Search in sources :

Example 1 with ClusterRerouteRequestBuilder

use of org.elasticsearch.action.admin.cluster.reroute.ClusterRerouteRequestBuilder in project elasticsearch by elastic.

the class PrimaryAllocationIT method testForceStaleReplicaToBePromotedToPrimary.

public void testForceStaleReplicaToBePromotedToPrimary() throws Exception {
    // if true, use stale replica, otherwise a completely empty copy
    boolean useStaleReplica = randomBoolean();
    createStaleReplicaScenario();
    logger.info("--> explicitly promote old primary shard");
    final String idxName = "test";
    ImmutableOpenIntMap<List<IndicesShardStoresResponse.StoreStatus>> storeStatuses = client().admin().indices().prepareShardStores(idxName).get().getStoreStatuses().get(idxName);
    ClusterRerouteRequestBuilder rerouteBuilder = client().admin().cluster().prepareReroute();
    for (IntObjectCursor<List<IndicesShardStoresResponse.StoreStatus>> shardStoreStatuses : storeStatuses) {
        int shardId = shardStoreStatuses.key;
        IndicesShardStoresResponse.StoreStatus storeStatus = randomFrom(shardStoreStatuses.value);
        logger.info("--> adding allocation command for shard {}", shardId);
        // force allocation based on node id
        if (useStaleReplica) {
            rerouteBuilder.add(new AllocateStalePrimaryAllocationCommand(idxName, shardId, storeStatus.getNode().getId(), true));
        } else {
            rerouteBuilder.add(new AllocateEmptyPrimaryAllocationCommand(idxName, shardId, storeStatus.getNode().getId(), true));
        }
    }
    rerouteBuilder.get();
    logger.info("--> check that the stale primary shard gets allocated and that documents are available");
    ensureYellow(idxName);
    if (useStaleReplica == false) {
        // When invoking AllocateEmptyPrimaryAllocationCommand, due to the UnassignedInfo.Reason being changed to INDEX_CREATION,
        // its possible that the shard has not completed initialization, even though the cluster health is yellow, so the
        // search can throw an "all shards failed" exception.  We will wait until the shard initialization has completed before
        // verifying the search hit count.
        assertBusy(() -> assertTrue(client().admin().cluster().prepareState().get().getState().routingTable().index(idxName).allPrimaryShardsActive()));
    }
    assertHitCount(client().prepareSearch(idxName).setSize(0).setQuery(matchAllQuery()).get(), useStaleReplica ? 1L : 0L);
    // allocation id of old primary was cleaned from the in-sync set
    ClusterState state = client().admin().cluster().prepareState().get().getState();
    assertEquals(Collections.singleton(state.routingTable().index(idxName).shard(0).primary.allocationId().getId()), state.metaData().index(idxName).inSyncAllocationIds(0));
}
Also used : ClusterState(org.elasticsearch.cluster.ClusterState) IndicesShardStoresResponse(org.elasticsearch.action.admin.indices.shards.IndicesShardStoresResponse) ClusterRerouteRequestBuilder(org.elasticsearch.action.admin.cluster.reroute.ClusterRerouteRequestBuilder) AllocateStalePrimaryAllocationCommand(org.elasticsearch.cluster.routing.allocation.command.AllocateStalePrimaryAllocationCommand) AllocateEmptyPrimaryAllocationCommand(org.elasticsearch.cluster.routing.allocation.command.AllocateEmptyPrimaryAllocationCommand) List(java.util.List)

Example 2 with ClusterRerouteRequestBuilder

use of org.elasticsearch.action.admin.cluster.reroute.ClusterRerouteRequestBuilder in project crate by crate.

the class PartitionedTableConcurrentIntegrationTest method testSelectWhileShardsAreRelocating.

/**
     * Test depends on 2 data nodes
     */
@Test
public void testSelectWhileShardsAreRelocating() throws Throwable {
    execute("create table t (name string, p string) " + "clustered into 2 shards " + "partitioned by (p) with (number_of_replicas = 0)");
    ensureYellow();
    execute("insert into t (name, p) values (?, ?)", new Object[][] { new Object[] { "Marvin", "a" }, new Object[] { "Trillian", "a" } });
    execute("refresh table t");
    execute("set global stats.enabled=true");
    final AtomicReference<Throwable> lastThrowable = new AtomicReference<>();
    final CountDownLatch selects = new CountDownLatch(100);
    Thread t = new Thread(new Runnable() {

        @Override
        public void run() {
            while (selects.getCount() > 0) {
                try {
                    execute("select * from t");
                } catch (Throwable t) {
                    // The failed job should have three started operations
                    SQLResponse res = execute("select id from sys.jobs_log where error is not null order by started desc limit 1");
                    if (res.rowCount() > 0) {
                        String id = (String) res.rows()[0][0];
                        res = execute("select count(*) from sys.operations_log where name=? or name = ?and job_id = ?", new Object[] { "collect", "fetchContext", id });
                        if ((long) res.rows()[0][0] < 3) {
                            // set the error if there where less than three attempts
                            lastThrowable.set(t);
                        }
                    }
                } finally {
                    selects.countDown();
                }
            }
        }
    });
    t.start();
    PartitionName partitionName = new PartitionName("t", Collections.singletonList(new BytesRef("a")));
    final String indexName = partitionName.asIndexName();
    ClusterService clusterService = internalCluster().getInstance(ClusterService.class);
    DiscoveryNodes nodes = clusterService.state().nodes();
    List<String> nodeIds = new ArrayList<>(2);
    for (DiscoveryNode node : nodes) {
        if (node.dataNode()) {
            nodeIds.add(node.getId());
        }
    }
    final Map<String, String> nodeSwap = new HashMap<>(2);
    nodeSwap.put(nodeIds.get(0), nodeIds.get(1));
    nodeSwap.put(nodeIds.get(1), nodeIds.get(0));
    final CountDownLatch relocations = new CountDownLatch(20);
    Thread relocatingThread = new Thread(new Runnable() {

        @Override
        public void run() {
            while (relocations.getCount() > 0) {
                ClusterStateResponse clusterStateResponse = admin().cluster().prepareState().setIndices(indexName).execute().actionGet();
                List<ShardRouting> shardRoutings = clusterStateResponse.getState().routingTable().allShards(indexName);
                ClusterRerouteRequestBuilder clusterRerouteRequestBuilder = admin().cluster().prepareReroute();
                int numMoves = 0;
                for (ShardRouting shardRouting : shardRoutings) {
                    if (shardRouting.currentNodeId() == null) {
                        continue;
                    }
                    if (shardRouting.state() != ShardRoutingState.STARTED) {
                        continue;
                    }
                    String toNode = nodeSwap.get(shardRouting.currentNodeId());
                    clusterRerouteRequestBuilder.add(new MoveAllocationCommand(shardRouting.shardId(), shardRouting.currentNodeId(), toNode));
                    numMoves++;
                }
                if (numMoves > 0) {
                    clusterRerouteRequestBuilder.execute().actionGet();
                    client().admin().cluster().prepareHealth().setWaitForEvents(Priority.LANGUID).setWaitForRelocatingShards(0).setTimeout(ACCEPTABLE_RELOCATION_TIME).execute().actionGet();
                    relocations.countDown();
                }
            }
        }
    });
    relocatingThread.start();
    relocations.await(SQLTransportExecutor.REQUEST_TIMEOUT.getSeconds() + 1, TimeUnit.SECONDS);
    selects.await(SQLTransportExecutor.REQUEST_TIMEOUT.getSeconds() + 1, TimeUnit.SECONDS);
    Throwable throwable = lastThrowable.get();
    if (throwable != null) {
        throw throwable;
    }
    t.join();
    relocatingThread.join();
}
Also used : DiscoveryNode(org.elasticsearch.cluster.node.DiscoveryNode) ClusterRerouteRequestBuilder(org.elasticsearch.action.admin.cluster.reroute.ClusterRerouteRequestBuilder) ClusterStateResponse(org.elasticsearch.action.admin.cluster.state.ClusterStateResponse) MoveAllocationCommand(org.elasticsearch.cluster.routing.allocation.command.MoveAllocationCommand) AtomicReference(java.util.concurrent.atomic.AtomicReference) SQLResponse(io.crate.testing.SQLResponse) CountDownLatch(java.util.concurrent.CountDownLatch) PartitionName(io.crate.metadata.PartitionName) ClusterService(org.elasticsearch.cluster.ClusterService) ShardRouting(org.elasticsearch.cluster.routing.ShardRouting) BytesRef(org.apache.lucene.util.BytesRef) DiscoveryNodes(org.elasticsearch.cluster.node.DiscoveryNodes) Test(org.junit.Test)

Aggregations

ClusterRerouteRequestBuilder (org.elasticsearch.action.admin.cluster.reroute.ClusterRerouteRequestBuilder)2 PartitionName (io.crate.metadata.PartitionName)1 SQLResponse (io.crate.testing.SQLResponse)1 List (java.util.List)1 CountDownLatch (java.util.concurrent.CountDownLatch)1 AtomicReference (java.util.concurrent.atomic.AtomicReference)1 BytesRef (org.apache.lucene.util.BytesRef)1 ClusterStateResponse (org.elasticsearch.action.admin.cluster.state.ClusterStateResponse)1 IndicesShardStoresResponse (org.elasticsearch.action.admin.indices.shards.IndicesShardStoresResponse)1 ClusterService (org.elasticsearch.cluster.ClusterService)1 ClusterState (org.elasticsearch.cluster.ClusterState)1 DiscoveryNode (org.elasticsearch.cluster.node.DiscoveryNode)1 DiscoveryNodes (org.elasticsearch.cluster.node.DiscoveryNodes)1 ShardRouting (org.elasticsearch.cluster.routing.ShardRouting)1 AllocateEmptyPrimaryAllocationCommand (org.elasticsearch.cluster.routing.allocation.command.AllocateEmptyPrimaryAllocationCommand)1 AllocateStalePrimaryAllocationCommand (org.elasticsearch.cluster.routing.allocation.command.AllocateStalePrimaryAllocationCommand)1 MoveAllocationCommand (org.elasticsearch.cluster.routing.allocation.command.MoveAllocationCommand)1 Test (org.junit.Test)1