Examples with ReplicaId com.github.ambry.clustermap.ReplicaId used on opensource projects

Example 66 with ReplicaId

use of com.github.ambry.clustermap.ReplicaId in project ambry by linkedin.

the class StorageManagerTest method storeStartFailureTest.

/**
 * Tests that {@link StorageManager} can start even when certain stores cannot be started. Checks that these stores
 * are not accessible. We can make the replica path non-readable to induce a store starting failure.
 * @throws Exception
 */
@Test
public void storeStartFailureTest() throws Exception {
    MockDataNodeId dataNode = clusterMap.getDataNodes().get(0);
    List<ReplicaId> replicas = clusterMap.getReplicaIds(dataNode);
    Set<Integer> badReplicaIndexes = new HashSet<>(Arrays.asList(2, 7));
    for (Integer badReplicaIndex : badReplicaIndexes) {
        new File(replicas.get(badReplicaIndex).getReplicaPath()).setReadable(false);
    }
    StorageManager storageManager = createStorageManager(dataNode, metricRegistry, null);
    storageManager.start();
    assertEquals("There should be no unexpected partitions reported", 0, getNumUnrecognizedPartitionsReported());
    Map<String, Counter> counters = metricRegistry.getCounters();
    assertEquals(0, getCounterValue(counters, DiskSpaceAllocator.class.getName(), "DiskSpaceAllocatorInitFailureCount"));
    assertEquals(badReplicaIndexes.size(), getCounterValue(counters, DiskManager.class.getName(), "TotalStoreStartFailures"));
    assertEquals(0, getCounterValue(counters, DiskManager.class.getName(), "DiskMountPathFailures"));
    for (int i = 0; i < replicas.size(); i++) {
        ReplicaId replica = replicas.get(i);
        PartitionId id = replica.getPartitionId();
        if (badReplicaIndexes.contains(i)) {
            assertNull("This store should not be accessible.", storageManager.getStore(id, false));
            assertFalse("Compaction should not be scheduled", storageManager.scheduleNextForCompaction(id));
        } else {
            Store store = storageManager.getStore(id, false);
            assertTrue("Store should be started", store.isStarted());
            assertTrue("Compaction should be scheduled", storageManager.scheduleNextForCompaction(id));
        }
    }
    assertEquals("Compaction thread count is incorrect", dataNode.getMountPaths().size(), TestUtils.numThreadsByThisName(CompactionManager.THREAD_NAME_PREFIX));
    verifyCompactionThreadCount(storageManager, dataNode.getMountPaths().size());
    shutdownAndAssertStoresInaccessible(storageManager, replicas);
    assertEquals("Compaction thread count is incorrect", 0, storageManager.getCompactionThreadCount());
}

Example 67 with ReplicaId

use of com.github.ambry.clustermap.ReplicaId in project ambry by linkedin.

the class StorageManagerTest method setBlobStoreStoppedStateFailureTest.

/**
 * Test set stopped state of blobstore with given list of {@link PartitionId} in failure cases.
 */
@Test
public void setBlobStoreStoppedStateFailureTest() throws Exception {
    MockDataNodeId dataNode = clusterMap.getDataNodes().get(0);
    List<ReplicaId> replicas = clusterMap.getReplicaIds(dataNode);
    List<MockDataNodeId> dataNodes = new ArrayList<>();
    dataNodes.add(dataNode);
    MockPartitionId invalidPartition = new MockPartitionId(Long.MAX_VALUE, MockClusterMap.DEFAULT_PARTITION_CLASS, dataNodes, 0);
    List<? extends ReplicaId> invalidPartitionReplicas = invalidPartition.getReplicaIds();
    StorageManager storageManager = createStorageManager(dataNode, metricRegistry, null);
    storageManager.start();
    assertEquals("There should be 1 unexpected partition reported", 1, getNumUnrecognizedPartitionsReported());
    // test set the state of store whose replicaStatusDelegate is null
    ReplicaId replica = replicas.get(0);
    PartitionId id = replica.getPartitionId();
    storageManager.getDiskManager(id).shutdown();
    List<PartitionId> failToUpdateList = storageManager.setBlobStoreStoppedState(Arrays.asList(id), true);
    assertEquals("Set store stopped state should fail on given store whose replicaStatusDelegate is null", id, failToUpdateList.get(0));
    // test invalid partition case (where diskManager == null)
    replica = invalidPartitionReplicas.get(0);
    id = replica.getPartitionId();
    failToUpdateList = storageManager.setBlobStoreStoppedState(Arrays.asList(id), true);
    assertEquals("Set store stopped state should fail on given invalid replica", id, failToUpdateList.get(0));
    shutdownAndAssertStoresInaccessible(storageManager, replicas);
}

Example 68 with ReplicaId

use of com.github.ambry.clustermap.ReplicaId in project ambry by linkedin.

the class StorageManagerTest method removeBlobStoreTest.

/**
 * Test remove blob store with given {@link PartitionId}
 * @throws Exception
 */
@Test
public void removeBlobStoreTest() throws Exception {
    MockDataNodeId dataNode = clusterMap.getDataNodes().get(0);
    List<ReplicaId> replicas = clusterMap.getReplicaIds(dataNode);
    List<MockDataNodeId> dataNodes = new ArrayList<>();
    dataNodes.add(dataNode);
    MockPartitionId invalidPartition = new MockPartitionId(Long.MAX_VALUE, MockClusterMap.DEFAULT_PARTITION_CLASS, dataNodes, 0);
    StorageManager storageManager = createStorageManager(dataNode, metricRegistry, null);
    storageManager.start();
    // Replica[1] will be used to test removing a started store. Replica[2] will be used to test a store with compaction enabled
    for (int i = 3; i < replicas.size(); i++) {
        ReplicaId replica = replicas.get(i);
        PartitionId id = replica.getPartitionId();
        assertTrue("Disable compaction should succeed", storageManager.controlCompactionForBlobStore(id, false));
        assertTrue("Shutdown should succeed on given store", storageManager.shutdownBlobStore(id));
        assertTrue("Removing store should succeed", storageManager.removeBlobStore(id));
        assertNull("The store should not exist", storageManager.getStore(id, false));
    }
    // test remove store that compaction is still enabled on it, even though it is shutdown
    PartitionId id = replicas.get(2).getPartitionId();
    assertTrue("Shutdown should succeed on given store", storageManager.shutdownBlobStore(id));
    assertFalse("Removing store should fail because compaction is enabled on this store", storageManager.removeBlobStore(id));
    // test remove store that is still started
    id = replicas.get(1).getPartitionId();
    assertFalse("Removing store should fail because store is still started", storageManager.removeBlobStore(id));
    // test remove store that the disk manager is not running
    id = replicas.get(0).getPartitionId();
    storageManager.getDiskManager(id).shutdown();
    assertFalse("Removing store should fail because disk manager is not running", storageManager.removeBlobStore(id));
    // test a store that doesn't exist
    assertFalse("Removing not-found store should return false", storageManager.removeBlobStore(invalidPartition));
    shutdownAndAssertStoresInaccessible(storageManager, replicas);
    // test that remove store when compaction executor is not instantiated
    // by default, storeCompactionTriggers = "" which makes compaction executor = null during initialization
    VerifiableProperties vProps = new VerifiableProperties(new Properties());
    storageManager = new StorageManager(new StoreConfig(vProps), diskManagerConfig, Utils.newScheduler(1, false), metricRegistry, new MockIdFactory(), clusterMap, dataNode, new DummyMessageStoreHardDelete(), null, SystemTime.getInstance(), new DummyMessageStoreRecovery(), new InMemAccountService(false, false));
    storageManager.start();
    for (ReplicaId replica : replicas) {
        id = replica.getPartitionId();
        assertTrue("Disable compaction should succeed", storageManager.controlCompactionForBlobStore(id, false));
        assertTrue("Shutdown should succeed on given store", storageManager.shutdownBlobStore(id));
        assertTrue("Removing store should succeed", storageManager.removeBlobStore(id));
        assertNull("The store should not exist", storageManager.getStore(id, false));
    }
    shutdownAndAssertStoresInaccessible(storageManager, replicas);
}

Example 69 with ReplicaId

use of com.github.ambry.clustermap.ReplicaId in project ambry by linkedin.

the class StorageManagerTest method replicaFromInactiveToOfflineTest.

/**
 * Test shutting down blob store failure during Inactive-To-Offline transition.
 * @throws Exception
 */
@Test
public void replicaFromInactiveToOfflineTest() throws Exception {
    generateConfigs(true, false);
    MockDataNodeId localNode = clusterMap.getDataNodes().get(0);
    List<ReplicaId> localReplicas = clusterMap.getReplicaIds(localNode);
    ReplicaId testReplica = localReplicas.get(0);
    MockClusterParticipant mockHelixParticipant = new MockClusterParticipant();
    StorageManager storageManager = createStorageManager(localNode, metricRegistry, Collections.singletonList(mockHelixParticipant));
    storageManager.start();
    // test shutdown store failure (this is induced by shutting down disk manager)
    storageManager.getDiskManager(testReplica.getPartitionId()).shutdown();
    mockHelixParticipant.getReplicaSyncUpManager().initiateDisconnection(testReplica);
    CountDownLatch participantLatch = new CountDownLatch(1);
    Utils.newThread(() -> {
        try {
            mockHelixParticipant.onPartitionBecomeOfflineFromInactive(testReplica.getPartitionId().toPathString());
            fail("should fail because of shutting down store failure");
        } catch (StateTransitionException e) {
            assertEquals("Error code doesn't match", ReplicaOperationFailure, e.getErrorCode());
            participantLatch.countDown();
        }
    }, false).start();
    // make sync-up complete to let code proceed and encounter exception in storage manager.
    mockHelixParticipant.getReplicaSyncUpManager().onDisconnectionComplete(testReplica);
    assertTrue("Helix participant transition didn't get invoked within 1 sec", participantLatch.await(1, TimeUnit.SECONDS));
    shutdownAndAssertStoresInaccessible(storageManager, localReplicas);
}

Example 70 with ReplicaId

use of com.github.ambry.clustermap.ReplicaId in project ambry by linkedin.

the class StorageManagerTest method multiParticipantsMarkStoreInErrorStateTest.

/**
 * Test that, if store is not started, all participants on this node are able to mark it in ERROR state during
 * OFFLINE -> BOOTSTRAP transition.
 * @throws Exception
 */
@Test
public void multiParticipantsMarkStoreInErrorStateTest() throws Exception {
    MockDataNodeId dataNode = clusterMap.getDataNodes().get(0);
    List<ReplicaId> replicas = clusterMap.getReplicaIds(dataNode);
    List<ClusterParticipant> participants = Arrays.asList(new MockClusterParticipant(), new MockClusterParticipant());
    StorageManager storageManager = createStorageManager(dataNode, metricRegistry, participants);
    storageManager.start();
    // stop one of the stores to induce transition failure
    PartitionId id = replicas.get(0).getPartitionId();
    storageManager.shutdownBlobStore(id);
    // verify that both participants throw exception during OFFLINE -> BOOTSTRAP transition
    for (ClusterParticipant participant : participants) {
        try {
            ((MockClusterParticipant) participant).onPartitionBecomeBootstrapFromOffline(id.toPathString());
            fail("should fail because store is not started");
        } catch (StateTransitionException e) {
            assertEquals("Error code doesn't match", StoreNotStarted, e.getErrorCode());
        }
    }
    shutdownAndAssertStoresInaccessible(storageManager, replicas);
}