOffsetToReturn = largestPartitionOffset;
    existingSegments = new ArrayList<>(segmentManager._idealStateEntries.keySet());
    segmentManager._paths.clear();
    segmentManager._records.clear();
    segmentManager.createConsumingSegment(rtTableName, nonConsumingPartitions, existingSegments, tableConfig);
    Assert.assertEquals(segmentManager._paths.size(), 3);
    Assert.assertEquals(segmentManager._records.size(), 3);
    Assert.assertEquals(segmentManager._oldSegmentNameStr.size(), 3);
    Assert.assertEquals(segmentManager._newSegmentNameStr.size(), 3);
    int found = 0;
    int index = 0;
    while (index < segmentManager._paths.size()) {
        String znodePath = segmentManager._paths.get(index);
        int slash = znodePath.lastIndexOf('/');
        String segmentNameStr = znodePath.substring(slash + 1);
        LLCSegmentName segmentName = new LLCSegmentName(segmentNameStr);
        ZNRecord znRecord;
        LLCRealtimeSegmentZKMetadata metadata;
        switch(segmentName.getPartitionId()) {
            case partitionToBeFixed:
                // We had left this partition with one segment. So, a second one should be created with a sequence number one
                // higher than starting. Its start offset should be what kafka returns.
                found++;
                Assert.assertEquals(segmentName.getSequenceNumber(), PinotLLCRealtimeSegmentManager.STARTING_SEQUENCE_NUMBER + 1);
                znRecord = segmentManager._records.get(index);
                metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
                Assert.assertEquals(metadata.getNumReplicas(), 2);
                Assert.assertEquals(metadata.getStartOffset(), smallestPartitionOffset);
                break;
            case emptyPartition:
                // We had removed any segments in this partition. A new one should be created with the offset as returned
                // by kafka and with the starting sequence number.
                found++;
                Assert.assertEquals(segmentName.getSequenceNumber(), PinotLLCRealtimeSegmentManager.STARTING_SEQUENCE_NUMBER);
                znRecord = segmentManager._records.get(index);
                metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
                Assert.assertEquals(metadata.getNumReplicas(), 2);
                if (tableConfigStartOffset.equals("smallest")) {
                    Assert.assertEquals(metadata.getStartOffset(), smallestPartitionOffset);
                } else {
                    Assert.assertEquals(metadata.getStartOffset(), largestPartitionOffset);
                }
                break;
            case partitionWithHigherOffset:
                // We had left this partition with one segment. In addition, we had the end-offset of the first segment set to
                // a value higher than that returned by kafka. So, a second one should be created with a sequence number one
                // equal to the end offset of the first one.
                found++;
                Assert.assertEquals(segmentName.getSequenceNumber(), PinotLLCRealtimeSegmentManager.STARTING_SEQUENCE_NUMBER + 1);
                znRecord = segmentManager._records.get(index);
                metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
                Assert.assertEquals(metadata.getNumReplicas(), 2);
                Assert.assertEquals(metadata.getStartOffset(), higherOffset);
                break;
        }
        index++;
    }
    // We should see all three cases here.
    Assert.assertEquals(3, found);
    // Now, if we make 'partitionToBeFixed' a non-consuming partition, a second one should get added with the same start offset as
    // as the first one, since the kafka offset to return has not changed.
    Set<Integer> ncPartitions = new HashSet<>(1);
    ncPartitions.add(partitionToBeFixed);
    segmentManager.createConsumingSegment(rtTableName, ncPartitions, segmentManager.getExistingSegments(rtTableName), tableConfig);
    Assert.assertEquals(segmentManager._paths.size(), 4);
    Assert.assertEquals(segmentManager._records.size(), 4);
    Assert.assertEquals(segmentManager._oldSegmentNameStr.size(), 4);
    Assert.assertEquals(segmentManager._newSegmentNameStr.size(), 4);
    // The latest zn record should be that of the new one we added.
    ZNRecord znRecord = segmentManager._records.get(3);
    LLCRealtimeSegmentZKMetadata metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
    Assert.assertEquals(metadata.getNumReplicas(), 2);
    Assert.assertEquals(metadata.getStartOffset(), smallestPartitionOffset);
    LLCSegmentName llcSegmentName = new LLCSegmentName(metadata.getSegmentName());
    Assert.assertEquals(llcSegmentName.getSequenceNumber(), PinotLLCRealtimeSegmentManager.STARTING_SEQUENCE_NUMBER + 2);
    Assert.assertEquals(llcSegmentName.getPartitionId(), partitionToBeFixed);
    // Now pretend the prev segment ended successfully, and set the end offset
    metadata.setEndOffset(metadata.getStartOffset() + 10);
    metadata.setStatus(CommonConstants.Segment.Realtime.Status.DONE);
    segmentManager._records.remove(3);
    segmentManager._records.add(metadata.toZNRecord());
    segmentManager._metadataMap.put(metadata.getSegmentName(), metadata);
    segmentManager._kafkaLargestOffsetToReturn *= 2;
    segmentManager._kafkaSmallestOffsetToReturn *= 2;
    ncPartitions.clear();
    ncPartitions.add(partitionToBeFixed);
    segmentManager.createConsumingSegment(rtTableName, ncPartitions, segmentManager.getExistingSegments(rtTableName), tableConfig);
    Assert.assertEquals(segmentManager._paths.size(), 5);
    Assert.assertEquals(segmentManager._records.size(), 5);
    Assert.assertEquals(segmentManager._oldSegmentNameStr.size(), 5);
    Assert.assertEquals(segmentManager._newSegmentNameStr.size(), 5);
    znRecord = segmentManager._records.get(4);
    metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
    Assert.assertEquals(metadata.getNumReplicas(), 2);
    // In this case, since we have data loss, we will always put the smallest kafka partition available.
    Assert.assertEquals(metadata.getStartOffset(), segmentManager.getKafkaPartitionOffset(null, "smallest", partitionToBeFixed));
    llcSegmentName = new LLCSegmentName(metadata.getSegmentName());
    Assert.assertEquals(llcSegmentName.getSequenceNumber(), PinotLLCRealtimeSegmentManager.STARTING_SEQUENCE_NUMBER + 3);
    Assert.assertEquals(llcSegmentName.getPartitionId(), partitionToBeFixed);
}

Example 28 with IdealState

use of org.apache.helix.model.IdealState in project pinot by linkedin.

the class PinotLLCRealtimeSegmentManagerTest method testInitialSegmentAssignments.

private void testInitialSegmentAssignments(final int nPartitions, final int nInstances, final int nReplicas, boolean existingIS) {
    FakePinotLLCRealtimeSegmentManager segmentManager = new FakePinotLLCRealtimeSegmentManager(true, null);
    final String topic = "someTopic";
    final String rtTableName = "table_REALTIME";
    List<String> instances = getInstanceList(nInstances);
    final String startOffset = KAFKA_OFFSET;
    IdealState idealState = PinotTableIdealStateBuilder.buildEmptyKafkaConsumerRealtimeIdealStateFor(rtTableName, nReplicas);
    segmentManager.setupHelixEntries(topic, rtTableName, nPartitions, instances, nReplicas, startOffset, DUMMY_HOST, idealState, !existingIS, 1000000);
    final String actualRtTableName = segmentManager._realtimeTableName;
    final Map<String, List<String>> idealStateEntries = segmentManager._idealStateEntries;
    final int idealStateNReplicas = segmentManager._nReplicas;
    final List<String> propStorePaths = segmentManager._paths;
    final List<ZNRecord> propStoreEntries = segmentManager._records;
    final boolean createNew = segmentManager._createNew;
    Assert.assertEquals(propStorePaths.size(), nPartitions);
    Assert.assertEquals(propStoreEntries.size(), nPartitions);
    Assert.assertEquals(idealStateEntries.size(), nPartitions);
    Assert.assertEquals(actualRtTableName, rtTableName);
    Assert.assertEquals(createNew, !existingIS);
    Assert.assertEquals(idealStateNReplicas, nReplicas);
    Map<Integer, ZNRecord> segmentPropStoreMap = new HashMap<>(propStorePaths.size());
    Map<Integer, String> segmentPathsMap = new HashMap<>(propStorePaths.size());
    for (String path : propStorePaths) {
        String segNameStr = path.split("/")[3];
        int partition = new LLCSegmentName(segNameStr).getPartitionId();
        segmentPathsMap.put(partition, path);
    }
    for (ZNRecord znRecord : propStoreEntries) {
        LLCRealtimeSegmentZKMetadata metadata = new LLCRealtimeSegmentZKMetadata(znRecord);
        segmentPropStoreMap.put(new LLCSegmentName(metadata.getSegmentName()).getPartitionId(), znRecord);
    }
    Assert.assertEquals(segmentPathsMap.size(), nPartitions);
    Assert.assertEquals(segmentPropStoreMap.size(), nPartitions);
    for (int partition = 0; partition < nPartitions; partition++) {
        final LLCRealtimeSegmentZKMetadata metadata = new LLCRealtimeSegmentZKMetadata(segmentPropStoreMap.get(partition));
        // Just for coverage
        metadata.toString();
        ZNRecord znRecord = metadata.toZNRecord();
        LLCRealtimeSegmentZKMetadata metadataCopy = new LLCRealtimeSegmentZKMetadata(znRecord);
        Assert.assertEquals(metadata, metadataCopy);
        final String path = segmentPathsMap.get(partition);
        final String segmentName = metadata.getSegmentName();
        Assert.assertEquals(metadata.getStartOffset(), -1L);
        Assert.assertEquals(path, "/SEGMENTS/" + rtTableName + "/" + segmentName);
        LLCSegmentName llcSegmentName = new LLCSegmentName(segmentName);
        Assert.assertEquals(llcSegmentName.getPartitionId(), partition);
        Assert.assertEquals(llcSegmentName.getTableName(), TableNameBuilder.extractRawTableName(rtTableName));
        Assert.assertEquals(metadata.getNumReplicas(), nReplicas);
    }
}

Example 29 with IdealState

use of org.apache.helix.model.IdealState in project pinot by linkedin.

the class PinotLLCRealtimeSegmentManagerTest method testPreExistingSegments.

@Test
public void testPreExistingSegments() throws Exception {
    LLCSegmentName existingSegmentName = new LLCSegmentName("someTable", 1, 31, 12355L);
    String[] existingSegs = { existingSegmentName.getSegmentName() };
    FakePinotLLCRealtimeSegmentManager segmentManager = new FakePinotLLCRealtimeSegmentManager(true, Arrays.asList(existingSegs));
    final String topic = "someTopic";
    final String rtTableName = "table_REALTIME";
    List<String> instances = getInstanceList(3);
    final String startOffset = KAFKA_OFFSET;
    IdealState idealState = PinotTableIdealStateBuilder.buildEmptyKafkaConsumerRealtimeIdealStateFor(rtTableName, 10);
    try {
        segmentManager.setupHelixEntries(topic, rtTableName, 8, instances, 3, startOffset, DUMMY_HOST, idealState, false, 10000);
        Assert.fail("Did not get expected exception when setting up helix with existing segments in propertystore");
    } catch (RuntimeException e) {
    // Expected
    }
    try {
        segmentManager.setupHelixEntries(topic, rtTableName, 8, instances, 3, startOffset, DUMMY_HOST, idealState, true, 10000);
        Assert.fail("Did not get expected exception when setting up helix with existing segments in propertystore");
    } catch (RuntimeException e) {
    // Expected
    }
}

Example 30 with IdealState

use of org.apache.helix.model.IdealState in project pinot by linkedin.

the class RetentionManagerTest method setupRealtimeTable.

// The most recent will be in
private Set<String> setupRealtimeTable(final int nSegments, final long now) throws Exception {
    final int replicaCount = 1;
    createRealtimeTableConfig(_realtimeTableName, replicaCount);
    Set<String> remainingSegments = new HashSet<>();
    IdealState idealState = PinotTableIdealStateBuilder.buildEmptyKafkaConsumerRealtimeIdealStateFor(_realtimeTableName, replicaCount);
    final int kafkaPartition = 5;
    final long millisInDays = TimeUnit.DAYS.toMillis(1);
    final String serverName = "Server_localhost_0";
    // If we set the segment creation time to a certain value and compare it as being X ms old,
    // then we could get unpredictable results depending on whether it takes more or less than
    // one millisecond to get to RetentionManager time comparison code. To be safe, set the
    // milliseconds off by 1/2 day.
    long segmentCreationTime = now - (nSegments + 1) * millisInDays + millisInDays / 2;
    List<LLCRealtimeSegmentZKMetadata> segmentZKMetadatas = new ArrayList<>();
    for (int seq = 1; seq <= nSegments; seq++) {
        segmentCreationTime += millisInDays;
        LLCRealtimeSegmentZKMetadata segmentMetadata = createSegmentMetadata(replicaCount, segmentCreationTime);
        LLCSegmentName llcSegmentName = new LLCSegmentName(_testTableName, kafkaPartition, seq, segmentCreationTime);
        final String segName = llcSegmentName.getSegmentName();
        segmentMetadata.setSegmentName(segName);
        if (seq == nSegments) {
            // create consuming segment
            segmentMetadata.setStatus(CommonConstants.Segment.Realtime.Status.IN_PROGRESS);
            idealState.setPartitionState(segName, serverName, "CONSUMING");
            remainingSegments.add(segName);
        } else if (seq % 2 == 0) {
            // create ONLINE segment
            segmentMetadata.setStatus(CommonConstants.Segment.Realtime.Status.DONE);
            idealState.setPartitionState(segName, serverName, "ONLINE");
            remainingSegments.add(segName);
        } else {
            segmentMetadata.setStatus(CommonConstants.Segment.Realtime.Status.IN_PROGRESS);
            idealState.setPartitionState(segName, serverName, "OFFLINE");
            if (now - segmentCreationTime < TimeUnit.DAYS.toMillis(RetentionManager.getRetentionTimeForOldLLCSegmentsDays())) {
                remainingSegments.add(segName);
            }
        }
        final String znodePath = ZKMetadataProvider.constructPropertyStorePathForSegment(_realtimeTableName, segName);
        _propertyStore.set(znodePath, segmentMetadata.toZNRecord(), AccessOption.PERSISTENT);
    }
    _helixAdmin.addResource(HELIX_CLUSTER_NAME, _realtimeTableName, idealState);
    return remainingSegments;
}