use of org.apache.tez.dag.api.client.StatusGetOpts in project hive by apache.
the class DAGSummary method hiveInputRecordsFromTezCounters.
private long hiveInputRecordsFromTezCounters(String vertexName, String inputVertexName) {
// Get the counters for the input vertex.
Set<StatusGetOpts> statusOptions = Collections.singleton(StatusGetOpts.GET_COUNTERS);
VertexStatus inputVertexStatus = vertexStatus(statusOptions, inputVertexName);
final TezCounters inputVertexCounters = inputVertexStatus.getVertexCounters();
// eg, group name TaskCounter_Map_7_OUTPUT_Reducer_8, counter name OUTPUT_RECORDS
String groupName = formattedName("TaskCounter", inputVertexName, vertexName);
String counterName = "OUTPUT_RECORDS";
// Do not create counter if it does not exist -
// instead fall back to default behavior for determining input records.
TezCounter tezCounter = inputVertexCounters.getGroup(groupName).findCounter(counterName, false);
if (tezCounter == null) {
return -1;
} else {
return tezCounter.getValue();
}
}
use of org.apache.tez.dag.api.client.StatusGetOpts in project hive by apache.
the class LLAPioSummary method print.
@Override
public void print(SessionState.LogHelper console) {
console.printInfo("");
console.printInfo(LLAP_IO_SUMMARY_HEADER);
SortedSet<String> keys = new TreeSet<>(progressMap.keySet());
Set<StatusGetOpts> statusOptions = Collections.singleton(StatusGetOpts.GET_COUNTERS);
String counterGroup = LlapIOCounters.class.getName();
for (String vertexName : keys) {
// Reducers do not benefit from LLAP IO so no point in printing
if (vertexName.startsWith("Reducer")) {
continue;
}
TezCounters vertexCounters = vertexCounter(statusOptions, vertexName);
if (vertexCounters != null) {
if (!first) {
console.printInfo(SEPARATOR);
console.printInfo(LLAP_SUMMARY_HEADER);
console.printInfo(SEPARATOR);
first = true;
}
console.printInfo(vertexSummary(vertexName, counterGroup, vertexCounters));
}
}
console.printInfo(SEPARATOR);
console.printInfo("");
}
use of org.apache.tez.dag.api.client.StatusGetOpts in project hive by apache.
the class FSCountersSummary method print.
@Override
public void print(SessionState.LogHelper console) {
console.printInfo("FileSystem Counters Summary");
SortedSet<String> keys = new TreeSet<>(progressMap.keySet());
Set<StatusGetOpts> statusOptions = Collections.singleton(StatusGetOpts.GET_COUNTERS);
// as well. If not, we need a way to get all the schemes that are accessed by the tez task/llap.
for (FileSystem.Statistics statistics : FileSystem.getAllStatistics()) {
final String scheme = statistics.getScheme().toUpperCase();
console.printInfo("");
console.printInfo("Scheme: " + scheme);
console.printInfo(SEPARATOR);
console.printInfo(HEADER);
console.printInfo(SEPARATOR);
for (String vertexName : keys) {
TezCounters vertexCounters = vertexCounters(statusOptions, vertexName);
if (vertexCounters != null) {
console.printInfo(summary(scheme, vertexName, vertexCounters));
}
}
console.printInfo(SEPARATOR);
}
}
use of org.apache.tez.dag.api.client.StatusGetOpts in project hive by apache.
the class TezTask method execute.
@Override
public int execute(DriverContext driverContext) {
int rc = 1;
boolean cleanContext = false;
Context ctx = null;
Ref<TezSessionState> sessionRef = Ref.from(null);
try {
// Get or create Context object. If we create it we have to clean it later as well.
ctx = driverContext.getCtx();
if (ctx == null) {
ctx = new Context(conf);
cleanContext = true;
// some DDL task that directly executes a TezTask does not setup Context and hence TriggerContext.
// Setting queryId is messed up. Some DDL tasks have executionId instead of proper queryId.
String queryId = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYID);
WmContext wmContext = new WmContext(System.currentTimeMillis(), queryId);
ctx.setWmContext(wmContext);
}
// Need to remove this static hack. But this is the way currently to get a session.
SessionState ss = SessionState.get();
// Note: given that we return pool sessions to the pool in the finally block below, and that
// we need to set the global to null to do that, this "reuse" may be pointless.
TezSessionState session = sessionRef.value = ss.getTezSession();
if (session != null && !session.isOpen()) {
LOG.warn("The session: " + session + " has not been opened");
}
// We only need a username for UGI to use for groups; getGroups will fetch the groups
// based on Hadoop configuration, as documented at
// https://hadoop.apache.org/docs/r2.8.0/hadoop-project-dist/hadoop-common/GroupsMapping.html
String userName = ss.getUserName();
List<String> groups = null;
if (userName == null) {
userName = "anonymous";
} else {
groups = UserGroupInformation.createRemoteUser(ss.getUserName()).getGroups();
}
MappingInput mi = new MappingInput(userName, groups, ss.getHiveVariables().get("wmpool"), ss.getHiveVariables().get("wmapp"));
WmContext wmContext = ctx.getWmContext();
// jobConf will hold all the configuration for hadoop, tez, and hive
JobConf jobConf = utils.createConfiguration(conf);
// Get all user jars from work (e.g. input format stuff).
String[] allNonConfFiles = work.configureJobConfAndExtractJars(jobConf);
// DAG scratch dir. We get a session from the pool so it may be different from Tez one.
// TODO: we could perhaps reuse the same directory for HiveResources?
Path scratchDir = utils.createTezDir(ctx.getMRScratchDir(), conf);
CallerContext callerContext = CallerContext.create("HIVE", queryPlan.getQueryId(), "HIVE_QUERY_ID", queryPlan.getQueryStr());
perfLogger.PerfLogBegin(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
session = sessionRef.value = WorkloadManagerFederation.getSession(sessionRef.value, conf, mi, getWork().getLlapMode(), wmContext);
perfLogger.PerfLogEnd(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
try {
ss.setTezSession(session);
LOG.info("Subscribed to counters: {} for queryId: {}", wmContext.getSubscribedCounters(), wmContext.getQueryId());
// Ensure the session is open and has the necessary local resources.
// This would refresh any conf resources and also local resources.
ensureSessionHasResources(session, allNonConfFiles);
// This is a combination of the jar stuff from conf, and not from conf.
List<LocalResource> allNonAppResources = session.getLocalizedResources();
logResources(allNonAppResources);
Map<String, LocalResource> allResources = DagUtils.createTezLrMap(session.getAppJarLr(), allNonAppResources);
// next we translate the TezWork to a Tez DAG
DAG dag = build(jobConf, work, scratchDir, ctx, allResources);
dag.setCallerContext(callerContext);
// Check isShutdown opportunistically; it's never unset.
if (this.isShutdown) {
throw new HiveException("Operation cancelled");
}
DAGClient dagClient = submit(jobConf, dag, sessionRef);
session = sessionRef.value;
boolean wasShutdown = false;
synchronized (dagClientLock) {
assert this.dagClient == null;
wasShutdown = this.isShutdown;
if (!wasShutdown) {
this.dagClient = dagClient;
}
}
if (wasShutdown) {
closeDagClientOnCancellation(dagClient);
throw new HiveException("Operation cancelled");
}
// finally monitor will print progress until the job is done
TezJobMonitor monitor = new TezJobMonitor(work.getAllWork(), dagClient, conf, dag, ctx);
rc = monitor.monitorExecution();
if (rc != 0) {
this.setException(new HiveException(monitor.getDiagnostics()));
}
// fetch the counters
try {
Set<StatusGetOpts> statusGetOpts = EnumSet.of(StatusGetOpts.GET_COUNTERS);
counters = dagClient.getDAGStatus(statusGetOpts).getDAGCounters();
} catch (Exception err) {
// Don't fail execution due to counters - just don't print summary info
LOG.warn("Failed to get counters. Ignoring, summary info will be incomplete. " + err, err);
counters = null;
}
} finally {
// Note: due to TEZ-3846, the session may actually be invalid in case of some errors.
// Currently, reopen on an attempted reuse will take care of that; we cannot tell
// if the session is usable until we try.
// We return this to the pool even if it's unusable; reopen is supposed to handle this.
wmContext = ctx.getWmContext();
try {
if (sessionRef.value != null) {
sessionRef.value.returnToSessionManager();
}
} catch (Exception e) {
LOG.error("Failed to return session: {} to pool", session, e);
throw e;
}
if (!conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("none") && wmContext != null) {
if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("json")) {
wmContext.printJson(console);
} else if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("text")) {
wmContext.print(console);
}
}
}
if (LOG.isInfoEnabled() && counters != null && (HiveConf.getBoolVar(conf, HiveConf.ConfVars.TEZ_EXEC_SUMMARY) || Utilities.isPerfOrAboveLogging(conf))) {
for (CounterGroup group : counters) {
LOG.info(group.getDisplayName() + ":");
for (TezCounter counter : group) {
LOG.info(" " + counter.getDisplayName() + ": " + counter.getValue());
}
}
}
} catch (Exception e) {
LOG.error("Failed to execute tez graph.", e);
// rc will be 1 at this point indicating failure.
} finally {
Utilities.clearWork(conf);
// Clear gWorkMap
for (BaseWork w : work.getAllWork()) {
JobConf workCfg = workToConf.get(w);
if (workCfg != null) {
Utilities.clearWorkMapForConf(workCfg);
}
}
if (cleanContext) {
try {
ctx.clear();
} catch (Exception e) {
/*best effort*/
LOG.warn("Failed to clean up after tez job", e);
}
}
// need to either move tmp files or remove them
DAGClient dagClient = null;
synchronized (dagClientLock) {
dagClient = this.dagClient;
this.dagClient = null;
}
// DagClient as such should have no bearing on jobClose.
if (dagClient != null) {
// rc will only be overwritten if close errors out
rc = close(work, rc, dagClient);
}
}
return rc;
}
use of org.apache.tez.dag.api.client.StatusGetOpts in project tez by apache.
the class TestTezJobs method testPerIOCounterAggregation.
@Test(timeout = 60000)
public void testPerIOCounterAggregation() throws Exception {
String baseDir = "/tmp/perIOCounterAgg/";
Path inPath1 = new Path(baseDir + "inPath1");
Path inPath2 = new Path(baseDir + "inPath2");
Path outPath = new Path(baseDir + "outPath");
final Set<String> expectedResults = generateSortMergeJoinInput(inPath1, inPath2);
Path stagingDirPath = new Path("/tmp/tez-staging-dir");
remoteFs.mkdirs(stagingDirPath);
TezConfiguration conf = new TezConfiguration(mrrTezCluster.getConfig());
conf.setBoolean(TezConfiguration.TEZ_TASK_GENERATE_COUNTERS_PER_IO, true);
TezClient tezClient = TezClient.create(SortMergeJoinHelper.class.getSimpleName(), conf);
tezClient.start();
SortMergeJoinHelper sortMergeJoinHelper = new SortMergeJoinHelper(tezClient);
sortMergeJoinHelper.setConf(conf);
String[] args = new String[] { "-D" + TezConfiguration.TEZ_AM_STAGING_DIR + "=" + stagingDirPath.toString(), "-counter", inPath1.toString(), inPath2.toString(), "1", outPath.toString() };
assertEquals(0, sortMergeJoinHelper.run(conf, args, tezClient));
verifySortMergeJoinInput(outPath, expectedResults);
String joinerVertexName = "joiner";
String input1Name = "input1";
String input2Name = "input2";
String joinOutputName = "joinOutput";
Set<StatusGetOpts> statusOpts = new HashSet<StatusGetOpts>();
statusOpts.add(StatusGetOpts.GET_COUNTERS);
VertexStatus joinerVertexStatus = sortMergeJoinHelper.dagClient.getVertexStatus(joinerVertexName, statusOpts);
final TezCounters joinerCounters = joinerVertexStatus.getVertexCounters();
final CounterGroup aggregatedGroup = joinerCounters.getGroup(TaskCounter.class.getCanonicalName());
final CounterGroup input1Group = joinerCounters.getGroup(TaskCounter.class.getSimpleName() + "_" + joinerVertexName + "_INPUT_" + input1Name);
final CounterGroup input2Group = joinerCounters.getGroup(TaskCounter.class.getSimpleName() + "_" + joinerVertexName + "_INPUT_" + input2Name);
assertTrue("aggregated counter group cannot be empty", aggregatedGroup.size() > 0);
assertTrue("per io group for input1 cannot be empty", input1Group.size() > 0);
assertTrue("per io group for input1 cannot be empty", input2Group.size() > 0);
List<TaskCounter> countersToVerifyAgg = Arrays.asList(TaskCounter.ADDITIONAL_SPILLS_BYTES_READ, TaskCounter.ADDITIONAL_SPILLS_BYTES_WRITTEN, TaskCounter.COMBINE_INPUT_RECORDS, TaskCounter.MERGED_MAP_OUTPUTS, TaskCounter.NUM_DISK_TO_DISK_MERGES, TaskCounter.NUM_FAILED_SHUFFLE_INPUTS, TaskCounter.NUM_MEM_TO_DISK_MERGES, TaskCounter.NUM_SHUFFLED_INPUTS, TaskCounter.NUM_SKIPPED_INPUTS, TaskCounter.REDUCE_INPUT_GROUPS, TaskCounter.REDUCE_INPUT_RECORDS, TaskCounter.SHUFFLE_BYTES, TaskCounter.SHUFFLE_BYTES_DECOMPRESSED, TaskCounter.SHUFFLE_BYTES_DISK_DIRECT, TaskCounter.SHUFFLE_BYTES_TO_DISK, TaskCounter.SHUFFLE_BYTES_TO_MEM, TaskCounter.SPILLED_RECORDS);
int nonZeroCounters = 0;
// verify that the sum of the counter values for edges add up to the aggregated counter value.
for (TaskCounter c : countersToVerifyAgg) {
TezCounter aggregatedCounter = aggregatedGroup.findCounter(c.name(), false);
TezCounter input1Counter = input1Group.findCounter(c.name(), false);
TezCounter input2Counter = input2Group.findCounter(c.name(), false);
assertNotNull("aggregated counter cannot be null " + c.name(), aggregatedCounter);
assertNotNull("input1 counter cannot be null " + c.name(), input1Counter);
assertNotNull("input2 counter cannot be null " + c.name(), input2Counter);
assertEquals("aggregated counter does not match sum of input counters " + c.name(), aggregatedCounter.getValue(), input1Counter.getValue() + input2Counter.getValue());
if (aggregatedCounter.getValue() > 0) {
nonZeroCounters++;
}
}
// ensure that at least one of the counters tested above were non-zero.
assertTrue("At least one of the counter should be non-zero. invalid test ", nonZeroCounters > 0);
CounterGroup joinerOutputGroup = joinerCounters.getGroup(TaskCounter.class.getSimpleName() + "_" + joinerVertexName + "_OUTPUT_" + joinOutputName);
String outputCounterName = TaskCounter.OUTPUT_RECORDS.name();
TezCounter aggregateCounter = aggregatedGroup.findCounter(outputCounterName, false);
TezCounter joinerOutputCounter = joinerOutputGroup.findCounter(outputCounterName, false);
assertNotNull("aggregated counter cannot be null " + outputCounterName, aggregateCounter);
assertNotNull("output counter cannot be null " + outputCounterName, joinerOutputCounter);
assertTrue("counter value is zero. test is invalid", aggregateCounter.getValue() > 0);
assertEquals("aggregated counter does not match sum of output counters " + outputCounterName, aggregateCounter.getValue(), joinerOutputCounter.getValue());
}
Aggregations