use of org.elasticsearch.common.unit.ByteSizeValue in project elasticsearch by elastic.
the class GoogleCloudStorageBlobStoreRepositoryTests method testChunkSize.
public void testChunkSize() {
// default chunk size
RepositoryMetaData repositoryMetaData = new RepositoryMetaData("repo", GoogleCloudStorageRepository.TYPE, Settings.EMPTY);
ByteSizeValue chunkSize = GoogleCloudStorageRepository.getSetting(GoogleCloudStorageRepository.CHUNK_SIZE, repositoryMetaData);
assertEquals(GoogleCloudStorageRepository.MAX_CHUNK_SIZE, chunkSize);
// chunk size in settings
int size = randomIntBetween(1, 100);
repositoryMetaData = new RepositoryMetaData("repo", GoogleCloudStorageRepository.TYPE, Settings.builder().put("chunk_size", size + "mb").build());
chunkSize = GoogleCloudStorageRepository.getSetting(GoogleCloudStorageRepository.CHUNK_SIZE, repositoryMetaData);
assertEquals(new ByteSizeValue(size, ByteSizeUnit.MB), chunkSize);
// zero bytes is not allowed
IllegalArgumentException e = expectThrows(IllegalArgumentException.class, () -> {
RepositoryMetaData repoMetaData = new RepositoryMetaData("repo", GoogleCloudStorageRepository.TYPE, Settings.builder().put("chunk_size", "0").build());
GoogleCloudStorageRepository.getSetting(GoogleCloudStorageRepository.CHUNK_SIZE, repoMetaData);
});
assertEquals("Failed to parse value [0] for setting [chunk_size] must be >= 1b", e.getMessage());
// negative bytes not allowed
e = expectThrows(IllegalArgumentException.class, () -> {
RepositoryMetaData repoMetaData = new RepositoryMetaData("repo", GoogleCloudStorageRepository.TYPE, Settings.builder().put("chunk_size", "-1").build());
GoogleCloudStorageRepository.getSetting(GoogleCloudStorageRepository.CHUNK_SIZE, repoMetaData);
});
assertEquals("Failed to parse value [-1] for setting [chunk_size] must be >= 1b", e.getMessage());
// greater than max chunk size not allowed
e = expectThrows(IllegalArgumentException.class, () -> {
RepositoryMetaData repoMetaData = new RepositoryMetaData("repo", GoogleCloudStorageRepository.TYPE, Settings.builder().put("chunk_size", "101mb").build());
GoogleCloudStorageRepository.getSetting(GoogleCloudStorageRepository.CHUNK_SIZE, repoMetaData);
});
assertEquals("Failed to parse value [101mb] for setting [chunk_size] must be <= 100mb", e.getMessage());
}
use of org.elasticsearch.common.unit.ByteSizeValue in project elasticsearch by elastic.
the class ZenFaultDetectionTests method setUp.
@Override
@Before
public void setUp() throws Exception {
super.setUp();
Settings settings = Settings.builder().put(HierarchyCircuitBreakerService.IN_FLIGHT_REQUESTS_CIRCUIT_BREAKER_LIMIT_SETTING.getKey(), new ByteSizeValue(0)).build();
ClusterSettings clusterSettings = new ClusterSettings(settings, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS);
threadPool = new TestThreadPool(getClass().getName());
circuitBreakerService = new HierarchyCircuitBreakerService(settings, clusterSettings);
settingsA = Settings.builder().put("node.name", "TS_A").put(settings).build();
serviceA = build(settingsA, version0);
nodeA = serviceA.getLocalDiscoNode();
settingsB = Settings.builder().put("node.name", "TS_B").put(settings).build();
serviceB = build(settingsB, version1);
nodeB = serviceB.getLocalDiscoNode();
clusterServiceA = createClusterService(settingsA, threadPool, nodeA);
clusterServiceB = createClusterService(settingsB, threadPool, nodeB);
// wait till all nodes are properly connected and the event has been sent, so tests in this class
// will not get this callback called on the connections done in this setup
final CountDownLatch latch = new CountDownLatch(2);
TransportConnectionListener waitForConnection = new TransportConnectionListener() {
@Override
public void onNodeConnected(DiscoveryNode node) {
latch.countDown();
}
@Override
public void onNodeDisconnected(DiscoveryNode node) {
fail("disconnect should not be called " + node);
}
};
serviceA.addConnectionListener(waitForConnection);
serviceB.addConnectionListener(waitForConnection);
serviceA.connectToNode(nodeB);
serviceA.connectToNode(nodeA);
serviceB.connectToNode(nodeA);
serviceB.connectToNode(nodeB);
assertThat("failed to wait for all nodes to connect", latch.await(5, TimeUnit.SECONDS), equalTo(true));
serviceA.removeConnectionListener(waitForConnection);
serviceB.removeConnectionListener(waitForConnection);
}
use of org.elasticsearch.common.unit.ByteSizeValue in project elasticsearch by elastic.
the class IndexWithShadowReplicasIT method testIndexWithFewDocuments.
@TestLogging("org.elasticsearch.gateway:TRACE")
public void testIndexWithFewDocuments() throws Exception {
final Path dataPath = createTempDir();
Settings nodeSettings = nodeSettings(dataPath);
internalCluster().startNodes(3, nodeSettings);
final String IDX = "test";
Settings idxSettings = Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 2).put(IndexSettings.INDEX_TRANSLOG_FLUSH_THRESHOLD_SIZE_SETTING.getKey(), new ByteSizeValue(1, ByteSizeUnit.PB)).put(IndexMetaData.SETTING_DATA_PATH, dataPath.toAbsolutePath().toString()).put(IndexMetaData.SETTING_SHADOW_REPLICAS, true).put(IndexMetaData.SETTING_SHARED_FILESYSTEM, true).build();
prepareCreate(IDX).setSettings(idxSettings).addMapping("doc", "foo", "type=text").get();
ensureGreen(IDX);
// So basically, the primary should fail and the replica will need to
// replay the translog, this is what this tests
client().prepareIndex(IDX, "doc", "1").setSource("foo", "bar").get();
client().prepareIndex(IDX, "doc", "2").setSource("foo", "bar").get();
IndicesStatsResponse indicesStatsResponse = client().admin().indices().prepareStats(IDX).clear().setTranslog(true).get();
assertEquals(2, indicesStatsResponse.getIndex(IDX).getPrimaries().getTranslog().estimatedNumberOfOperations());
assertEquals(2, indicesStatsResponse.getIndex(IDX).getTotal().getTranslog().estimatedNumberOfOperations());
Index index = resolveIndex(IDX);
for (IndicesService service : internalCluster().getInstances(IndicesService.class)) {
IndexService indexService = service.indexService(index);
if (indexService != null) {
IndexShard shard = indexService.getShard(0);
TranslogStats translogStats = shard.translogStats();
assertTrue(translogStats != null || shard instanceof ShadowIndexShard);
if (translogStats != null) {
assertEquals(2, translogStats.estimatedNumberOfOperations());
}
}
}
// Check that we can get doc 1 and 2, because we are doing realtime
// gets and getting from the primary
GetResponse gResp1 = client().prepareGet(IDX, "doc", "1").get();
GetResponse gResp2 = client().prepareGet(IDX, "doc", "2").get();
assertThat(gResp1.getSource().get("foo"), equalTo("bar"));
assertThat(gResp2.getSource().get("foo"), equalTo("bar"));
flushAndRefresh(IDX);
client().prepareIndex(IDX, "doc", "3").setSource("foo", "bar").get();
client().prepareIndex(IDX, "doc", "4").setSource("foo", "bar").get();
refresh();
// Check that we can get doc 1 and 2 without realtime
gResp1 = client().prepareGet(IDX, "doc", "1").setRealtime(false).get();
gResp2 = client().prepareGet(IDX, "doc", "2").setRealtime(false).get();
assertThat(gResp1.getSource().get("foo"), equalTo("bar"));
assertThat(gResp2.getSource().get("foo"), equalTo("bar"));
logger.info("--> restarting all nodes");
if (randomBoolean()) {
logger.info("--> rolling restart");
internalCluster().rollingRestart();
} else {
logger.info("--> full restart");
internalCluster().fullRestart();
}
client().admin().cluster().prepareHealth().setWaitForNodes("3").get();
ensureGreen(IDX);
flushAndRefresh(IDX);
logger.info("--> performing query");
SearchResponse resp = client().prepareSearch(IDX).setQuery(matchAllQuery()).get();
assertHitCount(resp, 4);
logger.info("--> deleting index");
assertAcked(client().admin().indices().prepareDelete(IDX));
}
use of org.elasticsearch.common.unit.ByteSizeValue in project elasticsearch by elastic.
the class CorruptedFileIT method testReplicaCorruption.
/**
* This test verifies that if we corrupt a replica, we can still get to green, even though
* listing its store fails. Note, we need to make sure that replicas are allocated on all data
* nodes, so that replica won't be sneaky and allocated on a node that doesn't have a corrupted
* replica.
*/
public void testReplicaCorruption() throws Exception {
int numDocs = scaledRandomIntBetween(100, 1000);
internalCluster().ensureAtLeastNumDataNodes(2);
assertAcked(prepareCreate("test").setSettings(Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, cluster().numDataNodes() - 1).put(MergePolicyConfig.INDEX_MERGE_ENABLED, false).put(MockFSIndexStore.INDEX_CHECK_INDEX_ON_CLOSE_SETTING.getKey(), // no checkindex - we corrupt shards on purpose
false).put(IndexSettings.INDEX_TRANSLOG_FLUSH_THRESHOLD_SIZE_SETTING.getKey(), // no translog based flush - it might change the .liv / segments.N files
new ByteSizeValue(1, ByteSizeUnit.PB))));
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).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);
// disable allocations of replicas post restart (the restart will change replicas to primaries, so we have
// to capture replicas post restart)
assertAcked(client().admin().cluster().prepareUpdateSettings().setPersistentSettings(Settings.builder().put(EnableAllocationDecider.CLUSTER_ROUTING_ALLOCATION_ENABLE_SETTING.getKey(), "primaries")));
internalCluster().fullRestart();
ensureYellow();
final Index index = resolveIndex("test");
final IndicesShardStoresResponse stores = client().admin().indices().prepareShardStores(index.getName()).get();
for (IntObjectCursor<List<IndicesShardStoresResponse.StoreStatus>> shards : stores.getStoreStatuses().get(index.getName())) {
for (IndicesShardStoresResponse.StoreStatus store : shards.value) {
final ShardId shardId = new ShardId(index, shards.key);
if (store.getAllocationStatus().equals(IndicesShardStoresResponse.StoreStatus.AllocationStatus.UNUSED)) {
for (Path path : findFilesToCorruptOnNode(store.getNode().getName(), shardId)) {
try (OutputStream os = Files.newOutputStream(path)) {
os.write(0);
}
logger.info("corrupting file {} on node {}", path, store.getNode().getName());
}
}
}
}
// enable allocation
assertAcked(client().admin().cluster().prepareUpdateSettings().setPersistentSettings(Settings.builder().putNull(EnableAllocationDecider.CLUSTER_ROUTING_ALLOCATION_ENABLE_SETTING.getKey())));
ensureGreen();
}
use of org.elasticsearch.common.unit.ByteSizeValue in project elasticsearch by elastic.
the class CorruptedFileIT method testCorruptFileAndRecover.
/**
* Tests that we can actually recover from a corruption on the primary given that we have replica shards around.
*/
public void testCorruptFileAndRecover() throws ExecutionException, InterruptedException, IOException {
int numDocs = scaledRandomIntBetween(100, 1000);
// have enough space for 3 copies
internalCluster().ensureAtLeastNumDataNodes(3);
if (cluster().numDataNodes() == 3) {
logger.info("--> cluster has [3] data nodes, corrupted primary will be overwritten");
}
assertThat(cluster().numDataNodes(), greaterThanOrEqualTo(3));
assertAcked(prepareCreate("test").setSettings(Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, "1").put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "1").put(MergePolicyConfig.INDEX_MERGE_ENABLED, false).put(MockFSIndexStore.INDEX_CHECK_INDEX_ON_CLOSE_SETTING.getKey(), // no checkindex - we corrupt shards on purpose
false).put(IndexSettings.INDEX_TRANSLOG_FLUSH_THRESHOLD_SIZE_SETTING.getKey(), // no translog based flush - it might change the .liv / segments.N files
new ByteSizeValue(1, ByteSizeUnit.PB))));
ensureGreen();
disableAllocation("test");
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).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 int numShards = numShards("test");
ShardRouting corruptedShardRouting = corruptRandomPrimaryFile();
logger.info("--> {} corrupted", corruptedShardRouting);
enableAllocation("test");
/*
* we corrupted the primary shard - now lets make sure we never recover from it successfully
*/
Settings build = Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, "2").build();
client().admin().indices().prepareUpdateSettings("test").setSettings(build).get();
ClusterHealthResponse health = client().admin().cluster().health(Requests.clusterHealthRequest("test").waitForGreenStatus().timeout(// sometimes due to cluster rebalacing and random settings default timeout is just not enough.
"5m").waitForNoRelocatingShards(true)).actionGet();
if (health.isTimedOut()) {
logger.info("cluster state:\n{}\n{}", client().admin().cluster().prepareState().get().getState(), client().admin().cluster().preparePendingClusterTasks().get());
assertThat("timed out waiting for green state", health.isTimedOut(), equalTo(false));
}
assertThat(health.getStatus(), equalTo(ClusterHealthStatus.GREEN));
final int numIterations = scaledRandomIntBetween(5, 20);
for (int i = 0; i < numIterations; i++) {
SearchResponse response = client().prepareSearch().setSize(numDocs).get();
assertHitCount(response, numDocs);
}
/*
* now hook into the IndicesService and register a close listener to
* run the checkindex. if the corruption is still there we will catch it.
*/
// primary + 2 replicas
final CountDownLatch latch = new CountDownLatch(numShards * 3);
final CopyOnWriteArrayList<Exception> exception = new CopyOnWriteArrayList<>();
final IndexEventListener listener = new IndexEventListener() {
@Override
public void afterIndexShardClosed(ShardId sid, @Nullable IndexShard indexShard, Settings indexSettings) {
if (indexShard != null) {
Store store = indexShard.store();
store.incRef();
try {
if (!Lucene.indexExists(store.directory()) && indexShard.state() == IndexShardState.STARTED) {
return;
}
try (CheckIndex checkIndex = new CheckIndex(store.directory())) {
BytesStreamOutput os = new BytesStreamOutput();
PrintStream out = new PrintStream(os, false, StandardCharsets.UTF_8.name());
checkIndex.setInfoStream(out);
out.flush();
CheckIndex.Status status = checkIndex.checkIndex();
if (!status.clean) {
logger.warn("check index [failure]\n{}", os.bytes().utf8ToString());
throw new IOException("index check failure");
}
}
} catch (Exception e) {
exception.add(e);
} finally {
store.decRef();
latch.countDown();
}
}
}
};
for (MockIndexEventListener.TestEventListener eventListener : internalCluster().getDataNodeInstances(MockIndexEventListener.TestEventListener.class)) {
eventListener.setNewDelegate(listener);
}
try {
client().admin().indices().prepareDelete("test").get();
latch.await();
assertThat(exception, empty());
} finally {
for (MockIndexEventListener.TestEventListener eventListener : internalCluster().getDataNodeInstances(MockIndexEventListener.TestEventListener.class)) {
eventListener.setNewDelegate(null);
}
}
}
Aggregations