use of voldemort.cluster.Cluster in project voldemort by voldemort.
the class ZonedRebalanceNonContiguousZonesTest method testShuffleZ1Z3AndShuffleAgain.
@Test(timeout = 600000)
public void testShuffleZ1Z3AndShuffleAgain() throws Exception {
logger.info("Starting testShuffleZZAndShuffleAgain");
// Hacky work around of TOCTOU bind Exception issues. Each test that invokes this method brings servers
// up & down on the same ports. The OS seems to need a rest between subsequent tests...
Thread.sleep(TimeUnit.SECONDS.toMillis(5));
Cluster interimCluster = RebalanceUtils.getInterimCluster(z1z3Current, z1z3Shuffle);
// start all the servers
List<Integer> serverList = new ArrayList<Integer>(interimCluster.getNodeIds());
Map<String, String> configProps = new HashMap<String, String>();
configProps.put("admin.max.threads", "5");
interimCluster = startServers(interimCluster, z1z3StoresXml, serverList, configProps);
// Populate cluster with data
for (StoreDefinition storeDef : z1z3Stores) {
populateData(z1z3Current, storeDef);
}
String bootstrapUrl = getBootstrapUrl(interimCluster, 3);
// Shuffle cluster
ClusterTestUtils.RebalanceKit rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, z1z3Shuffle, z1z3Stores);
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, serverList);
checkConsistentMetadata(z1z3Shuffle, serverList);
// Now, go from shuffled state, back to the original to confirm subsequent rebalances can be invoked.
rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, z1z3Current, z1z3Stores);
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, serverList);
checkConsistentMetadata(z1z3Current, serverList);
// Done.
stopServer(serverList);
}
use of voldemort.cluster.Cluster in project voldemort by voldemort.
the class ZonedRebalanceNonContiguousZonesTest method testRWRebalance.
@Test(timeout = 600000)
public void testRWRebalance() throws Exception {
logger.info("Starting testRWRebalance");
try {
int[] zoneIds = new int[] { 1, 3 };
int[][] nodesPerZone = new int[][] { { 3, 4 }, { 9, 10 } };
int[][] partitionMap = new int[][] { { 0, 2, 4, 6 }, {}, { 1, 3, 5, 7 }, {} };
Cluster currentCluster = ServerTestUtils.getLocalNonContiguousZonedCluster(zoneIds, nodesPerZone, partitionMap, ClusterTestUtils.getClusterPorts());
Cluster finalCluster = UpdateClusterUtils.createUpdatedCluster(currentCluster, 10, Lists.newArrayList(2, 6));
finalCluster = UpdateClusterUtils.createUpdatedCluster(finalCluster, 4, Lists.newArrayList(3, 7));
// start all the servers
List<Integer> serverList = Arrays.asList(3, 4, 9, 10);
Map<String, String> configProps = new HashMap<String, String>();
configProps.put("admin.max.threads", "5");
currentCluster = startServers(currentCluster, storeDefFileWithoutReplication, serverList, configProps);
String bootstrapUrl = getBootstrapUrl(currentCluster, 3);
ClusterTestUtils.RebalanceKit rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, finalCluster);
try {
populateData(currentCluster, rwStoreDefWithoutReplication);
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, Arrays.asList(4, 9));
checkConsistentMetadata(finalCluster, serverList);
} finally {
stopServer(serverList);
}
} catch (AssertionError ae) {
logger.error("Assertion broken in testRWRebalance ", ae);
throw ae;
}
}
use of voldemort.cluster.Cluster in project voldemort by voldemort.
the class ZonedRebalanceNonContiguousZonesTest method testProxyPutDuringRebalancing.
@Test(timeout = 600000)
public void testProxyPutDuringRebalancing() throws Exception {
logger.info("Starting testProxyPutDuringRebalancing");
try {
int[] zoneIds = new int[] { 1, 3 };
int[][] nodesPerZone = new int[][] { { 3, 4, 5 }, { 9, 10, 11 } };
int[][] partitionMap = new int[][] { { 0 }, { 1, 6 }, { 2 }, { 3 }, { 4, 7 }, { 5 } };
Cluster currentCluster = ServerTestUtils.getLocalNonContiguousZonedCluster(zoneIds, nodesPerZone, partitionMap, ClusterTestUtils.getClusterPorts());
Cluster finalCluster = UpdateClusterUtils.createUpdatedCluster(currentCluster, 5, Lists.newArrayList(7));
finalCluster = UpdateClusterUtils.createUpdatedCluster(finalCluster, 11, Lists.newArrayList(6));
/**
* Original partition map
*
* [s3 : p0] [s4 : p1, p6] [s5 : p2]
*
* [s9 : p3] [s10 : p4, p7] [s11 : p5]
*
* final server partition ownership
*
* [s3 : p0] [s4 : p1] [s5 : p2, p7]
*
* [s9 : p3] [s10 : p4] [s11 : p5, p6]
*
* Note that rwStoreDefFileWithReplication is a "2/1/1" store def.
*
* Original server n-ary partition ownership
*
* [s3 : p0, p3-7] [s4 : p0-p7] [s5 : p1-2]
*
* [s9 : p0-3, p6-7] [s10 : p0-p7] [s11 : p4-5]
*
* final server n-ary partition ownership
*
* [s3 : p0, p2-7] [s4 : p0-1] [s5 : p1-p7]
*
* [s9 : p0-3, p5-7] [s10 : p0-4, p7] [s11 : p4-6]
*/
List<Integer> serverList = Arrays.asList(3, 4, 5, 9, 10, 11);
Map<String, String> configProps = new HashMap<String, String>();
configProps.put("admin.max.threads", "5");
final Cluster updatedCurrentCluster = startServers(currentCluster, rwStoreDefFileWithReplication, serverList, configProps);
ExecutorService executors = Executors.newFixedThreadPool(2);
final AtomicBoolean rebalancingComplete = new AtomicBoolean(false);
final List<Exception> exceptions = Collections.synchronizedList(new ArrayList<Exception>());
// Its is imperative that we test in a single shot since multiple batches would mean the proxy bridges
// being torn down and established multiple times and we cannot test against the source
// cluster topology then. getRebalanceKit uses batch size of infinite, so this should be fine.
String bootstrapUrl = getBootstrapUrl(updatedCurrentCluster, 3);
int maxParallel = 2;
final ClusterTestUtils.RebalanceKit rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, maxParallel, finalCluster);
populateData(currentCluster, rwStoreDefWithReplication);
final AdminClient adminClient = rebalanceKit.controller.getAdminClient();
// the plan would cause these partitions to move:
// Partition : Donor -> stealer
//
// p2 (Z-SEC) : s4 -> s3
// p3-6 (Z-PRI) : s4 -> s5
// p7 (Z-PRI) : s3 -> s5
//
// p5 (Z-SEC): s10 -> s9
// p6 (Z-PRI): s10 -> s11
//
// Rebalancing will run on servers 3, 5, 9, & 11
final List<ByteArray> movingKeysList = sampleKeysFromPartition(adminClient, 4, rwStoreDefWithReplication.getName(), Arrays.asList(6), 20);
assertTrue("Empty list of moving keys...", movingKeysList.size() > 0);
final AtomicBoolean rebalancingStarted = new AtomicBoolean(false);
final AtomicBoolean proxyWritesDone = new AtomicBoolean(false);
final HashMap<String, String> baselineTuples = new HashMap<String, String>(testEntries);
final HashMap<String, VectorClock> baselineVersions = new HashMap<String, VectorClock>();
for (String key : baselineTuples.keySet()) {
baselineVersions.put(key, new VectorClock());
}
final CountDownLatch latch = new CountDownLatch(2);
// start get operation.
executors.execute(new Runnable() {
@Override
public void run() {
SocketStoreClientFactory factory = null;
try {
// wait for the rebalancing to begin
List<VoldemortServer> serverList = Lists.newArrayList(serverMap.get(3), serverMap.get(5), serverMap.get(9), serverMap.get(11));
while (!rebalancingComplete.get()) {
Iterator<VoldemortServer> serverIterator = serverList.iterator();
while (serverIterator.hasNext()) {
VoldemortServer server = serverIterator.next();
if (ByteUtils.getString(server.getMetadataStore().get(MetadataStore.SERVER_STATE_KEY, null).get(0).getValue(), "UTF-8").compareTo(VoldemortState.REBALANCING_MASTER_SERVER.toString()) == 0) {
logger.info("Server " + server.getIdentityNode().getId() + " transitioned into REBALANCING MODE");
serverIterator.remove();
}
}
if (serverList.size() == 0) {
rebalancingStarted.set(true);
break;
}
}
if (rebalancingStarted.get()) {
factory = new SocketStoreClientFactory(new ClientConfig().setBootstrapUrls(getBootstrapUrl(updatedCurrentCluster, 3)).setEnableLazy(false).setSocketTimeout(120, TimeUnit.SECONDS).setClientZoneId(3));
final StoreClient<String, String> storeClientRW = new DefaultStoreClient<String, String>(testStoreNameRW, null, factory, 3);
// Initially, all data now with zero vector clock
for (ByteArray movingKey : movingKeysList) {
try {
String keyStr = ByteUtils.getString(movingKey.get(), "UTF-8");
String valStr = "proxy_write";
storeClientRW.put(keyStr, valStr);
baselineTuples.put(keyStr, valStr);
baselineVersions.get(keyStr).incrementVersion(11, System.currentTimeMillis());
proxyWritesDone.set(true);
if (rebalancingComplete.get()) {
break;
}
} catch (InvalidMetadataException e) {
logger.error("Encountered an invalid metadata exception.. ", e);
}
}
}
} catch (Exception e) {
logger.error("Exception in proxy write thread..", e);
exceptions.add(e);
} finally {
if (factory != null)
factory.close();
latch.countDown();
}
}
});
executors.execute(new Runnable() {
@Override
public void run() {
try {
rebalanceKit.rebalance();
} catch (Exception e) {
logger.error("Error in rebalancing... ", e);
exceptions.add(e);
} finally {
rebalancingComplete.set(true);
latch.countDown();
}
}
});
latch.await();
executors.shutdown();
executors.awaitTermination(300, TimeUnit.SECONDS);
assertEquals("Client did not see all server transition into rebalancing state", rebalancingStarted.get(), true);
assertEquals("Not enough time to begin proxy writing", proxyWritesDone.get(), true);
checkEntriesPostRebalance(updatedCurrentCluster, finalCluster, Lists.newArrayList(rwStoreDefWithReplication), Arrays.asList(3, 4, 5, 9, 10, 11), baselineTuples, baselineVersions);
checkConsistentMetadata(finalCluster, serverList);
// check No Exception
if (exceptions.size() > 0) {
for (Exception e : exceptions) {
e.printStackTrace();
}
fail("Should not see any exceptions.");
}
// check that the proxy writes were made to the original donor, node 4
List<ClockEntry> clockEntries = new ArrayList<ClockEntry>(serverList.size());
for (Integer nodeid : serverList) clockEntries.add(new ClockEntry(nodeid.shortValue(), System.currentTimeMillis()));
VectorClock clusterXmlClock = new VectorClock(clockEntries, System.currentTimeMillis());
for (Integer nodeid : serverList) adminClient.metadataMgmtOps.updateRemoteCluster(nodeid, currentCluster, clusterXmlClock);
adminClient.setAdminClientCluster(currentCluster);
checkForTupleEquivalence(adminClient, 4, testStoreNameRW, movingKeysList, baselineTuples, baselineVersions);
// stop servers
try {
stopServer(serverList);
} catch (Exception e) {
throw new RuntimeException(e);
}
} catch (AssertionError ae) {
logger.error("Assertion broken in testProxyPutDuringRebalancing ", ae);
throw ae;
}
}
use of voldemort.cluster.Cluster in project voldemort by voldemort.
the class AbstractZonedRebalanceTest method testShuffleZZAndShuffleAgain.
@Test(timeout = 600000)
public void testShuffleZZAndShuffleAgain() throws Exception {
logger.info("Starting testShuffleZZAndShuffleAgain");
// Hacky work around of TOCTOU bind Exception issues. Each test that
// invokes this method brings servers up & down on the same ports. The
// OS seems to need a rest between subsequent tests...
Thread.sleep(TimeUnit.SECONDS.toMillis(5));
Cluster interimCluster = RebalanceUtils.getInterimCluster(zzCurrent, zzShuffle);
// start all the servers
List<Integer> serverList = new ArrayList<Integer>(interimCluster.getNodeIds());
Map<String, String> configProps = new HashMap<String, String>();
configProps.put("admin.max.threads", "5");
interimCluster = startServers(interimCluster, zzStoresXml, serverList, configProps);
// Populate cluster with data
for (StoreDefinition storeDef : zzStores) {
populateData(zzCurrent, storeDef);
}
String bootstrapUrl = getBootstrapUrl(interimCluster, 0);
// Shuffle cluster
ClusterTestUtils.RebalanceKit rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, zzShuffle, zzStores);
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, serverList);
checkConsistentMetadata(zzShuffle, serverList);
// Now, go from shuffled state, back to the original to ocnfirm
// subsequent rebalances can be invoked.
rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, zzCurrent, zzStores);
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, serverList);
checkConsistentMetadata(zzCurrent, serverList);
// Done.
stopServer(serverList);
}
use of voldemort.cluster.Cluster in project voldemort by voldemort.
the class AbstractZonedRebalanceTest method testZonedRebalance.
// TODO: The tests based on this method are susceptible to TOCTOU
// BindException issue since findFreePorts is used to determine the ports
// for localhost:PORT of each node.
/**
* Scripts the execution of a specific type of zoned rebalance test: sets up
* cluster based on cCluster plus any new nodes/zones in fCluster,
* rebalances to fCluster, verifies rebalance was correct.
*
* @param testTag For pretty printing
* @param cCluster current cluster
* @param fCluster final cluster
* @param cStoresXml XML file with current stores xml
* @param fStoresXml Unused parameter. Included for symmetry in method
* declaration.
* @param cStoreDefs store defs for current cluster (from on cStoresXml)
* @param fStoreDefs store defs for final cluster.
* @throws Exception
*/
public void testZonedRebalance(String testTag, Cluster cCluster, Cluster fCluster, String cStoresXml, String fStoresXml, List<StoreDefinition> cStoreDefs, List<StoreDefinition> fStoreDefs) throws Exception {
logger.info("Starting " + testTag);
// Hacky work around of TOCTOU bind Exception issues. Each test that
// invokes this method brings servers up & down on the same ports. The
// OS seems to need a rest between subsequent tests...
Thread.sleep(TimeUnit.SECONDS.toMillis(5));
try {
Cluster interimCluster = RebalanceUtils.getInterimCluster(cCluster, fCluster);
// start all the servers
List<Integer> serverList = new ArrayList<Integer>(interimCluster.getNodeIds());
Map<String, String> configProps = new HashMap<String, String>();
configProps.put("admin.max.threads", "5");
interimCluster = startServers(interimCluster, cStoresXml, serverList, configProps);
String bootstrapUrl = getBootstrapUrl(interimCluster, 0);
ClusterTestUtils.RebalanceKit rebalanceKit = ClusterTestUtils.getRebalanceKit(bootstrapUrl, fCluster, fStoreDefs);
try {
for (StoreDefinition storeDef : cStoreDefs) {
populateData(cCluster, storeDef);
}
rebalanceAndCheck(rebalanceKit.plan, rebalanceKit.controller, serverList);
checkConsistentMetadata(fCluster, serverList);
} finally {
// stop servers
stopServer(serverList);
}
} catch (AssertionError ae) {
logger.error("Assertion broken in " + testTag + " : ", ae);
throw ae;
}
}
Aggregations