/**
* This test triggers a corrupt index exception during finalization size if an empty commit point
* is transferred during recovery we don't know the version of the segments_N file because it has
* no segments we can take it from. This simulates recoveries from old indices or even without
* checksums and makes sure if we fail during finalization we also check if the primary is ok.
* Without the relevant checks this test fails with a RED cluster
*/
public void testCorruptionOnNetworkLayerFinalizingRecovery()
throws ExecutionException, InterruptedException, IOException {
internalCluster().ensureAtLeastNumDataNodes(2);
NodesStatsResponse nodeStats = client().admin().cluster().prepareNodesStats().get();
List<NodeStats> dataNodeStats = new ArrayList<>();
for (NodeStats stat : nodeStats.getNodes()) {
if (stat.getNode().isDataNode()) {
dataNodeStats.add(stat);
}
}
assertThat(dataNodeStats.size(), greaterThanOrEqualTo(2));
Collections.shuffle(dataNodeStats, random());
NodeStats primariesNode = dataNodeStats.get(0);
NodeStats unluckyNode = dataNodeStats.get(1);
assertAcked(
prepareCreate("test")
.setSettings(
Settings.builder()
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "0")
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 1)
.put(
"index.routing.allocation.include._name", primariesNode.getNode().getName())
.put(
EnableAllocationDecider.INDEX_ROUTING_REBALANCE_ENABLE_SETTING.getKey(),
EnableAllocationDecider.Rebalance.NONE)
.put("index.allocation.max_retries", Integer.MAX_VALUE) // keep on retrying
));
ensureGreen(); // allocated with empty commit
final AtomicBoolean corrupt = new AtomicBoolean(true);
final CountDownLatch hasCorrupted = new CountDownLatch(1);
for (NodeStats dataNode : dataNodeStats) {
MockTransportService mockTransportService =
((MockTransportService)
internalCluster().getInstance(TransportService.class, dataNode.getNode().getName()));
mockTransportService.addDelegate(
internalCluster().getInstance(TransportService.class, unluckyNode.getNode().getName()),
new MockTransportService.DelegateTransport(mockTransportService.original()) {
@Override
public void sendRequest(
DiscoveryNode node,
long requestId,
String action,
TransportRequest request,
TransportRequestOptions options)
throws IOException, TransportException {
if (corrupt.get() && action.equals(RecoveryTargetService.Actions.FILE_CHUNK)) {
RecoveryFileChunkRequest req = (RecoveryFileChunkRequest) request;
byte[] array = BytesRef.deepCopyOf(req.content().toBytesRef()).bytes;
int i = randomIntBetween(0, req.content().length() - 1);
array[i] = (byte) ~array[i]; // flip one byte in the content
hasCorrupted.countDown();
}
super.sendRequest(node, requestId, action, request, options);
}
});
}
Settings build =
Settings.builder()
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "1")
.put(
"index.routing.allocation.include._name",
primariesNode.getNode().getName() + "," + unluckyNode.getNode().getName())
.build();
client().admin().indices().prepareUpdateSettings("test").setSettings(build).get();
client().admin().cluster().prepareReroute().get();
hasCorrupted.await();
corrupt.set(false);
ensureGreen();
}
/**
* Tests corruption that happens on the network layer and that the primary does not get affected
* by corruption that happens on the way to the replica. The file on disk stays uncorrupted
*/
public void testCorruptionOnNetworkLayer() throws ExecutionException, InterruptedException {
int numDocs = scaledRandomIntBetween(100, 1000);
internalCluster().ensureAtLeastNumDataNodes(2);
if (cluster().numDataNodes() < 3) {
internalCluster()
.startNode(
Settings.builder()
.put(Node.NODE_DATA_SETTING.getKey(), true)
.put(Node.NODE_MASTER_SETTING.getKey(), false));
}
NodesStatsResponse nodeStats = client().admin().cluster().prepareNodesStats().get();
List<NodeStats> dataNodeStats = new ArrayList<>();
for (NodeStats stat : nodeStats.getNodes()) {
if (stat.getNode().isDataNode()) {
dataNodeStats.add(stat);
}
}
assertThat(dataNodeStats.size(), greaterThanOrEqualTo(2));
Collections.shuffle(dataNodeStats, random());
NodeStats primariesNode = dataNodeStats.get(0);
NodeStats unluckyNode = dataNodeStats.get(1);
assertAcked(
prepareCreate("test")
.setSettings(
Settings.builder()
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "0")
.put(
IndexMetaData.SETTING_NUMBER_OF_SHARDS,
between(1, 4)) // don't go crazy here it must recovery fast
// This does corrupt files on the replica, so we can't check:
.put(MockFSIndexStore.INDEX_CHECK_INDEX_ON_CLOSE_SETTING.getKey(), false)
.put(
"index.routing.allocation.include._name", primariesNode.getNode().getName())
.put(
EnableAllocationDecider.INDEX_ROUTING_REBALANCE_ENABLE_SETTING.getKey(),
EnableAllocationDecider.Rebalance.NONE)));
ensureGreen();
IndexRequestBuilder[] builders = new IndexRequestBuilder[numDocs];
for (int i = 0; i < builders.length; i++) {
builders[i] = client().prepareIndex("test", "type").setSource("field", "value");
}
indexRandom(true, builders);
ensureGreen();
assertAllSuccessful(
client()
.admin()
.indices()
.prepareFlush()
.setForce(true)
.setWaitIfOngoing(true)
.execute()
.actionGet());
// we have to flush at least once here since we don't corrupt the translog
SearchResponse countResponse = client().prepareSearch().setSize(0).get();
assertHitCount(countResponse, numDocs);
final boolean truncate = randomBoolean();
for (NodeStats dataNode : dataNodeStats) {
MockTransportService mockTransportService =
((MockTransportService)
internalCluster().getInstance(TransportService.class, dataNode.getNode().getName()));
mockTransportService.addDelegate(
internalCluster().getInstance(TransportService.class, unluckyNode.getNode().getName()),
new MockTransportService.DelegateTransport(mockTransportService.original()) {
@Override
public void sendRequest(
DiscoveryNode node,
long requestId,
String action,
TransportRequest request,
TransportRequestOptions options)
throws IOException, TransportException {
if (action.equals(RecoveryTargetService.Actions.FILE_CHUNK)) {
RecoveryFileChunkRequest req = (RecoveryFileChunkRequest) request;
if (truncate && req.length() > 1) {
BytesRef bytesRef = req.content().toBytesRef();
BytesArray array =
new BytesArray(bytesRef.bytes, bytesRef.offset, (int) req.length() - 1);
request =
new RecoveryFileChunkRequest(
req.recoveryId(),
req.shardId(),
req.metadata(),
req.position(),
array,
req.lastChunk(),
req.totalTranslogOps(),
req.sourceThrottleTimeInNanos());
} else {
assert req.content().toBytesRef().bytes == req.content().toBytesRef().bytes
: "no internal reference!!";
final byte[] array = req.content().toBytesRef().bytes;
int i = randomIntBetween(0, req.content().length() - 1);
array[i] = (byte) ~array[i]; // flip one byte in the content
}
}
super.sendRequest(node, requestId, action, request, options);
}
});
}
Settings build =
Settings.builder()
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "1")
.put("index.routing.allocation.include._name", "*")
.build();
client().admin().indices().prepareUpdateSettings("test").setSettings(build).get();
client().admin().cluster().prepareReroute().get();
ClusterHealthResponse actionGet =
client()
.admin()
.cluster()
.health(Requests.clusterHealthRequest("test").waitForGreenStatus())
.actionGet();
if (actionGet.isTimedOut()) {
logger.info(
"ensureGreen timed out, cluster state:\n{}\n{}",
client().admin().cluster().prepareState().get().getState().prettyPrint(),
client().admin().cluster().preparePendingClusterTasks().get().prettyPrint());
assertThat("timed out waiting for green state", actionGet.isTimedOut(), equalTo(false));
}
// we are green so primaries got not corrupted.
// ensure that no shard is actually allocated on the unlucky node
ClusterStateResponse clusterStateResponse = client().admin().cluster().prepareState().get();
for (IndexShardRoutingTable table :
clusterStateResponse.getState().getRoutingTable().index("test")) {
for (ShardRouting routing : table) {
if (unluckyNode.getNode().getId().equals(routing.currentNodeId())) {
assertThat(routing.state(), not(equalTo(ShardRoutingState.STARTED)));
assertThat(routing.state(), not(equalTo(ShardRoutingState.RELOCATING)));
}
}
}
final int numIterations = scaledRandomIntBetween(5, 20);
for (int i = 0; i < numIterations; i++) {
SearchResponse response = client().prepareSearch().setSize(numDocs).get();
assertHitCount(response, numDocs);
}
}
public void testLimitsRequestSize() throws Exception {
ByteSizeValue inFlightRequestsLimit = new ByteSizeValue(8, ByteSizeUnit.KB);
if (noopBreakerUsed()) {
logger.info("--> noop breakers used, skipping test");
return;
}
internalCluster().ensureAtLeastNumDataNodes(2);
NodesStatsResponse nodeStats = client().admin().cluster().prepareNodesStats().get();
List<NodeStats> dataNodeStats = new ArrayList<>();
for (NodeStats stat : nodeStats.getNodes()) {
if (stat.getNode().isDataNode()) {
dataNodeStats.add(stat);
}
}
assertThat(dataNodeStats.size(), greaterThanOrEqualTo(2));
Collections.shuffle(dataNodeStats, random());
// send bulk request from source node to target node later. The sole shard is bound to the
// target node.
NodeStats targetNode = dataNodeStats.get(0);
NodeStats sourceNode = dataNodeStats.get(1);
assertAcked(
prepareCreate("index")
.setSettings(
Settings.builder()
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0)
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 1)
.put("index.routing.allocation.include._name", targetNode.getNode().getName())
.put(
EnableAllocationDecider.INDEX_ROUTING_REBALANCE_ENABLE_SETTING.getKey(),
EnableAllocationDecider.Rebalance.NONE)));
Client client = client(sourceNode.getNode().getName());
// we use the limit size as a (very) rough indication on how many requests we should sent to hit
// the limit
int numRequests = inFlightRequestsLimit.bytesAsInt();
BulkRequest bulkRequest = new BulkRequest();
for (int i = 0; i < numRequests; i++) {
IndexRequest indexRequest = new IndexRequest("index", "type", Integer.toString(i));
indexRequest.source("field", "value", "num", i);
bulkRequest.add(indexRequest);
}
Settings limitSettings =
Settings.builder()
.put(
HierarchyCircuitBreakerService.IN_FLIGHT_REQUESTS_CIRCUIT_BREAKER_LIMIT_SETTING
.getKey(),
inFlightRequestsLimit)
.build();
assertAcked(
client().admin().cluster().prepareUpdateSettings().setTransientSettings(limitSettings));
// can either fail directly with an exception or the response contains exceptions (depending on
// client)
try {
BulkResponse response = client.bulk(bulkRequest).actionGet();
if (!response.hasFailures()) {
fail("Should have thrown CircuitBreakingException");
} else {
// each item must have failed with CircuitBreakingException
for (BulkItemResponse bulkItemResponse : response) {
Throwable cause = ExceptionsHelper.unwrapCause(bulkItemResponse.getFailure().getCause());
assertThat(cause, instanceOf(CircuitBreakingException.class));
assertEquals(
((CircuitBreakingException) cause).getByteLimit(), inFlightRequestsLimit.bytes());
}
}
} catch (CircuitBreakingException ex) {
assertEquals(ex.getByteLimit(), inFlightRequestsLimit.bytes());
}
}
