use of com.github.ambry.messageformat.ValidatingTransformer in project ambry by linkedin.
the class ReplicationTest method remoteReplicaInfoAddRemoveTest.
/**
* Tests add/remove replicaInfo to {@link ReplicaThread}
* @throws Exception
*/
@Test
public void remoteReplicaInfoAddRemoveTest() throws Exception {
MockClusterMap clusterMap = new MockClusterMap();
Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
MockHost localHost = localAndRemoteHosts.getFirst();
MockHost remoteHost = localAndRemoteHosts.getSecond();
StoreKeyFactory storeKeyFactory = Utils.getObj("com.github.ambry.commons.BlobIdFactory", clusterMap);
MockStoreKeyConverterFactory mockStoreKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
mockStoreKeyConverterFactory.setReturnInputIfAbsent(true);
mockStoreKeyConverterFactory.setConversionMap(new HashMap<>());
StoreKeyConverter storeKeyConverter = mockStoreKeyConverterFactory.getStoreKeyConverter();
Transformer transformer = new ValidatingTransformer(storeKeyFactory, storeKeyConverter);
ReplicationMetrics replicationMetrics = new ReplicationMetrics(new MetricRegistry(), clusterMap.getReplicaIds(localHost.dataNodeId));
replicationMetrics.populateSingleColoMetrics(remoteHost.dataNodeId.getDatacenterName());
List<RemoteReplicaInfo> remoteReplicaInfoList = localHost.getRemoteReplicaInfos(remoteHost, null);
Map<DataNodeId, MockHost> hosts = new HashMap<>();
hosts.put(remoteHost.dataNodeId, remoteHost);
MockConnectionPool connectionPool = new MockConnectionPool(hosts, clusterMap, 4);
ReplicaThread replicaThread = new ReplicaThread("threadtest", new MockFindTokenHelper(storeKeyFactory, replicationConfig), clusterMap, new AtomicInteger(0), localHost.dataNodeId, connectionPool, replicationConfig, replicationMetrics, null, mockStoreKeyConverterFactory.getStoreKeyConverter(), transformer, clusterMap.getMetricRegistry(), false, localHost.dataNodeId.getDatacenterName(), new ResponseHandler(clusterMap), time, null, null, null);
for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfoList) {
replicaThread.addRemoteReplicaInfo(remoteReplicaInfo);
}
List<RemoteReplicaInfo> actualRemoteReplicaInfoList = replicaThread.getRemoteReplicaInfos().get(remoteHost.dataNodeId);
Comparator<RemoteReplicaInfo> remoteReplicaInfoComparator = Comparator.comparing(info -> info.getReplicaId().getPartitionId().toPathString());
Collections.sort(remoteReplicaInfoList, remoteReplicaInfoComparator);
Collections.sort(actualRemoteReplicaInfoList, remoteReplicaInfoComparator);
assertEquals("getRemoteReplicaInfos not correct", remoteReplicaInfoList, actualRemoteReplicaInfoList);
// Test remove remoteReplicaInfo.
replicaThread.removeRemoteReplicaInfo(remoteReplicaInfoList.get(remoteReplicaInfoList.size() - 1));
actualRemoteReplicaInfoList = replicaThread.getRemoteReplicaInfos().get(remoteHost.dataNodeId);
Collections.sort(actualRemoteReplicaInfoList, remoteReplicaInfoComparator);
remoteReplicaInfoList.remove(remoteReplicaInfoList.size() - 1);
assertEquals("getRemoteReplicaInfos not correct", remoteReplicaInfoList, actualRemoteReplicaInfoList);
}
use of com.github.ambry.messageformat.ValidatingTransformer in project ambry by linkedin.
the class ReplicationTest method replicationPauseTest.
/**
* Tests pausing replication for all and individual partitions. Also tests replication will pause on store that is not
* started and resume when store restarted.
* @throws Exception
*/
@Test
public void replicationPauseTest() throws Exception {
MockClusterMap clusterMap = new MockClusterMap();
Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
MockHost localHost = localAndRemoteHosts.getFirst();
MockHost remoteHost = localAndRemoteHosts.getSecond();
List<PartitionId> partitionIds = clusterMap.getAllPartitionIds(null);
for (PartitionId partitionId : partitionIds) {
// add 10 messages to the remote host only
addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHost), 10);
}
StoreKeyFactory storeKeyFactory = new BlobIdFactory(clusterMap);
MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
storeKeyConverterFactory.setConversionMap(new HashMap<>());
storeKeyConverterFactory.setReturnInputIfAbsent(true);
StoreKeyConverter storeKeyConverter = storeKeyConverterFactory.getStoreKeyConverter();
Transformer transformer = new ValidatingTransformer(storeKeyFactory, storeKeyConverter);
int batchSize = 4;
Pair<Map<DataNodeId, List<RemoteReplicaInfo>>, ReplicaThread> replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, null, null);
Map<DataNodeId, List<RemoteReplicaInfo>> replicasToReplicate = replicasAndThread.getFirst();
ReplicaThread replicaThread = replicasAndThread.getSecond();
Map<PartitionId, Integer> progressTracker = new HashMap<>();
PartitionId partitionToResumeFirst = clusterMap.getAllPartitionIds(null).get(0);
PartitionId partitionToShutdownLocally = clusterMap.getAllPartitionIds(null).get(1);
boolean allStopped = false;
boolean onlyOneResumed = false;
boolean allReenabled = false;
boolean shutdownStoreRestarted = false;
Set<PartitionId> expectedPaused = new HashSet<>();
assertEquals("There should be no disabled partitions", expectedPaused, replicaThread.getReplicationDisabledPartitions());
while (true) {
replicaThread.replicate();
boolean replicationDone = true;
for (RemoteReplicaInfo replicaInfo : replicasToReplicate.get(remoteHost.dataNodeId)) {
PartitionId id = replicaInfo.getReplicaId().getPartitionId();
MockFindToken token = (MockFindToken) replicaInfo.getToken();
int lastProgress = progressTracker.computeIfAbsent(id, id1 -> 0);
int currentProgress = token.getIndex();
boolean partDone = currentProgress + 1 == remoteHost.infosByPartition.get(id).size();
if (allStopped || (onlyOneResumed && !id.equals(partitionToResumeFirst)) || (allReenabled && !shutdownStoreRestarted && id.equals(partitionToShutdownLocally))) {
assertEquals("There should have been no progress", lastProgress, currentProgress);
} else if (!partDone) {
assertTrue("There has been no progress", currentProgress > lastProgress);
progressTracker.put(id, currentProgress);
}
replicationDone = replicationDone && partDone;
}
if (!allStopped && !onlyOneResumed && !allReenabled && !shutdownStoreRestarted) {
replicaThread.controlReplicationForPartitions(clusterMap.getAllPartitionIds(null), false);
expectedPaused.addAll(clusterMap.getAllPartitionIds(null));
assertEquals("Disabled partitions sets do not match", expectedPaused, replicaThread.getReplicationDisabledPartitions());
allStopped = true;
} else if (!onlyOneResumed && !allReenabled && !shutdownStoreRestarted) {
// resume replication for first partition
replicaThread.controlReplicationForPartitions(Collections.singletonList(partitionIds.get(0)), true);
expectedPaused.remove(partitionIds.get(0));
assertEquals("Disabled partitions sets do not match", expectedPaused, replicaThread.getReplicationDisabledPartitions());
allStopped = false;
onlyOneResumed = true;
} else if (!allReenabled && !shutdownStoreRestarted) {
// not removing the first partition
replicaThread.controlReplicationForPartitions(clusterMap.getAllPartitionIds(null), true);
// shutdown one local store to pause replication against that store
localHost.storesByPartition.get(partitionToShutdownLocally).shutdown();
onlyOneResumed = false;
allReenabled = true;
expectedPaused.clear();
assertEquals("Disabled partitions sets do not match", expectedPaused, replicaThread.getReplicationDisabledPartitions());
} else if (!shutdownStoreRestarted) {
localHost.storesByPartition.get(partitionToShutdownLocally).start();
shutdownStoreRestarted = true;
}
if (replicationDone) {
break;
}
}
Map<PartitionId, List<MessageInfo>> missingInfos = remoteHost.getMissingInfos(localHost.infosByPartition);
for (Map.Entry<PartitionId, List<MessageInfo>> entry : missingInfos.entrySet()) {
assertEquals("No infos should be missing", 0, entry.getValue().size());
}
Map<PartitionId, List<ByteBuffer>> missingBuffers = remoteHost.getMissingBuffers(localHost.buffersByPartition);
for (Map.Entry<PartitionId, List<ByteBuffer>> entry : missingBuffers.entrySet()) {
assertEquals("No buffers should be missing", 0, entry.getValue().size());
}
}
use of com.github.ambry.messageformat.ValidatingTransformer in project ambry by linkedin.
the class ReplicationTest method limitMaxPartitionCountPerRequestTest.
/**
* Test that max partition count per request is honored in {@link ReplicaThread} if there are too many partitions to
* replicate from the remote node.
* @throws Exception
*/
@Test
public void limitMaxPartitionCountPerRequestTest() throws Exception {
MockClusterMap clusterMap = new MockClusterMap();
Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
MockHost localHost = localAndRemoteHosts.getFirst();
MockHost remoteHost = localAndRemoteHosts.getSecond();
List<PartitionId> partitionIds = clusterMap.getAllPartitionIds(null);
for (PartitionId partitionId : partitionIds) {
// add 5 messages into each partition and place it on remote host only
addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHost), 5);
}
StoreKeyFactory storeKeyFactory = Utils.getObj("com.github.ambry.commons.BlobIdFactory", clusterMap);
MockStoreKeyConverterFactory mockStoreKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
mockStoreKeyConverterFactory.setReturnInputIfAbsent(true);
mockStoreKeyConverterFactory.setConversionMap(new HashMap<>());
// we set batchSize to 10 in order to get all messages from one partition within single replication cycle
int batchSize = 10;
StoreKeyConverter storeKeyConverter = mockStoreKeyConverterFactory.getStoreKeyConverter();
Transformer transformer = new ValidatingTransformer(storeKeyFactory, storeKeyConverter);
// we set max partition count per request to 5, which forces thread to replicate replicas in two cycles. (Note that
// number of partition to replicate is 10, they will be replicated in two batches)
ReplicationConfig initialReplicationConfig = replicationConfig;
properties.setProperty("replication.max.partition.count.per.request", String.valueOf(5));
replicationConfig = new ReplicationConfig(new VerifiableProperties(properties));
CountDownLatch replicationCompleted = new CountDownLatch(partitionIds.size());
AtomicReference<Exception> exception = new AtomicReference<>();
Pair<Map<DataNodeId, List<RemoteReplicaInfo>>, ReplicaThread> replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, (store, messageInfos) -> {
try {
replicationCompleted.countDown();
// for each partition, replication should complete within single cycle (fetch once should suffice), so
// we shut down local store once blobs are written. This can avoid unnecessary metadata requests sent to
// remote host.
store.shutdown();
} catch (Exception e) {
exception.set(e);
}
}, null);
ReplicaThread replicaThread = replicasAndThread.getSecond();
Thread thread = Utils.newThread(replicaThread, false);
thread.start();
assertTrue("Replication didn't complete within 10 secs", replicationCompleted.await(10, TimeUnit.SECONDS));
// verify the # of replicas per metadata request is limited to 5 (note that there are 10 replicas to replicate, they
// are split into to 2 small batches and get replicated in separate requests)
assertEquals("There should be 2 metadata requests and each has 5 replicas to replicate", Arrays.asList(5, 5), remoteHost.replicaCountPerRequestTracker);
// shutdown
replicaThread.shutdown();
if (exception.get() != null) {
throw exception.get();
}
replicationConfig = initialReplicationConfig;
}
use of com.github.ambry.messageformat.ValidatingTransformer in project ambry by linkedin.
the class ReplicationTest method replicationAllPauseTest.
/**
* Tests pausing all partitions and makes sure that the replica thread pauses. Also tests that it resumes when one
* eligible partition is re-enabled and that replication completes successfully.
* @throws Exception
*/
@Test
public void replicationAllPauseTest() throws Exception {
MockClusterMap clusterMap = new MockClusterMap();
Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
MockHost localHost = localAndRemoteHosts.getFirst();
MockHost remoteHost = localAndRemoteHosts.getSecond();
List<PartitionId> partitionIds = clusterMap.getAllPartitionIds(null);
for (PartitionId partitionId : partitionIds) {
// add 10 messages into each partition and place it on remote host only
addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHost), 10);
}
StoreKeyFactory storeKeyFactory = Utils.getObj("com.github.ambry.commons.BlobIdFactory", clusterMap);
MockStoreKeyConverterFactory mockStoreKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
mockStoreKeyConverterFactory.setReturnInputIfAbsent(true);
mockStoreKeyConverterFactory.setConversionMap(new HashMap<>());
int batchSize = 4;
StoreKeyConverter storeKeyConverter = mockStoreKeyConverterFactory.getStoreKeyConverter();
Transformer transformer = new ValidatingTransformer(storeKeyFactory, storeKeyConverter);
CountDownLatch readyToPause = new CountDownLatch(1);
CountDownLatch readyToProceed = new CountDownLatch(1);
AtomicReference<CountDownLatch> reachedLimitLatch = new AtomicReference<>(new CountDownLatch(1));
AtomicReference<Exception> exception = new AtomicReference<>();
Pair<Map<DataNodeId, List<RemoteReplicaInfo>>, ReplicaThread> replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, (store, messageInfos) -> {
try {
readyToPause.countDown();
readyToProceed.await();
if (store.messageInfos.size() == remoteHost.infosByPartition.get(store.id).size()) {
reachedLimitLatch.get().countDown();
}
} catch (Exception e) {
exception.set(e);
}
}, null);
ReplicaThread replicaThread = replicasAndThread.getSecond();
Thread thread = Utils.newThread(replicaThread, false);
thread.start();
assertEquals("There should be no disabled partitions", 0, replicaThread.getReplicationDisabledPartitions().size());
// wait to pause replication
readyToPause.await(10, TimeUnit.SECONDS);
replicaThread.controlReplicationForPartitions(clusterMap.getAllPartitionIds(null), false);
Set<PartitionId> expectedPaused = new HashSet<>(clusterMap.getAllPartitionIds(null));
assertEquals("Disabled partitions sets do not match", expectedPaused, replicaThread.getReplicationDisabledPartitions());
// signal the replica thread to move forward
readyToProceed.countDown();
// wait for the thread to go into waiting state
assertTrue("Replica thread did not go into waiting state", TestUtils.waitUntilExpectedState(thread, Thread.State.WAITING, 10000));
// unpause one partition
replicaThread.controlReplicationForPartitions(Collections.singletonList(partitionIds.get(0)), true);
expectedPaused.remove(partitionIds.get(0));
assertEquals("Disabled partitions sets do not match", expectedPaused, replicaThread.getReplicationDisabledPartitions());
// wait for it to catch up
reachedLimitLatch.get().await(10, TimeUnit.SECONDS);
// reset limit
reachedLimitLatch.set(new CountDownLatch(partitionIds.size() - 1));
// unpause all partitions
replicaThread.controlReplicationForPartitions(clusterMap.getAllPartitionIds(null), true);
assertEquals("There should be no disabled partitions", 0, replicaThread.getReplicationDisabledPartitions().size());
// wait until all catch up
reachedLimitLatch.get().await(10, TimeUnit.SECONDS);
// shutdown
replicaThread.shutdown();
if (exception.get() != null) {
throw exception.get();
}
Map<PartitionId, List<MessageInfo>> missingInfos = remoteHost.getMissingInfos(localHost.infosByPartition);
for (Map.Entry<PartitionId, List<MessageInfo>> entry : missingInfos.entrySet()) {
assertEquals("No infos should be missing", 0, entry.getValue().size());
}
Map<PartitionId, List<ByteBuffer>> missingBuffers = remoteHost.getMissingBuffers(localHost.buffersByPartition);
for (Map.Entry<PartitionId, List<ByteBuffer>> entry : missingBuffers.entrySet()) {
assertEquals("No buffers should be missing", 0, entry.getValue().size());
}
}
Aggregations