Examples with IndexShardNotStartedException org.elasticsearch.index.shard.IndexShardNotStartedException used on opensource projects

Example 1 with IndexShardNotStartedException

use of org.elasticsearch.index.shard.IndexShardNotStartedException in project elasticsearch by elastic.

the class ReplicationOperationTests method testReplication.

public void testReplication() throws Exception {
    final String index = "test";
    final ShardId shardId = new ShardId(index, "_na_", 0);
    ClusterState state = stateWithActivePrimary(index, true, randomInt(5));
    IndexMetaData indexMetaData = state.getMetaData().index(index);
    final long primaryTerm = indexMetaData.primaryTerm(0);
    final IndexShardRoutingTable indexShardRoutingTable = state.getRoutingTable().shardRoutingTable(shardId);
    ShardRouting primaryShard = indexShardRoutingTable.primaryShard();
    if (primaryShard.relocating() && randomBoolean()) {
        // simulate execution of the replication phase on the relocation target node after relocation source was marked as relocated
        state = ClusterState.builder(state).nodes(DiscoveryNodes.builder(state.nodes()).localNodeId(primaryShard.relocatingNodeId())).build();
        primaryShard = primaryShard.getTargetRelocatingShard();
    }
    // add a few in-sync allocation ids that don't have corresponding routing entries
    Set<String> staleAllocationIds = Sets.newHashSet(generateRandomStringArray(4, 10, false));
    state = ClusterState.builder(state).metaData(MetaData.builder(state.metaData()).put(IndexMetaData.builder(indexMetaData).putInSyncAllocationIds(0, Sets.union(indexMetaData.inSyncAllocationIds(0), staleAllocationIds)))).build();
    final Set<ShardRouting> expectedReplicas = getExpectedReplicas(shardId, state);
    final Map<ShardRouting, Exception> expectedFailures = new HashMap<>();
    final Set<ShardRouting> expectedFailedShards = new HashSet<>();
    for (ShardRouting replica : expectedReplicas) {
        if (randomBoolean()) {
            Exception t;
            boolean criticalFailure = randomBoolean();
            if (criticalFailure) {
                t = new CorruptIndexException("simulated", (String) null);
            } else {
                t = new IndexShardNotStartedException(shardId, IndexShardState.RECOVERING);
            }
            logger.debug("--> simulating failure on {} with [{}]", replica, t.getClass().getSimpleName());
            expectedFailures.put(replica, t);
            if (criticalFailure) {
                expectedFailedShards.add(replica);
            }
        }
    }
    Request request = new Request(shardId);
    PlainActionFuture<TestPrimary.Result> listener = new PlainActionFuture<>();
    final ClusterState finalState = state;
    final TestReplicaProxy replicasProxy = new TestReplicaProxy(expectedFailures);
    final TestPrimary primary = new TestPrimary(primaryShard, primaryTerm);
    final TestReplicationOperation op = new TestReplicationOperation(request, primary, listener, replicasProxy, () -> finalState);
    op.execute();
    assertThat(request.primaryTerm(), equalTo(primaryTerm));
    assertThat("request was not processed on primary", request.processedOnPrimary.get(), equalTo(true));
    assertThat(request.processedOnReplicas, equalTo(expectedReplicas));
    assertThat(replicasProxy.failedReplicas, equalTo(expectedFailedShards));
    assertThat(replicasProxy.markedAsStaleCopies, equalTo(staleAllocationIds));
    assertTrue("listener is not marked as done", listener.isDone());
    ShardInfo shardInfo = listener.actionGet().getShardInfo();
    assertThat(shardInfo.getFailed(), equalTo(expectedFailedShards.size()));
    assertThat(shardInfo.getFailures(), arrayWithSize(expectedFailedShards.size()));
    assertThat(shardInfo.getSuccessful(), equalTo(1 + expectedReplicas.size() - expectedFailures.size()));
    final List<ShardRouting> unassignedShards = indexShardRoutingTable.shardsWithState(ShardRoutingState.UNASSIGNED);
    final int totalShards = 1 + expectedReplicas.size() + unassignedShards.size();
    assertThat(shardInfo.getTotal(), equalTo(totalShards));
    assertThat(primary.knownLocalCheckpoints.remove(primaryShard.allocationId().getId()), equalTo(primary.localCheckpoint));
    assertThat(primary.knownLocalCheckpoints, equalTo(replicasProxy.generatedLocalCheckpoints));
}

Example 2 with IndexShardNotStartedException

use of org.elasticsearch.index.shard.IndexShardNotStartedException in project crate by crate.

the class ReplicationOperationTests method testReplication.

@Test
public void testReplication() throws Exception {
    final String index = "test";
    final ShardId shardId = new ShardId(index, "_na_", 0);
    ClusterState initialState = stateWithActivePrimary(index, true, randomInt(5));
    IndexMetadata indexMetadata = initialState.getMetadata().index(index);
    final long primaryTerm = indexMetadata.primaryTerm(0);
    final IndexShardRoutingTable indexShardRoutingTable = initialState.getRoutingTable().shardRoutingTable(shardId);
    ShardRouting primaryShard = indexShardRoutingTable.primaryShard();
    if (primaryShard.relocating() && randomBoolean()) {
        // simulate execution of the replication phase on the relocation target node after relocation source was marked as relocated
        initialState = ClusterState.builder(initialState).nodes(DiscoveryNodes.builder(initialState.nodes()).localNodeId(primaryShard.relocatingNodeId())).build();
        primaryShard = primaryShard.getTargetRelocatingShard();
    }
    // add a few in-sync allocation ids that don't have corresponding routing entries
    final Set<String> staleAllocationIds = Set.of(generateRandomStringArray(4, 10, false));
    final Set<String> inSyncAllocationIds = Sets.union(indexMetadata.inSyncAllocationIds(0), staleAllocationIds);
    final Set<String> trackedShards = new HashSet<>();
    final Set<String> untrackedShards = new HashSet<>();
    addTrackingInfo(indexShardRoutingTable, primaryShard, trackedShards, untrackedShards);
    trackedShards.addAll(staleAllocationIds);
    final ReplicationGroup replicationGroup = new ReplicationGroup(indexShardRoutingTable, inSyncAllocationIds, trackedShards);
    final Set<ShardRouting> expectedReplicas = getExpectedReplicas(shardId, initialState, trackedShards);
    final Map<ShardRouting, Exception> simulatedFailures = new HashMap<>();
    final Map<ShardRouting, Exception> reportedFailures = new HashMap<>();
    for (ShardRouting replica : expectedReplicas) {
        if (randomBoolean()) {
            Exception t;
            boolean criticalFailure = randomBoolean();
            if (criticalFailure) {
                t = new CorruptIndexException("simulated", (String) null);
                reportedFailures.put(replica, t);
            } else {
                t = new IndexShardNotStartedException(shardId, IndexShardState.RECOVERING);
            }
            logger.debug("--> simulating failure on {} with [{}]", replica, t.getClass().getSimpleName());
            simulatedFailures.put(replica, t);
        }
    }
    Request request = new Request(shardId);
    PlainActionFuture<TestPrimary.Result> listener = new PlainActionFuture<>();
    final TestReplicaProxy replicasProxy = new TestReplicaProxy(simulatedFailures);
    final TestPrimary primary = new TestPrimary(primaryShard, () -> replicationGroup);
    final TestReplicationOperation op = new TestReplicationOperation(request, primary, listener, replicasProxy, primaryTerm);
    op.execute();
    assertThat("request was not processed on primary", request.processedOnPrimary.get(), equalTo(true));
    assertThat(request.processedOnReplicas, equalTo(expectedReplicas));
    assertThat(replicasProxy.failedReplicas, equalTo(simulatedFailures.keySet()));
    assertThat(replicasProxy.markedAsStaleCopies, equalTo(staleAllocationIds));
    assertThat("post replication operations not run on primary", request.runPostReplicationActionsOnPrimary.get(), equalTo(true));
    assertTrue("listener is not marked as done", listener.isDone());
    ShardInfo shardInfo = listener.actionGet().getShardInfo();
    assertThat(shardInfo.getFailed(), equalTo(reportedFailures.size()));
    assertThat(shardInfo.getFailures(), arrayWithSize(reportedFailures.size()));
    assertThat(shardInfo.getSuccessful(), equalTo(1 + expectedReplicas.size() - simulatedFailures.size()));
    final List<ShardRouting> unassignedShards = indexShardRoutingTable.shardsWithState(ShardRoutingState.UNASSIGNED);
    final int totalShards = 1 + expectedReplicas.size() + unassignedShards.size() + untrackedShards.size();
    assertThat(replicationGroup.toString(), shardInfo.getTotal(), equalTo(totalShards));
    assertThat(primary.knownLocalCheckpoints.remove(primaryShard.allocationId().getId()), equalTo(primary.localCheckpoint));
    assertThat(primary.knownLocalCheckpoints, equalTo(replicasProxy.generatedLocalCheckpoints));
    assertThat(primary.knownGlobalCheckpoints.remove(primaryShard.allocationId().getId()), equalTo(primary.globalCheckpoint));
    assertThat(primary.knownGlobalCheckpoints, equalTo(replicasProxy.generatedGlobalCheckpoints));
}