Examples with StoreRoutingPlan voldemort.routing.StoreRoutingPlan used on opensource projects

Example 6 with StoreRoutingPlan

use of voldemort.routing.StoreRoutingPlan in project voldemort by voldemort.

the class ClusterForkLiftToolTest method testForkLiftOverWrite.

@Test
public void testForkLiftOverWrite() throws Exception {
    StoreRoutingPlan srcStoreInstance = new StoreRoutingPlan(srcCluster, globallyResolvingStoreDef);
    // populate data on the source cluster..
    for (Map.Entry<String, String> entry : kvPairs.entrySet()) {
        srcGloballyResolvingStoreClient.put(entry.getKey(), entry.getValue());
    }
    // generate a conflict on the primary and a secondary
    List<Integer> nodeList = srcStoreInstance.getReplicationNodeList(srcStoreInstance.getMasterPartitionId(conflictKey.getBytes("UTF-8")));
    VectorClock losingClock = new VectorClock(Lists.newArrayList(new ClockEntry((short) 0, 5)), System.currentTimeMillis());
    VectorClock winningClock = new VectorClock(Lists.newArrayList(new ClockEntry((short) 1, 5)), losingClock.getTimestamp() + 1);
    srcAdminClient.storeOps.putNodeKeyValue(GLOBALLY_RESOLVING_STORE_NAME, new NodeValue<ByteArray, byte[]>(nodeList.get(0), new ByteArray(conflictKey.getBytes("UTF-8")), new Versioned<byte[]>("losing value".getBytes("UTF-8"), losingClock)));
    srcAdminClient.storeOps.putNodeKeyValue(GLOBALLY_RESOLVING_STORE_NAME, new NodeValue<ByteArray, byte[]>(nodeList.get(1), new ByteArray(conflictKey.getBytes("UTF-8")), new Versioned<byte[]>("winning value".getBytes("UTF-8"), winningClock)));
    // *** do a write to destination cluster ***
    // This is the main test , where after fork lift this value should be
    // overwritten. This is the only difference.
    dstGloballyResolvingStoreClient.put(firstKey, "before forklift");
    // Make the current thread sleep , so when the new clock is generated
    // using milliSeconds it is greater and must be overwritten.
    Thread.sleep(2);
    // perform the forklifting..
    ClusterForkLiftTool forkLiftTool = new ClusterForkLiftTool(srcBootStrapUrl, dstBootStrapUrl, // OverWrite
    true, // ignore
    false, // mismatch
    10000, 1, 1000, Lists.newArrayList(GLOBALLY_RESOLVING_STORE_NAME), null, ClusterForkLiftTool.ForkLiftTaskMode.global_resolution);
    forkLiftTool.run();
    // do a write to destination cluster
    dstGloballyResolvingStoreClient.put(lastKey, "after forklift");
    // verify data on the destination is as expected
    for (Map.Entry<String, String> entry : kvPairs.entrySet()) {
        String dstClusterValue = dstGloballyResolvingStoreClient.get(entry.getKey()).getValue();
        if (entry.getKey().equals(lastKey)) {
            assertEquals("can't update value after forklift", dstClusterValue, "after forklift");
        } else if (entry.getKey().equals(conflictKey)) {
            assertEquals("Conflict resolution incorrect", dstClusterValue, "winning value");
        } else {
            if (!dstClusterValue.equals(entry.getValue())) {
                assertEquals("fork lift data missing", dstClusterValue, entry.getValue());
            }
        }
    }
}

Example 7 with StoreRoutingPlan

use of voldemort.routing.StoreRoutingPlan in project voldemort by voldemort.

the class ClusterForkLiftToolTest method testGloballyResolvingForkLift.

@Test
public void testGloballyResolvingForkLift() throws Exception {
    StoreRoutingPlan srcStoreInstance = new StoreRoutingPlan(srcCluster, globallyResolvingStoreDef);
    // populate data on the source cluster..
    for (Map.Entry<String, String> entry : kvPairs.entrySet()) {
        srcGloballyResolvingStoreClient.put(entry.getKey(), entry.getValue());
    }
    // generate a conflict on the primary and a secondary
    List<Integer> nodeList = srcStoreInstance.getReplicationNodeList(srcStoreInstance.getMasterPartitionId(conflictKey.getBytes("UTF-8")));
    VectorClock losingClock = new VectorClock(Lists.newArrayList(new ClockEntry((short) 0, 5)), System.currentTimeMillis());
    VectorClock winningClock = new VectorClock(Lists.newArrayList(new ClockEntry((short) 1, 5)), losingClock.getTimestamp() + 1);
    srcAdminClient.storeOps.putNodeKeyValue(GLOBALLY_RESOLVING_STORE_NAME, new NodeValue<ByteArray, byte[]>(nodeList.get(0), new ByteArray(conflictKey.getBytes("UTF-8")), new Versioned<byte[]>("losing value".getBytes("UTF-8"), losingClock)));
    srcAdminClient.storeOps.putNodeKeyValue(GLOBALLY_RESOLVING_STORE_NAME, new NodeValue<ByteArray, byte[]>(nodeList.get(1), new ByteArray(conflictKey.getBytes("UTF-8")), new Versioned<byte[]>("winning value".getBytes("UTF-8"), winningClock)));
    // do a write to destination cluster
    dstGloballyResolvingStoreClient.put(firstKey, "before forklift");
    // perform the forklifting..
    ClusterForkLiftTool forkLiftTool = new ClusterForkLiftTool(srcBootStrapUrl, dstBootStrapUrl, false, // ignoreSchemaMismatch
    false, 10000, 1, 1000, Lists.newArrayList(GLOBALLY_RESOLVING_STORE_NAME), null, ClusterForkLiftTool.ForkLiftTaskMode.global_resolution);
    forkLiftTool.run();
    // do a write to destination cluster
    dstGloballyResolvingStoreClient.put(lastKey, "after forklift");
    // verify data on the destination is as expected
    for (Map.Entry<String, String> entry : kvPairs.entrySet()) {
        if (entry.getKey().equals(firstKey)) {
            assertEquals("Online write overwritten", dstGloballyResolvingStoreClient.get(firstKey).getValue(), "before forklift");
        } else if (entry.getKey().equals(lastKey)) {
            assertEquals("can't update value after forklift", dstGloballyResolvingStoreClient.get(lastKey).getValue(), "after forklift");
        } else if (entry.getKey().equals(conflictKey)) {
            assertEquals("Conflict resolution incorrect", dstGloballyResolvingStoreClient.get(conflictKey).getValue(), "winning value");
        } else {
            assertEquals("fork lift data missing", dstGloballyResolvingStoreClient.get(entry.getKey()).getValue(), entry.getValue());
        }
    }
}

Example 8 with StoreRoutingPlan

use of voldemort.routing.StoreRoutingPlan in project voldemort by voldemort.

the class RepairJob method operate.

@Override
public void operate() throws Exception {
    for (StoreDefinition storeDef : metadataStore.getStoreDefList()) {
        if (isWritableStore(storeDef)) {
            // Lets generate routing strategy for this storage engine
            StoreRoutingPlan routingPlan = new StoreRoutingPlan(metadataStore.getCluster(), storeDef);
            logger.info("Repairing store " + storeDef.getName());
            StorageEngine<ByteArray, byte[], byte[]> engine = storeRepo.getStorageEngine(storeDef.getName());
            iterator = engine.keys();
            long itemsScanned = 0;
            long numDeletedKeys = 0;
            while (iterator.hasNext()) {
                ByteArray key = iterator.next();
                if (!routingPlan.checkKeyBelongsToNode(key.get(), metadataStore.getNodeId())) {
                    /**
                         * Blow away the entire key with all its versions..
                         */
                    engine.delete(key, null);
                    numDeletedKeys = this.numKeysUpdatedThisRun.incrementAndGet();
                }
                itemsScanned = this.numKeysScannedThisRun.incrementAndGet();
                throttler.maybeThrottle(1);
                if (itemsScanned % STAT_RECORDS_INTERVAL == 0) {
                    logger.info("#Scanned:" + itemsScanned + " #Deleted:" + numDeletedKeys);
                }
            }
            closeIterator(iterator);
            logger.info("Completed store " + storeDef.getName() + " #Scanned:" + itemsScanned + " #Deleted:" + numDeletedKeys);
        }
    }
}

Example 9 with StoreRoutingPlan

use of voldemort.routing.StoreRoutingPlan in project voldemort by voldemort.

the class AdminRebalanceTest method testRebalanceNodeRO.

@Test(timeout = 60000)
public void testRebalanceNodeRO() throws IOException {
    try {
        startFourNodeRO();
        int numChunks = 5;
        for (StoreDefinition storeDef : Lists.newArrayList(storeDef1, storeDef2)) {
            buildROStore(storeDef, numChunks);
        }
        // Set into rebalancing state
        for (RebalanceTaskInfo partitionPlan : plans) {
            getServer(partitionPlan.getStealerId()).getMetadataStore().put(MetadataStore.SERVER_STATE_KEY, MetadataStore.VoldemortState.REBALANCING_MASTER_SERVER);
            getServer(partitionPlan.getStealerId()).getMetadataStore().put(MetadataStore.REBALANCING_STEAL_INFO, new RebalancerState(Lists.newArrayList(RebalanceTaskInfo.create(partitionPlan.toJsonString()))));
            getServer(partitionPlan.getStealerId()).getMetadataStore().put(MetadataStore.REBALANCING_SOURCE_CLUSTER_XML, partitionPlan.getInitialCluster());
        }
        // Actually run it
        try {
            for (RebalanceTaskInfo currentPlan : plans) {
                int asyncId = adminClient.rebalanceOps.rebalanceNode(currentPlan);
                assertNotSame("Got a valid rebalanceAsyncId", -1, asyncId);
                getAdminClient().rpcOps.waitForCompletion(currentPlan.getStealerId(), asyncId, 300, TimeUnit.SECONDS);
                // Test that plan has been removed from the list
                assertFalse(getServer(currentPlan.getStealerId()).getMetadataStore().getRebalancerState().getAll().contains(currentPlan));
            }
        } catch (Exception e) {
            e.printStackTrace();
            fail("Should not throw any exceptions");
        }
        // Check if files have been copied
        for (StoreDefinition storeDef : Lists.newArrayList(storeDef1, storeDef2)) {
            String storeName = storeDef.getName();
            for (RebalanceTaskInfo currentPlan : plans) {
                MetadataStore metadataStore = getServer(currentPlan.getStealerId()).getMetadataStore();
                int nodeId = metadataStore.getNodeId();
                int zoneId = metadataStore.getCluster().getNodeById(nodeId).getZoneId();
                StoreRoutingPlan storeRoutingPlan = new StoreRoutingPlan(metadataStore.getCluster(), storeDef);
                File currentDir = new File(((ReadOnlyStorageEngine) getStore(currentPlan.getStealerId(), storeName)).getCurrentDirPath());
                if (currentPlan.getPartitionStores().contains(storeDef.getName())) {
                    for (Integer partitionId : currentPlan.getStoreToPartitionIds().get(storeName)) {
                        int zoneNary = -1;
                        // it means we don't want to consider that partition.
                        try {
                            zoneNary = storeRoutingPlan.getZoneNaryForNodesPartition(zoneId, nodeId, partitionId);
                        } catch (VoldemortException ve) {
                            continue;
                        }
                        if (zoneNary < storeDef.getReplicationFactor()) {
                            for (int chunkId = 0; chunkId < numChunks; chunkId++) {
                                assertTrue(new File(currentDir, partitionId + "_" + zoneNary + "_" + chunkId + ".data").exists());
                                assertTrue(new File(currentDir, partitionId + "_" + zoneNary + "_" + chunkId + ".index").exists());
                            }
                        }
                    }
                }
            }
        }
        // All servers should be back to normal state
        for (VoldemortServer server : servers) {
            assertEquals(server.getMetadataStore().getRebalancerState(), new RebalancerState(new ArrayList<RebalanceTaskInfo>()));
            assertEquals(server.getMetadataStore().getServerStateUnlocked(), MetadataStore.VoldemortState.NORMAL_SERVER);
        }
        // Test the "cluster + swap" changes
        // Test 1) Fail some swap by adding a dummy store
        servers[2].getMetadataStore().put(MetadataStore.STORES_KEY, Lists.newArrayList(storeDef1, storeDef2, new StoreDefinitionBuilder().setName("test3").setType(ReadOnlyStorageConfiguration.TYPE_NAME).setKeySerializer(new SerializerDefinition("string")).setValueSerializer(new SerializerDefinition("string")).setRoutingPolicy(RoutingTier.CLIENT).setRoutingStrategyType(RoutingStrategyType.CONSISTENT_STRATEGY).setReplicationFactor(2).setPreferredReads(1).setRequiredReads(1).setPreferredWrites(1).setRequiredWrites(1).build()));
        try {
            adminClient.rebalanceOps.rebalanceStateChange(currentCluster, finalCluster, servers[2].getMetadataStore().getStoreDefList(), servers[2].getMetadataStore().getStoreDefList(), plans, true, true, false, true, true);
            fail("Should have thrown an exception since one node doesn't have the store");
        } catch (VoldemortException e) {
        }
        servers[2].getMetadataStore().put(MetadataStore.STORES_KEY, Lists.newArrayList(storeDef1, storeDef2));
        // Test that all servers are still using the old cluster and have
        // swapped successfully
        checkRO(currentCluster);
        // Test 2) All passes scenario
        adminClient.rebalanceOps.rebalanceStateChange(currentCluster, finalCluster, servers[2].getMetadataStore().getStoreDefList(), servers[2].getMetadataStore().getStoreDefList(), plans, true, true, false, true, true);
        checkRO(finalCluster);
        // mmap-ed. Should fail...
        for (RebalanceTaskInfo partitionPlan : plans) {
            getServer(partitionPlan.getStealerId()).getMetadataStore().put(MetadataStore.SERVER_STATE_KEY, MetadataStore.VoldemortState.REBALANCING_MASTER_SERVER);
            getServer(partitionPlan.getStealerId()).getMetadataStore().put(MetadataStore.REBALANCING_STEAL_INFO, new RebalancerState(Lists.newArrayList(RebalanceTaskInfo.create(partitionPlan.toJsonString()))));
        }
        // Actually run it
        try {
            int asyncId = adminClient.rebalanceOps.rebalanceNode(plans.get(0));
            getAdminClient().rpcOps.waitForCompletion(plans.get(0).getStealerId(), asyncId, 300, TimeUnit.SECONDS);
            fail("Should throw an exception");
        } catch (Exception e) {
        }
    } finally {
        shutDown();
    }
}

Example 10 with StoreRoutingPlan

use of voldemort.routing.StoreRoutingPlan in project voldemort by voldemort.

the class RebalanceBatchPlan method constructBatchPlan.

/**
     * Determine the batch plan and return it. The batch plan has the following
     * properties:
     * 
     * 1) A stealer node does not steal any partition-stores it already hosts.
     * 
     * 2) Use current policy to decide which node to steal from: see getDonorId
     * method.
     * 
     * Currently, this batch plan avoids all unnecessary cross zone moves,
     * distributes cross zone moves into new zones evenly across existing zones,
     * and copies replicaFactor partition-stores into any new zone.
     * 
     * @return the batch plan
     */
private List<RebalanceTaskInfo> constructBatchPlan() {
    // Construct all store routing plans once.
    HashMap<String, StoreRoutingPlan> currentStoreRoutingPlans = new HashMap<String, StoreRoutingPlan>();
    for (StoreDefinition storeDef : currentStoreDefs) {
        currentStoreRoutingPlans.put(storeDef.getName(), new StoreRoutingPlan(currentCluster, storeDef));
    }
    HashMap<String, StoreRoutingPlan> finalStoreRoutingPlans = new HashMap<String, StoreRoutingPlan>();
    for (StoreDefinition storeDef : finalStoreDefs) {
        finalStoreRoutingPlans.put(storeDef.getName(), new StoreRoutingPlan(finalCluster, storeDef));
    }
    RebalanceTaskInfoBuilder rpiBuilder = new RebalanceTaskInfoBuilder();
    // For every node in the final cluster ...
    for (Node stealerNode : finalCluster.getNodes()) {
        int stealerZoneId = stealerNode.getZoneId();
        int stealerNodeId = stealerNode.getId();
        // Consider all store definitions ...
        for (StoreDefinition storeDef : finalStoreDefs) {
            StoreRoutingPlan currentSRP = currentStoreRoutingPlans.get(storeDef.getName());
            StoreRoutingPlan finalSRP = finalStoreRoutingPlans.get(storeDef.getName());
            for (int stealerPartitionId : finalSRP.getZoneNAryPartitionIds(stealerNodeId)) {
                // Optimization for RW stores: Do not steal a partition-store you already host!
                if (!storeDef.getType().equalsIgnoreCase(ReadOnlyStorageConfiguration.TYPE_NAME)) {
                    if (currentSRP.getReplicationNodeList(stealerPartitionId).contains(stealerNodeId)) {
                        continue;
                    }
                }
                // Determine which node to steal from. This will find node that hosts the appropriate zone-nary
                int donorNodeId = getDonorId(currentSRP, finalSRP, stealerZoneId, stealerNodeId, stealerPartitionId);
                // and stealerNodeId are the same, hence this check is needed for RO stores
                if (donorNodeId != stealerNodeId) {
                    rpiBuilder.addPartitionStoreMove(stealerNodeId, donorNodeId, storeDef.getName(), stealerPartitionId);
                }
            }
        }
    }
    return rpiBuilder.buildRebalanceTaskInfos();
}