Examples with Slice org.apache.solr.common.cloud.Slice used on opensource projects

Example 56 with Slice

use of org.apache.solr.common.cloud.Slice in project lucene-solr by apache.

the class AbstractFullDistribZkTestBase method ensureAllReplicasAreActive.

protected List<Replica> ensureAllReplicasAreActive(String testCollectionName, String shardId, int shards, int rf, int maxWaitSecs) throws Exception {
    final RTimer timer = new RTimer();
    Map<String, Replica> notLeaders = new HashMap<>();
    ZkStateReader zkr = cloudClient.getZkStateReader();
    // force the state to be fresh
    zkr.forceUpdateCollection(testCollectionName);
    ClusterState cs = zkr.getClusterState();
    Collection<Slice> slices = cs.getActiveSlices(testCollectionName);
    assertTrue(slices.size() == shards);
    boolean allReplicasUp = false;
    long waitMs = 0L;
    long maxWaitMs = maxWaitSecs * 1000L;
    Replica leader = null;
    while (waitMs < maxWaitMs && !allReplicasUp) {
        cs = cloudClient.getZkStateReader().getClusterState();
        assertNotNull(cs);
        Slice shard = cs.getSlice(testCollectionName, shardId);
        assertNotNull("No Slice for " + shardId, shard);
        // assume true
        allReplicasUp = true;
        Collection<Replica> replicas = shard.getReplicas();
        assertTrue("Did not find correct number of replicas. Expected:" + rf + " Found:" + replicas.size(), replicas.size() == rf);
        leader = shard.getLeader();
        assertNotNull(leader);
        log.info("Found " + replicas.size() + " replicas and leader on " + leader.getNodeName() + " for " + shardId + " in " + testCollectionName);
        // ensure all replicas are "active" and identify the non-leader replica
        for (Replica replica : replicas) {
            if (replica.getState() != Replica.State.ACTIVE) {
                log.info("Replica {} is currently {}", replica.getName(), replica.getState());
                allReplicasUp = false;
            }
            if (!leader.equals(replica))
                notLeaders.put(replica.getName(), replica);
        }
        if (!allReplicasUp) {
            try {
                Thread.sleep(500L);
            } catch (Exception ignoreMe) {
            }
            waitMs += 500L;
        }
    }
    if (!allReplicasUp)
        fail("Didn't see all replicas for shard " + shardId + " in " + testCollectionName + " come up within " + maxWaitMs + " ms! ClusterState: " + printClusterStateInfo());
    if (notLeaders.isEmpty())
        fail("Didn't isolate any replicas that are not the leader! ClusterState: " + printClusterStateInfo());
    log.info("Took {} ms to see all replicas become active.", timer.getTime());
    List<Replica> replicas = new ArrayList<>();
    replicas.addAll(notLeaders.values());
    return replicas;
}

Example 57 with Slice

use of org.apache.solr.common.cloud.Slice in project lucene-solr by apache.

the class AbstractFullDistribZkTestBase method checkCollectionExpectations.

private String checkCollectionExpectations(String collectionName, List<Integer> numShardsNumReplicaList, List<String> nodesAllowedToRunShards) {
    ClusterState clusterState = getCommonCloudSolrClient().getZkStateReader().getClusterState();
    int expectedSlices = numShardsNumReplicaList.get(0);
    // The Math.min thing is here, because we expect replication-factor to be reduced to if there are not enough live nodes to spread all shards of a collection over different nodes
    int expectedShardsPerSlice = numShardsNumReplicaList.get(1);
    int expectedTotalShards = expectedSlices * expectedShardsPerSlice;
    //          .getCollectionStates();
    if (clusterState.hasCollection(collectionName)) {
        Map<String, Slice> slices = clusterState.getCollection(collectionName).getSlicesMap();
        // did we find expectedSlices slices/shards?
        if (slices.size() != expectedSlices) {
            return "Found new collection " + collectionName + ", but mismatch on number of slices. Expected: " + expectedSlices + ", actual: " + slices.size();
        }
        int totalShards = 0;
        for (String sliceName : slices.keySet()) {
            for (Replica replica : slices.get(sliceName).getReplicas()) {
                if (nodesAllowedToRunShards != null && !nodesAllowedToRunShards.contains(replica.getStr(ZkStateReader.NODE_NAME_PROP))) {
                    return "Shard " + replica.getName() + " created on node " + replica.getNodeName() + " not allowed to run shards for the created collection " + collectionName;
                }
            }
            totalShards += slices.get(sliceName).getReplicas().size();
        }
        if (totalShards != expectedTotalShards) {
            return "Found new collection " + collectionName + " with correct number of slices, but mismatch on number of shards. Expected: " + expectedTotalShards + ", actual: " + totalShards;
        }
        return null;
    } else {
        return "Could not find new collection " + collectionName;
    }
}

Example 58 with Slice

use of org.apache.solr.common.cloud.Slice in project lucene-solr by apache.

the class AbstractFullDistribZkTestBase method assertDocCounts.

protected void assertDocCounts(boolean verbose) throws Exception {
    // and node/client to shard?
    if (verbose)
        System.err.println("control docs:" + controlClient.query(new SolrQuery("*:*")).getResults().getNumFound() + "\n\n");
    long controlCount = controlClient.query(new SolrQuery("*:*")).getResults().getNumFound();
    // do some really inefficient mapping...
    Map<String, Slice> slices = null;
    ClusterState clusterState;
    try (ZkStateReader zk = new ZkStateReader(zkServer.getZkAddress(), AbstractZkTestCase.TIMEOUT, AbstractZkTestCase.TIMEOUT)) {
        zk.createClusterStateWatchersAndUpdate();
        clusterState = zk.getClusterState();
        slices = clusterState.getSlicesMap(DEFAULT_COLLECTION);
    }
    if (slices == null) {
        throw new RuntimeException("Could not find collection " + DEFAULT_COLLECTION + " in " + clusterState.getCollectionsMap().keySet());
    }
    for (CloudJettyRunner cjetty : cloudJettys) {
        CloudSolrServerClient client = cjetty.client;
        for (Map.Entry<String, Slice> slice : slices.entrySet()) {
            Map<String, Replica> theShards = slice.getValue().getReplicasMap();
            for (Map.Entry<String, Replica> shard : theShards.entrySet()) {
                String shardName = new URI(((HttpSolrClient) client.solrClient).getBaseURL()).getPort() + "_solr_";
                if (verbose && shard.getKey().endsWith(shardName)) {
                    System.err.println("shard:" + slice.getKey());
                    System.err.println(shard.getValue());
                }
            }
        }
        ZkStateReader zkStateReader = cloudClient.getZkStateReader();
        long count = 0;
        final Replica.State currentState = Replica.State.getState(cjetty.info.getStr(ZkStateReader.STATE_PROP));
        if (currentState == Replica.State.ACTIVE && zkStateReader.getClusterState().liveNodesContain(cjetty.info.getStr(ZkStateReader.NODE_NAME_PROP))) {
            SolrQuery query = new SolrQuery("*:*");
            query.set("distrib", false);
            count = client.solrClient.query(query).getResults().getNumFound();
        }
        if (verbose)
            System.err.println("client docs:" + count + "\n\n");
    }
    if (verbose)
        System.err.println("control docs:" + controlClient.query(new SolrQuery("*:*")).getResults().getNumFound() + "\n\n");
    SolrQuery query = new SolrQuery("*:*");
    assertEquals("Doc Counts do not add up", controlCount, cloudClient.query(query).getResults().getNumFound());
}

Example 59 with Slice

use of org.apache.solr.common.cloud.Slice in project lucene-solr by apache.

the class AbstractFullDistribZkTestBase method waitForReplicationFromReplicas.

protected void waitForReplicationFromReplicas(String collectionName, ZkStateReader zkStateReader, TimeOut timeout) throws KeeperException, InterruptedException, IOException {
    zkStateReader.forceUpdateCollection(collectionName);
    DocCollection collection = zkStateReader.getClusterState().getCollection(collectionName);
    Map<String, CoreContainer> containers = new HashMap<>();
    for (JettySolrRunner runner : jettys) {
        if (!runner.isRunning()) {
            continue;
        }
        containers.put(runner.getNodeName(), runner.getCoreContainer());
    }
    for (Slice s : collection.getSlices()) {
        Replica leader = s.getLeader();
        long leaderIndexVersion = -1;
        while (!timeout.hasTimedOut()) {
            leaderIndexVersion = getIndexVersion(leader);
            if (leaderIndexVersion >= 0) {
                break;
            }
            Thread.sleep(1000);
        }
        if (timeout.hasTimedOut()) {
            fail("Unable to get leader indexVersion");
        }
        for (Replica pullReplica : s.getReplicas(EnumSet.of(Replica.Type.PULL, Replica.Type.TLOG))) {
            if (!zkStateReader.getClusterState().liveNodesContain(pullReplica.getNodeName())) {
                continue;
            }
            while (true) {
                long replicaIndexVersion = getIndexVersion(pullReplica);
                if (leaderIndexVersion == replicaIndexVersion) {
                    log.debug("Leader replica's version ({}) in sync with replica({}): {} == {}", leader.getName(), pullReplica.getName(), leaderIndexVersion, replicaIndexVersion);
                    // Make sure the host is serving the correct version
                    try (SolrCore core = containers.get(pullReplica.getNodeName()).getCore(pullReplica.getCoreName())) {
                        RefCounted<SolrIndexSearcher> ref = core.getRegisteredSearcher();
                        try {
                            SolrIndexSearcher searcher = ref.get();
                            String servingVersion = searcher.getIndexReader().getIndexCommit().getUserData().get(SolrIndexWriter.COMMIT_TIME_MSEC_KEY);
                            if (Long.parseLong(servingVersion) == replicaIndexVersion) {
                                break;
                            } else {
                                log.debug("Replica {} has the correct version replicated, but the searcher is not ready yet. Replicated version: {}, Serving version: {}", pullReplica.getName(), replicaIndexVersion, servingVersion);
                            }
                        } finally {
                            if (ref != null)
                                ref.decref();
                        }
                    }
                } else {
                    if (timeout.hasTimedOut()) {
                        logReplicaTypesReplicationInfo(collectionName, zkStateReader);
                        fail(String.format(Locale.ROOT, "Timed out waiting for replica %s (%d) to replicate from leader %s (%d)", pullReplica.getName(), replicaIndexVersion, leader.getName(), leaderIndexVersion));
                    }
                    if (leaderIndexVersion > replicaIndexVersion) {
                        log.debug("{} version is {} and leader's is {}, will wait for replication", pullReplica.getName(), replicaIndexVersion, leaderIndexVersion);
                    } else {
                        log.debug("Leader replica's version ({}) is lower than pull replica({}): {} < {}", leader.getName(), pullReplica.getName(), leaderIndexVersion, replicaIndexVersion);
                    }
                }
                Thread.sleep(1000);
            }
        }
    }
}

Example 60 with Slice

use of org.apache.solr.common.cloud.Slice in project lucene-solr by apache.

the class AbstractFullDistribZkTestBase method updateMappingsFromZk.

protected void updateMappingsFromZk(List<JettySolrRunner> jettys, List<SolrClient> clients, boolean allowOverSharding) throws Exception {
    ZkStateReader zkStateReader = cloudClient.getZkStateReader();
    zkStateReader.forceUpdateCollection(DEFAULT_COLLECTION);
    cloudJettys.clear();
    shardToJetty.clear();
    ClusterState clusterState = zkStateReader.getClusterState();
    DocCollection coll = clusterState.getCollection(DEFAULT_COLLECTION);
    List<CloudSolrServerClient> theClients = new ArrayList<>();
    for (SolrClient client : clients) {
        // find info for this client in zk 
        nextClient: // we find out state by simply matching ports...
        for (Slice slice : coll.getSlices()) {
            for (Replica replica : slice.getReplicas()) {
                int port = new URI(((HttpSolrClient) client).getBaseURL()).getPort();
                if (replica.getStr(ZkStateReader.BASE_URL_PROP).contains(":" + port)) {
                    CloudSolrServerClient csc = new CloudSolrServerClient();
                    csc.solrClient = client;
                    csc.port = port;
                    csc.shardName = replica.getStr(ZkStateReader.NODE_NAME_PROP);
                    csc.info = replica;
                    theClients.add(csc);
                    break nextClient;
                }
            }
        }
    }
    for (JettySolrRunner jetty : jettys) {
        int port = jetty.getLocalPort();
        if (port == -1) {
            throw new RuntimeException("Cannot find the port for jetty");
        }
        nextJetty: for (Slice slice : coll.getSlices()) {
            Set<Entry<String, Replica>> entries = slice.getReplicasMap().entrySet();
            for (Entry<String, Replica> entry : entries) {
                Replica replica = entry.getValue();
                if (replica.getStr(ZkStateReader.BASE_URL_PROP).contains(":" + port)) {
                    List<CloudJettyRunner> list = shardToJetty.get(slice.getName());
                    if (list == null) {
                        list = new ArrayList<>();
                        shardToJetty.put(slice.getName(), list);
                    }
                    boolean isLeader = slice.getLeader() == replica;
                    CloudJettyRunner cjr = new CloudJettyRunner();
                    cjr.jetty = jetty;
                    cjr.info = replica;
                    cjr.nodeName = replica.getStr(ZkStateReader.NODE_NAME_PROP);
                    cjr.coreNodeName = entry.getKey();
                    cjr.url = replica.getStr(ZkStateReader.BASE_URL_PROP) + "/" + replica.getStr(ZkStateReader.CORE_NAME_PROP);
                    cjr.client = findClientByPort(port, theClients);
                    list.add(cjr);
                    if (isLeader) {
                        shardToLeaderJetty.put(slice.getName(), cjr);
                    }
                    cloudJettys.add(cjr);
                    break nextJetty;
                }
            }
        }
    }
    // running jetty though
    for (Slice slice : coll.getSlices()) {
        // check that things look right
        List<CloudJettyRunner> jetties = shardToJetty.get(slice.getName());
        if (!allowOverSharding) {
            assertNotNull("Test setup problem: We found no jetties for shard: " + slice.getName() + " just:" + shardToJetty.keySet(), jetties);
            assertEquals("slice:" + slice.getName(), slice.getReplicas().size(), jetties.size());
        }
    }
}