use of org.apache.solr.util.DefaultSolrThreadFactory in project lucene-solr by apache.
the class TestInPlaceUpdatesDistrib method reorderedDeletesTest.
// The following should work: full update to doc 0, in-place update for doc 0, delete doc 0
private void reorderedDeletesTest() throws Exception {
clearIndex();
commit();
buildRandomIndex(0);
float inplace_updatable_float = 1;
// update doc, set
index("id", 0, "inplace_updatable_float", map("set", inplace_updatable_float));
LEADER.commit();
// RTG straight from the index
SolrDocument sdoc = LEADER.getById("0");
assertEquals(inplace_updatable_float, sdoc.get("inplace_updatable_float"));
assertEquals("title0", sdoc.get("title_s"));
long version0 = (long) sdoc.get("_version_");
// put replica out of sync
float newinplace_updatable_float = 100;
List<UpdateRequest> updates = new ArrayList<>();
// full update
updates.add(simulatedUpdateRequest(null, "id", 0, "title_s", "title0_new", "inplace_updatable_float", newinplace_updatable_float, "_version_", version0 + 1));
// inplace_updatable_float=101
updates.add(simulatedUpdateRequest(version0 + 1, "id", 0, "inplace_updatable_float", newinplace_updatable_float + 1, "_version_", version0 + 2));
updates.add(simulatedDeleteRequest(0, version0 + 3));
// order the updates correctly for NONLEADER 1
for (UpdateRequest update : updates) {
log.info("Issuing well ordered update: " + update.getDocuments());
NONLEADERS.get(1).request(update);
}
// Reordering needs to happen using parallel threads
ExecutorService threadpool = ExecutorUtil.newMDCAwareFixedThreadPool(updates.size() + 1, new DefaultSolrThreadFactory(getTestName()));
// re-order the updates for NONLEADER 0
List<UpdateRequest> reorderedUpdates = new ArrayList<>(updates);
Collections.shuffle(reorderedUpdates, r);
List<Future<UpdateResponse>> updateResponses = new ArrayList<>();
for (UpdateRequest update : reorderedUpdates) {
AsyncUpdateWithRandomCommit task = new AsyncUpdateWithRandomCommit(update, NONLEADERS.get(0), random().nextLong());
updateResponses.add(threadpool.submit(task));
// while we can't guarantee/trust what order the updates are executed in, since multiple threads
// are involved, but we're trying to bias the thread scheduling to run them in the order submitted
Thread.sleep(10);
}
threadpool.shutdown();
assertTrue("Thread pool didn't terminate within 15 secs", threadpool.awaitTermination(15, TimeUnit.SECONDS));
// assert all requests were successful
for (Future<UpdateResponse> resp : updateResponses) {
assertEquals(0, resp.get().getStatus());
}
// assert both replicas have same effect
for (SolrClient client : NONLEADERS) {
// 0th is re-ordered replica, 1st is well-ordered replica
SolrDocument doc = client.getById(String.valueOf(0), params("distrib", "false"));
assertNull("This doc was supposed to have been deleted, but was: " + doc, doc);
}
log.info("reorderedDeletesTest: This test passed fine...");
}
use of org.apache.solr.util.DefaultSolrThreadFactory in project lucene-solr by apache.
the class TestStressCloudBlindAtomicUpdates method createMiniSolrCloudCluster.
@BeforeClass
private static void createMiniSolrCloudCluster() throws Exception {
// NOTE: numDocsToCheck uses atLeast, so nightly & multiplier are alreayd a factor in index size
// no need to redundently factor them in here as well
DOC_ID_INCR = TestUtil.nextInt(random(), 1, 7);
NUM_THREADS = atLeast(3);
EXEC_SERVICE = ExecutorUtil.newMDCAwareFixedThreadPool(NUM_THREADS, new DefaultSolrThreadFactory(DEBUG_LABEL));
// at least 2, but don't go crazy on nightly/test.multiplier with "atLeast()"
final int numShards = TEST_NIGHTLY ? 5 : 2;
final int repFactor = 2;
final int numNodes = numShards * repFactor;
final String configName = DEBUG_LABEL + "_config-set";
final Path configDir = Paths.get(TEST_HOME(), "collection1", "conf");
configureCluster(numNodes).addConfig(configName, configDir).configure();
CLOUD_CLIENT = cluster.getSolrClient();
CLOUD_CLIENT.setDefaultCollection(COLLECTION_NAME);
CollectionAdminRequest.createCollection(COLLECTION_NAME, configName, numShards, repFactor).withProperty("config", "solrconfig-tlog.xml").withProperty("schema", "schema-minimal-atomic-stress.xml").process(CLOUD_CLIENT);
waitForRecoveriesToFinish(CLOUD_CLIENT);
for (JettySolrRunner jetty : cluster.getJettySolrRunners()) {
CLIENTS.add(getHttpSolrClient(jetty.getBaseUrl() + "/" + COLLECTION_NAME + "/"));
}
// sanity check no one broke the assumptions we make about our schema
checkExpectedSchemaType(map("name", "long", "class", "solr.TrieLongField", "multiValued", Boolean.FALSE, "indexed", Boolean.FALSE, "stored", Boolean.FALSE, "docValues", Boolean.FALSE));
}
use of org.apache.solr.util.DefaultSolrThreadFactory in project lucene-solr by apache.
the class EmptyRequestHandler method testRefCountMT.
@Test
public void testRefCountMT() throws Exception {
SolrCore core = h.getCore();
assertTrue("Refcount != 1", core.getOpenCount() == 1);
final ClosingRequestHandler handler1 = new ClosingRequestHandler();
handler1.inform(core);
String path = "/this/is A path /that won't be registered!";
SolrRequestHandler old = core.registerRequestHandler(path, handler1);
// should not be anything...
assertNull(old);
assertEquals(core.getRequestHandlers().get(path), handler1);
final int LOOP = 100;
final int MT = 16;
ExecutorService service = ExecutorUtil.newMDCAwareFixedThreadPool(MT, new DefaultSolrThreadFactory("refCountMT"));
List<Callable<Integer>> callees = new ArrayList<>(MT);
final CoreContainer cores = h.getCoreContainer();
for (int i = 0; i < MT; ++i) {
Callable<Integer> call = new Callable<Integer>() {
void yield(int n) {
try {
Thread.sleep(0, (n % 13 + 1) * 10);
} catch (InterruptedException xint) {
}
}
@Override
public Integer call() {
SolrCore core = null;
int r = 0;
try {
for (int l = 0; l < LOOP; ++l) {
r += 1;
core = cores.getCore(SolrTestCaseJ4.DEFAULT_TEST_CORENAME);
// sprinkle concurrency hinting...
yield(l);
assertTrue("Refcount < 1", core.getOpenCount() >= 1);
yield(l);
assertTrue("Refcount > 17", core.getOpenCount() <= 17);
yield(l);
assertTrue("Handler is closed", handler1.closed == false);
yield(l);
core.close();
core = null;
yield(l);
}
return r;
} finally {
if (core != null)
core.close();
}
}
};
callees.add(call);
}
List<Future<Integer>> results = service.invokeAll(callees);
for (Future<Integer> result : results) {
assertTrue("loop=" + result.get() + " < " + LOOP, result.get() >= LOOP);
}
cores.shutdown();
assertTrue("Refcount != 0", core.getOpenCount() == 0);
assertTrue("Handler not closed", core.isClosed() && handler1.closed == true);
service.shutdown();
assertTrue("Running for too long...", service.awaitTermination(60, TimeUnit.SECONDS));
}
use of org.apache.solr.util.DefaultSolrThreadFactory in project lucene-solr by apache.
the class EmptyRequestHandler method testReloadLeak.
/**
* Test that's meant to be run with many iterations to expose a leak of SolrIndexSearcher when a core is closed
* due to a reload. Without the fix, this test fails with most iters=1000 runs.
*/
@Test
public void testReloadLeak() throws Exception {
final ExecutorService executor = ExecutorUtil.newMDCAwareFixedThreadPool(1, new DefaultSolrThreadFactory("testReloadLeak"));
// Continuously open new searcher while core is not closed, and reload core to try to reproduce searcher leak.
// While in practice we never continuously open new searchers, this is trying to make up for the fact that opening
// a searcher in this empty core is very fast by opening new searchers continuously to increase the likelihood
// for race.
SolrCore core = h.getCore();
assertTrue("Refcount != 1", core.getOpenCount() == 1);
executor.execute(new NewSearcherRunnable(core));
// Since we called getCore() vs getCoreInc() and don't own a refCount, the container should decRef the core
// and close it when we call reload.
h.reload();
executor.shutdown();
executor.awaitTermination(1, TimeUnit.MINUTES);
// Check that all cores are closed and no searcher references are leaked.
assertTrue("SolrCore " + core + " is not closed", core.isClosed());
assertTrue(core.areAllSearcherReferencesEmpty());
}
use of org.apache.solr.util.DefaultSolrThreadFactory in project lucene-solr by apache.
the class IndexFetcher method fetchLatestIndex.
/**
* This command downloads all the necessary files from master to install a index commit point. Only changed files are
* downloaded. It also downloads the conf files (if they are modified).
*
* @param forceReplication force a replication in all cases
* @param forceCoreReload force a core reload in all cases
* @return true on success, false if slave is already in sync
* @throws IOException if an exception occurs
*/
IndexFetchResult fetchLatestIndex(boolean forceReplication, boolean forceCoreReload) throws IOException, InterruptedException {
boolean cleanupDone = false;
boolean successfulInstall = false;
markReplicationStart();
Directory tmpIndexDir = null;
String tmpIndex;
Directory indexDir = null;
String indexDirPath;
boolean deleteTmpIdxDir = true;
File tmpTlogDir = null;
if (!solrCore.getSolrCoreState().getLastReplicateIndexSuccess()) {
// if the last replication was not a success, we force a full replication
// when we are a bit more confident we may want to try a partial replication
// if the error is connection related or something, but we have to be careful
forceReplication = true;
LOG.info("Last replication failed, so I'll force replication");
}
try {
if (fetchFromLeader) {
assert !solrCore.isClosed() : "Replication should be stopped before closing the core";
Replica replica = getLeaderReplica();
CloudDescriptor cd = solrCore.getCoreDescriptor().getCloudDescriptor();
if (cd.getCoreNodeName().equals(replica.getName())) {
return IndexFetchResult.EXPECTING_NON_LEADER;
}
if (replica.getState() != Replica.State.ACTIVE) {
LOG.info("Replica {} is leader but it's state is {}, skipping replication", replica.getName(), replica.getState());
return IndexFetchResult.LEADER_IS_NOT_ACTIVE;
}
if (!solrCore.getCoreContainer().getZkController().getClusterState().liveNodesContain(replica.getNodeName())) {
LOG.info("Replica {} is leader but it's not hosted on a live node, skipping replication", replica.getName());
return IndexFetchResult.LEADER_IS_NOT_ACTIVE;
}
if (!replica.getCoreUrl().equals(masterUrl)) {
masterUrl = replica.getCoreUrl();
LOG.info("Updated masterUrl to {}", masterUrl);
// TODO: Do we need to set forceReplication = true?
} else {
LOG.debug("masterUrl didn't change");
}
}
//get the current 'replicateable' index version in the master
NamedList response;
try {
response = getLatestVersion();
} catch (Exception e) {
final String errorMsg = e.toString();
if (!Strings.isNullOrEmpty(errorMsg) && errorMsg.contains(INTERRUPT_RESPONSE_MESSAGE)) {
LOG.warn("Master at: " + masterUrl + " is not available. Index fetch failed by interrupt. Exception: " + errorMsg);
return new IndexFetchResult(IndexFetchResult.FAILED_BY_INTERRUPT_MESSAGE, false, e);
} else {
LOG.warn("Master at: " + masterUrl + " is not available. Index fetch failed by exception: " + errorMsg);
return new IndexFetchResult(IndexFetchResult.FAILED_BY_EXCEPTION_MESSAGE, false, e);
}
}
long latestVersion = (Long) response.get(CMD_INDEX_VERSION);
long latestGeneration = (Long) response.get(GENERATION);
LOG.info("Master's generation: " + latestGeneration);
LOG.info("Master's version: " + latestVersion);
// TODO: make sure that getLatestCommit only returns commit points for the main index (i.e. no side-car indexes)
IndexCommit commit = solrCore.getDeletionPolicy().getLatestCommit();
if (commit == null) {
// Presumably the IndexWriter hasn't been opened yet, and hence the deletion policy hasn't been updated with commit points
RefCounted<SolrIndexSearcher> searcherRefCounted = null;
try {
searcherRefCounted = solrCore.getNewestSearcher(false);
if (searcherRefCounted == null) {
LOG.warn("No open searcher found - fetch aborted");
return IndexFetchResult.NO_INDEX_COMMIT_EXIST;
}
commit = searcherRefCounted.get().getIndexReader().getIndexCommit();
} finally {
if (searcherRefCounted != null)
searcherRefCounted.decref();
}
}
LOG.info("Slave's generation: " + commit.getGeneration());
LOG.info("Slave's version: " + IndexDeletionPolicyWrapper.getCommitTimestamp(commit));
if (latestVersion == 0L) {
if (forceReplication && commit.getGeneration() != 0) {
// since we won't get the files for an empty index,
// we just clear ours and commit
LOG.info("New index in Master. Deleting mine...");
RefCounted<IndexWriter> iw = solrCore.getUpdateHandler().getSolrCoreState().getIndexWriter(solrCore);
try {
iw.get().deleteAll();
} finally {
iw.decref();
}
SolrQueryRequest req = new LocalSolrQueryRequest(solrCore, new ModifiableSolrParams());
solrCore.getUpdateHandler().commit(new CommitUpdateCommand(req, false));
}
//there is nothing to be replicated
successfulInstall = true;
LOG.debug("Nothing to replicate, master's version is 0");
return IndexFetchResult.MASTER_VERSION_ZERO;
}
// TODO: Should we be comparing timestamps (across machines) here?
if (!forceReplication && IndexDeletionPolicyWrapper.getCommitTimestamp(commit) == latestVersion) {
//master and slave are already in sync just return
LOG.info("Slave in sync with master.");
successfulInstall = true;
return IndexFetchResult.ALREADY_IN_SYNC;
}
LOG.info("Starting replication process");
// get the list of files first
fetchFileList(latestGeneration);
// this can happen if the commit point is deleted before we fetch the file list.
if (filesToDownload.isEmpty()) {
return IndexFetchResult.PEER_INDEX_COMMIT_DELETED;
}
LOG.info("Number of files in latest index in master: " + filesToDownload.size());
if (tlogFilesToDownload != null) {
LOG.info("Number of tlog files in master: " + tlogFilesToDownload.size());
}
// Create the sync service
fsyncService = ExecutorUtil.newMDCAwareSingleThreadExecutor(new DefaultSolrThreadFactory("fsyncService"));
// use a synchronized list because the list is read by other threads (to show details)
filesDownloaded = Collections.synchronizedList(new ArrayList<Map<String, Object>>());
// if the generation of master is older than that of the slave , it means they are not compatible to be copied
// then a new index directory to be created and all the files need to be copied
boolean isFullCopyNeeded = IndexDeletionPolicyWrapper.getCommitTimestamp(commit) >= latestVersion || commit.getGeneration() >= latestGeneration || forceReplication;
String timestamp = new SimpleDateFormat(SnapShooter.DATE_FMT, Locale.ROOT).format(new Date());
String tmpIdxDirName = "index." + timestamp;
tmpIndex = solrCore.getDataDir() + tmpIdxDirName;
tmpIndexDir = solrCore.getDirectoryFactory().get(tmpIndex, DirContext.DEFAULT, solrCore.getSolrConfig().indexConfig.lockType);
// tmp dir for tlog files
if (tlogFilesToDownload != null) {
tmpTlogDir = new File(solrCore.getUpdateHandler().getUpdateLog().getLogDir(), "tlog." + timestamp);
}
// cindex dir...
indexDirPath = solrCore.getIndexDir();
indexDir = solrCore.getDirectoryFactory().get(indexDirPath, DirContext.DEFAULT, solrCore.getSolrConfig().indexConfig.lockType);
try {
//in the metadata. If there is a mismatch for the same index file then we download the entire index again.
if (!isFullCopyNeeded && isIndexStale(indexDir)) {
isFullCopyNeeded = true;
}
if (!isFullCopyNeeded && !fetchFromLeader) {
// and wait until we are able to clean up all unused lucene files
if (solrCore.getCoreContainer().isZooKeeperAware()) {
solrCore.closeSearcher();
}
// rollback and reopen index writer and wait until all unused files
// are successfully deleted
solrCore.getUpdateHandler().newIndexWriter(true);
RefCounted<IndexWriter> writer = solrCore.getUpdateHandler().getSolrCoreState().getIndexWriter(null);
try {
IndexWriter indexWriter = writer.get();
int c = 0;
indexWriter.deleteUnusedFiles();
while (hasUnusedFiles(indexDir, commit)) {
indexWriter.deleteUnusedFiles();
LOG.info("Sleeping for 1000ms to wait for unused lucene index files to be delete-able");
Thread.sleep(1000);
c++;
if (c >= 30) {
LOG.warn("IndexFetcher unable to cleanup unused lucene index files so we must do a full copy instead");
isFullCopyNeeded = true;
break;
}
}
if (c > 0) {
LOG.info("IndexFetcher slept for " + (c * 1000) + "ms for unused lucene index files to be delete-able");
}
} finally {
writer.decref();
}
}
boolean reloadCore = false;
try {
// we have to be careful and do this after we know isFullCopyNeeded won't be flipped
if (!isFullCopyNeeded) {
solrCore.getUpdateHandler().getSolrCoreState().closeIndexWriter(solrCore, true);
}
LOG.info("Starting download (fullCopy={}) to {}", isFullCopyNeeded, tmpIndexDir);
successfulInstall = false;
long bytesDownloaded = downloadIndexFiles(isFullCopyNeeded, indexDir, tmpIndexDir, latestGeneration);
if (tlogFilesToDownload != null) {
bytesDownloaded += downloadTlogFiles(tmpTlogDir, latestGeneration);
// reload update log
reloadCore = true;
}
final long timeTakenSeconds = getReplicationTimeElapsed();
final Long bytesDownloadedPerSecond = (timeTakenSeconds != 0 ? new Long(bytesDownloaded / timeTakenSeconds) : null);
LOG.info("Total time taken for download (fullCopy={},bytesDownloaded={}) : {} secs ({} bytes/sec) to {}", isFullCopyNeeded, bytesDownloaded, timeTakenSeconds, bytesDownloadedPerSecond, tmpIndexDir);
Collection<Map<String, Object>> modifiedConfFiles = getModifiedConfFiles(confFilesToDownload);
if (!modifiedConfFiles.isEmpty()) {
reloadCore = true;
downloadConfFiles(confFilesToDownload, latestGeneration);
if (isFullCopyNeeded) {
successfulInstall = solrCore.modifyIndexProps(tmpIdxDirName);
if (successfulInstall)
deleteTmpIdxDir = false;
} else {
successfulInstall = moveIndexFiles(tmpIndexDir, indexDir);
}
if (tlogFilesToDownload != null) {
// move tlog files and refresh ulog only if we successfully installed a new index
successfulInstall &= moveTlogFiles(tmpTlogDir);
}
if (successfulInstall) {
if (isFullCopyNeeded) {
// may be closed
if (indexDir != null) {
solrCore.getDirectoryFactory().doneWithDirectory(indexDir);
// Cleanup all index files not associated with any *named* snapshot.
solrCore.deleteNonSnapshotIndexFiles(indexDirPath);
}
}
LOG.info("Configuration files are modified, core will be reloaded");
logReplicationTimeAndConfFiles(modifiedConfFiles, // write to a file time of replication and
successfulInstall);
// conf files.
}
} else {
terminateAndWaitFsyncService();
if (isFullCopyNeeded) {
successfulInstall = solrCore.modifyIndexProps(tmpIdxDirName);
if (successfulInstall)
deleteTmpIdxDir = false;
} else {
successfulInstall = moveIndexFiles(tmpIndexDir, indexDir);
}
if (tlogFilesToDownload != null) {
// move tlog files and refresh ulog only if we successfully installed a new index
successfulInstall &= moveTlogFiles(tmpTlogDir);
}
if (successfulInstall) {
logReplicationTimeAndConfFiles(modifiedConfFiles, successfulInstall);
}
}
} finally {
if (!isFullCopyNeeded) {
solrCore.getUpdateHandler().getSolrCoreState().openIndexWriter(solrCore);
}
}
// we must reload the core after we open the IW back up
if (successfulInstall && (reloadCore || forceCoreReload)) {
LOG.info("Reloading SolrCore {}", solrCore.getName());
reloadCore();
}
if (successfulInstall) {
if (isFullCopyNeeded) {
// may be closed
if (indexDir != null) {
LOG.info("removing old index directory " + indexDir);
solrCore.getDirectoryFactory().doneWithDirectory(indexDir);
solrCore.getDirectoryFactory().remove(indexDir);
}
}
if (isFullCopyNeeded) {
solrCore.getUpdateHandler().newIndexWriter(isFullCopyNeeded);
}
openNewSearcherAndUpdateCommitPoint();
}
if (!isFullCopyNeeded && !forceReplication && !successfulInstall) {
cleanup(solrCore, tmpIndexDir, indexDir, deleteTmpIdxDir, tmpTlogDir, successfulInstall);
cleanupDone = true;
// we try with a full copy of the index
LOG.warn("Replication attempt was not successful - trying a full index replication reloadCore={}", reloadCore);
successfulInstall = fetchLatestIndex(true, reloadCore).getSuccessful();
}
markReplicationStop();
return successfulInstall ? IndexFetchResult.INDEX_FETCH_SUCCESS : IndexFetchResult.INDEX_FETCH_FAILURE;
} catch (ReplicationHandlerException e) {
LOG.error("User aborted Replication");
return new IndexFetchResult(IndexFetchResult.FAILED_BY_EXCEPTION_MESSAGE, false, e);
} catch (SolrException e) {
throw e;
} catch (InterruptedException e) {
throw new InterruptedException("Index fetch interrupted");
} catch (Exception e) {
throw new SolrException(ErrorCode.SERVER_ERROR, "Index fetch failed : ", e);
}
} finally {
if (!cleanupDone) {
cleanup(solrCore, tmpIndexDir, indexDir, deleteTmpIdxDir, tmpTlogDir, successfulInstall);
}
}
}
Aggregations