use of com.github.ambry.store.StorageManager in project ambry by linkedin.
the class ReplicationTestHelper method createStorageManagerAndReplicationManager.
/**
* Helper method to create storage manager and replication manager
* @param clusterMap {@link ClusterMap} to use
* @param clusterMapConfig {@link ClusterMapConfig} to use
* @param clusterParticipant {@link com.github.ambry.clustermap.ClusterParticipant} for listener registration.
* @return a pair of storage manager and replication manager
* @throws Exception
*/
protected Pair<StorageManager, ReplicationManager> createStorageManagerAndReplicationManager(ClusterMap clusterMap, ClusterMapConfig clusterMapConfig, MockHelixParticipant clusterParticipant, ConnectionPool mockConnectionPool) throws Exception {
StoreConfig storeConfig = new StoreConfig(verifiableProperties);
DataNodeId dataNodeId = clusterMap.getDataNodeIds().get(0);
MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
storeKeyConverterFactory.setConversionMap(new HashMap<>());
storeKeyConverterFactory.setReturnInputIfAbsent(true);
StoreKeyFactory storeKeyFactory = new BlobIdFactory(clusterMap);
StorageManager storageManager = new StorageManager(storeConfig, new DiskManagerConfig(verifiableProperties), Utils.newScheduler(1, true), new MetricRegistry(), null, clusterMap, dataNodeId, null, clusterParticipant == null ? null : Collections.singletonList(clusterParticipant), new MockTime(), null, new InMemAccountService(false, false));
storageManager.start();
MockReplicationManager replicationManager = new MockReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, clusterMap, dataNodeId, storeKeyConverterFactory, clusterParticipant, mockConnectionPool, new MockFindTokenHelper(storeKeyFactory, replicationConfig), BlobIdTransformer.class.getName(), storeKeyFactory, time);
return new Pair<>(storageManager, replicationManager);
}
use of com.github.ambry.store.StorageManager in project ambry by linkedin.
the class AmbryServerRequests method handleStopStoreRequest.
/**
* Handles admin request that stops BlobStore
* @param partitionId the {@link PartitionId} associated with BlobStore
* @param numReplicasCaughtUpPerPartition the minimum number of peer replicas per partition that should catch up with
* local store before stopping it.
* @return {@link ServerErrorCode} represents result of handling admin request.
*/
private ServerErrorCode handleStopStoreRequest(PartitionId partitionId, short numReplicasCaughtUpPerPartition) {
ServerErrorCode error = validateRequest(partitionId, RequestOrResponseType.AdminRequest, false);
if (!error.equals(ServerErrorCode.No_Error)) {
logger.debug("Validate request fails for {} with error code {} when trying to stop store", partitionId, error);
return error;
}
if (numReplicasCaughtUpPerPartition < 0) {
logger.debug("The number of replicas to catch up should not be less than zero {}", numReplicasCaughtUpPerPartition);
return ServerErrorCode.Bad_Request;
}
if (!storeManager.controlCompactionForBlobStore(partitionId, false)) {
logger.error("Disable compaction fails on given BlobStore {}", partitionId);
return ServerErrorCode.Unknown_Error;
}
Collection<PartitionId> partitionIds = Collections.singletonList(partitionId);
controlRequestForPartitions(EnumSet.of(RequestOrResponseType.PutRequest, RequestOrResponseType.DeleteRequest, RequestOrResponseType.TtlUpdateRequest), partitionIds, false);
if (!replicationEngine.controlReplicationForPartitions(partitionIds, Collections.<String>emptyList(), false)) {
logger.error("Could not disable replication on {}", partitionIds);
return ServerErrorCode.Unknown_Error;
}
if (!isRemoteLagLesserOrEqual(partitionIds, 0, numReplicasCaughtUpPerPartition)) {
logger.debug("Catchup not done on {}", partitionIds);
return ServerErrorCode.Retry_After_Backoff;
}
controlRequestForPartitions(EnumSet.of(RequestOrResponseType.ReplicaMetadataRequest, RequestOrResponseType.GetRequest), partitionIds, false);
// Shutdown the BlobStore completely
if (storeManager.shutdownBlobStore(partitionId)) {
logger.info("Store is successfully shutdown for partition: {}", partitionId);
List<PartitionId> failToUpdateList = storeManager.setBlobStoreStoppedState(Collections.singletonList(partitionId), true);
if (!failToUpdateList.isEmpty() && clusterParticipant != null) {
logger.error("Fail to add BlobStore(s) {} to stopped list after stop operation completed", failToUpdateList.toArray());
error = ServerErrorCode.Unknown_Error;
}
// is adopted. The intention here is to force Helix to re-elect a new leader replica if necessary.
if (storeManager instanceof StorageManager && clusterParticipant != null) {
// Previous operation has guaranteed the store is not null
Store store = ((StorageManager) storeManager).getStore(partitionId, true);
// Setting disable state will trigger transition error at the very beginning of STANDBY -> INACTIVE transition.
// Thus it doesn't go through decommission process.
((BlobStore) store).setDisableState(true);
clusterParticipant.setReplicaDisabledState(storeManager.getReplica(partitionId.toPathString()), true);
}
} else {
error = ServerErrorCode.Unknown_Error;
logger.error("Shutting down BlobStore fails on {}", partitionId);
}
return error;
}
use of com.github.ambry.store.StorageManager in project ambry by linkedin.
the class StatsManagerTest method testReplicaFromOfflineToDropped.
/**
* Test Offline-To-Dropped transition (both failure and success cases)
* @throws Exception
*/
@Test
public void testReplicaFromOfflineToDropped() throws Exception {
ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
ReplicationConfig replicationConfig = new ReplicationConfig(verifiableProperties);
StoreConfig storeConfig = new StoreConfig(verifiableProperties);
MockClusterMap clusterMap = new MockClusterMap();
DataNodeId currentNode = clusterMap.getDataNodeIds().get(0);
List<ReplicaId> localReplicas = clusterMap.getReplicaIds(currentNode);
StorageManager storageManager = new StorageManager(storeConfig, new DiskManagerConfig(verifiableProperties), Utils.newScheduler(1, true), new MetricRegistry(), null, clusterMap, currentNode, null, Collections.singletonList(clusterParticipant), new MockTime(), null, new InMemAccountService(false, false));
storageManager.start();
MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
storeKeyConverterFactory.setConversionMap(new HashMap<>());
MockReplicationManager mockReplicationManager = new MockReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, clusterMap, currentNode, storeKeyConverterFactory, clusterParticipant);
MockStatsManager mockStatsManager = new MockStatsManager(storageManager, localReplicas, new MetricRegistry(), statsManagerConfig, clusterParticipant);
// 1. attempt to remove replica while store is still running (remove store failure case)
ReplicaId replicaToDrop = localReplicas.get(0);
try {
clusterParticipant.onPartitionBecomeDroppedFromOffline(replicaToDrop.getPartitionId().toPathString());
fail("should fail because store is still running");
} catch (StateTransitionException e) {
assertEquals("Error code doesn't match", ReplicaOperationFailure, e.getErrorCode());
}
// 2. shutdown the store but introduce file deletion failure (put a invalid dir in store dir)
storageManager.shutdownBlobStore(replicaToDrop.getPartitionId());
File invalidDir = new File(replicaToDrop.getReplicaPath(), "invalidDir");
invalidDir.deleteOnExit();
assertTrue("Couldn't create dir within store dir", invalidDir.mkdir());
assertTrue("Could not make unreadable", invalidDir.setReadable(false));
try {
clusterParticipant.onPartitionBecomeDroppedFromOffline(replicaToDrop.getPartitionId().toPathString());
fail("should fail because store deletion fails");
} catch (StateTransitionException e) {
assertEquals("Error code doesn't match", ReplicaOperationFailure, e.getErrorCode());
}
// reset permission to allow deletion to succeed.
assertTrue("Could not make readable", invalidDir.setReadable(true));
assertTrue("Could not delete invalid dir", invalidDir.delete());
// 3. success case (remove another replica because previous replica has been removed from in-mem data structures)
ReplicaId replica = localReplicas.get(1);
storageManager.shutdownBlobStore(replica.getPartitionId());
MockHelixParticipant mockHelixParticipant = Mockito.spy(clusterParticipant);
doNothing().when(mockHelixParticipant).setPartitionDisabledState(anyString(), anyBoolean());
mockHelixParticipant.onPartitionBecomeDroppedFromOffline(replica.getPartitionId().toPathString());
// verify that the replica is no longer present in StorageManager
assertNull("Store of removed replica should not exist", storageManager.getStore(replica.getPartitionId(), true));
// purposely remove the same replica in ReplicationManager again to verify it no longer exists
assertFalse("Should return false because replica no longer exists", mockReplicationManager.removeReplica(replica));
// purposely remove the same replica in StatsManager again to verify it no longer exists
assertFalse("Should return false because replica no longer exists", mockStatsManager.removeReplica(replica));
verify(mockHelixParticipant).setPartitionDisabledState(replica.getPartitionId().toPathString(), false);
storageManager.shutdown();
mockStatsManager.shutdown();
}
use of com.github.ambry.store.StorageManager in project ambry by linkedin.
the class StatsManagerTest method testAddAndRemoveReplica.
/**
* Test to verify the {@link StatsManager} behaves correctly when dynamically adding/removing {@link ReplicaId}.
* @throws Exception
*/
@Test
public void testAddAndRemoveReplica() throws Exception {
// setup testing environment
Map<PartitionId, Store> testStoreMap = new HashMap<>();
List<ReplicaId> testReplicas = new ArrayList<>();
DataNodeId dataNodeId = new MockDataNodeId(Collections.singletonList(new Port(6667, PortType.PLAINTEXT)), Collections.singletonList("/tmp"), "DC1");
for (int i = 0; i < 3; i++) {
PartitionId partitionId = new MockPartitionId(i, MockClusterMap.DEFAULT_PARTITION_CLASS, Collections.singletonList((MockDataNodeId) dataNodeId), 0);
testStoreMap.put(partitionId, new MockStore(new MockStoreStats(hostAccountStorageStats.getStorageStats().get(i), false)));
testReplicas.add(partitionId.getReplicaIds().get(0));
}
StorageManager mockStorageManager = new MockStorageManager(testStoreMap, dataNodeId);
StatsManager testStatsManager = new StatsManager(mockStorageManager, testReplicas, new MetricRegistry(), statsManagerConfig, new MockTime(), null, null, inMemoryAccountService);
// verify that adding an existing store to StatsManager should fail
assertFalse("Adding a store which already exists should fail", testStatsManager.addReplica(testReplicas.get(0)));
PartitionId partitionId3 = new MockPartitionId(3, MockClusterMap.DEFAULT_PARTITION_CLASS, Collections.singletonList((MockDataNodeId) dataNodeId), 0);
testStoreMap.put(partitionId3, new MockStore(new MockStoreStats(hostAccountStorageStats.getStorageStats().get(0), false)));
// verify that partitionId3 is not in stats report before adding to statsManager
StatsManager.AccountStatsPublisher publisher = testStatsManager.new AccountStatsPublisher(accountStatsStore);
publisher.run();
HostAccountStorageStatsWrapper statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
assertFalse("Partition3 should not present in stats report", statsWrapper.getStats().getStorageStats().containsKey(partitionId3.getId()));
// verify that after adding into statsManager, PartitionId3 is in stats report
testStatsManager.addReplica(partitionId3.getReplicaIds().get(0));
publisher.run();
statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
assertTrue("Partition3 should present in stats report", statsWrapper.getStats().getStorageStats().containsKey(partitionId3.getId()));
// verify that after removing PartitionId0 (corresponding to the first replica in replicas list), PartitionId0 is not in the stats report
PartitionId partitionId0 = testReplicas.get(0).getPartitionId();
assertTrue("Partition0 should present in stats report before removal", statsWrapper.getStats().getStorageStats().containsKey(partitionId0.getId()));
testStoreMap.remove(testReplicas.get(0).getPartitionId());
testStatsManager.removeReplica(testReplicas.get(0));
publisher.run();
statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
assertFalse("Partition0 should not present in stats report after removal", statsWrapper.getStats().getStorageStats().containsKey(partitionId0.getId()));
// verify that removing the PartitionId0 should fail because it no longer exists in StatsManager
assertFalse(testStatsManager.removeReplica(testReplicas.get(0)));
// concurrent remove test
CountDownLatch getStatsCountdown1 = new CountDownLatch(1);
CountDownLatch waitRemoveCountdown = new CountDownLatch(1);
((MockStorageManager) mockStorageManager).waitOperationCountdown = waitRemoveCountdown;
((MockStorageManager) mockStorageManager).firstCall = true;
((MockStorageManager) mockStorageManager).unreachablePartitions.clear();
for (Store store : testStoreMap.values()) {
((MockStore) store).getStatsCountdown = getStatsCountdown1;
((MockStore) store).isCollected = false;
}
List<PartitionId> partitionRemoved = new ArrayList<>();
Utils.newThread(() -> {
// wait until at least one store has been collected (this ensures stats aggregation using old snapshot of map)
try {
getStatsCountdown1.await();
} catch (InterruptedException e) {
throw new IllegalStateException("CountDown await was interrupted", e);
}
// find one store which hasn't been collected
ReplicaId replicaToRemove = null;
for (Map.Entry<PartitionId, Store> partitionToStore : testStoreMap.entrySet()) {
MockStore store = (MockStore) partitionToStore.getValue();
if (!store.isCollected) {
replicaToRemove = partitionToStore.getKey().getReplicaIds().get(0);
break;
}
}
if (replicaToRemove != null) {
testStatsManager.removeReplica(replicaToRemove);
testStoreMap.remove(replicaToRemove.getPartitionId());
partitionRemoved.add(replicaToRemove.getPartitionId());
// count down to allow stats aggregation to proceed
waitRemoveCountdown.countDown();
}
}, false).start();
publisher.run();
statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
// verify that the removed store is indeed unreachable during stats aggregation
assertTrue("The removed partition should be unreachable during aggregation", ((MockStorageManager) mockStorageManager).unreachablePartitions.contains(partitionRemoved.get(0)));
// verify unreachable store list doesn't contain the store which is removed.
List<String> unreachableStores = statsWrapper.getHeader().getUnreachableStores();
assertFalse("The removed partition should not present in unreachable list", unreachableStores.contains(partitionRemoved.get(0).toPathString()));
// concurrent add test
CountDownLatch getStatsCountdown2 = new CountDownLatch(1);
CountDownLatch waitAddCountdown = new CountDownLatch(1);
((MockStorageManager) mockStorageManager).waitOperationCountdown = waitAddCountdown;
((MockStorageManager) mockStorageManager).firstCall = true;
((MockStorageManager) mockStorageManager).unreachablePartitions.clear();
for (Store store : testStoreMap.values()) {
((MockStore) store).getStatsCountdown = getStatsCountdown2;
((MockStore) store).isCollected = false;
}
PartitionId partitionId4 = new MockPartitionId(4, MockClusterMap.DEFAULT_PARTITION_CLASS, Collections.singletonList((MockDataNodeId) dataNodeId), 0);
Utils.newThread(() -> {
// wait until at least one store has been collected (this ensures stats aggregation using old snapshot of map)
try {
getStatsCountdown2.await();
} catch (InterruptedException e) {
throw new IllegalStateException("CountDown await was interrupted", e);
}
testStatsManager.addReplica(partitionId4.getReplicaIds().get(0));
testStoreMap.put(partitionId4, new MockStore(new MockStoreStats(hostAccountStorageStats.getStorageStats().get(0), false)));
// count down to allow stats aggregation to proceed
waitAddCountdown.countDown();
}, false).start();
publisher.run();
statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
// verify that new added PartitionId4 is not in report for this round of aggregation
assertFalse("Partition4 should not present in stats report", statsWrapper.getStats().getStorageStats().containsKey(partitionId4.getId()));
// verify that new added PartitionId4 will be collected for next round of aggregation
publisher.run();
statsWrapper = accountStatsStore.queryHostAccountStorageStatsByHost("localhost", 0);
assertTrue("Partition4 should present in stats report", statsWrapper.getStats().getStorageStats().containsKey(partitionId4.getId()));
}
Aggregations