/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hive.ql.exec;
import java.io.Serializable;
import java.util.ArrayList;
import java.util.Collection;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.concurrent.Callable;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hive.common.StatsSetupConst;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.conf.HiveConf.ConfVars;
import org.apache.hadoop.hive.metastore.MetaStoreUtils;
import org.apache.hadoop.hive.metastore.Warehouse;
import org.apache.hadoop.hive.metastore.api.EnvironmentContext;
import org.apache.hadoop.hive.metastore.api.MetaException;
import org.apache.hadoop.hive.metastore.api.StorageDescriptor;
import org.apache.hadoop.hive.ql.DriverContext;
import org.apache.hadoop.hive.ql.ErrorMsg;
import org.apache.hadoop.hive.ql.io.AcidUtils;
import org.apache.hadoop.hive.ql.metadata.Hive;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.metadata.Partition;
import org.apache.hadoop.hive.ql.metadata.Table;
import org.apache.hadoop.hive.ql.parse.BaseSemanticAnalyzer.TableSpec;
import org.apache.hadoop.hive.ql.parse.ExplainConfiguration.AnalyzeState;
import org.apache.hadoop.hive.ql.plan.DynamicPartitionCtx;
import org.apache.hadoop.hive.ql.plan.LoadTableDesc;
import org.apache.hadoop.hive.ql.plan.StatsWork;
import org.apache.hadoop.hive.ql.plan.api.StageType;
import org.apache.hadoop.hive.ql.stats.StatsAggregator;
import org.apache.hadoop.hive.ql.stats.StatsCollectionContext;
import org.apache.hadoop.hive.ql.stats.StatsFactory;
import org.apache.hadoop.hive.ql.stats.StatsPublisher;
import org.apache.hadoop.util.StringUtils;
import com.google.common.collect.Lists;
import com.google.common.util.concurrent.ThreadFactoryBuilder;
/**
* StatsTask implementation. StatsTask mainly deals with "collectable" stats. These are
* stats that require data scanning and are collected during query execution (unless the user
* explicitly requests data scanning just for the purpose of stats computation using the "ANALYZE"
* command. All other stats are computed directly by the MetaStore. The rationale being that the
* MetaStore layer covers all Thrift calls and provides better guarantees about the accuracy of
* those stats.
**/
public class StatsTask extends Task<StatsWork> implements Serializable {
private static final long serialVersionUID = 1L;
private static transient final Logger LOG = LoggerFactory.getLogger(StatsTask.class);
private Table table;
private Collection<Partition> dpPartSpecs;
public StatsTask() {
super();
dpPartSpecs = null;
}
@Override
protected void receiveFeed(FeedType feedType, Object feedValue) {
// this method should be called by MoveTask when there are dynamic partitions generated
if (feedType == FeedType.DYNAMIC_PARTITIONS) {
dpPartSpecs = (Collection<Partition>) feedValue;
}
}
@Override
public int execute(DriverContext driverContext) {
if (driverContext.getCtx().getExplainAnalyze() == AnalyzeState.RUNNING) {
return 0;
}
LOG.info("Executing stats task");
// Make sure that it is either an ANALYZE, INSERT OVERWRITE (maybe load) or CTAS command
short workComponentsPresent = 0;
if (work.getLoadTableDesc() != null) {
workComponentsPresent++;
}
if (work.getTableSpecs() != null) {
workComponentsPresent++;
}
if (work.getLoadFileDesc() != null) {
workComponentsPresent++;
}
assert (workComponentsPresent == 1);
String tableName = "";
Hive hive = getHive();
try {
if (work.getLoadTableDesc() != null) {
tableName = work.getLoadTableDesc().getTable().getTableName();
} else if (work.getTableSpecs() != null){
tableName = work.getTableSpecs().tableName;
} else {
tableName = work.getLoadFileDesc().getDestinationCreateTable();
}
table = hive.getTable(tableName);
} catch (HiveException e) {
LOG.error("Cannot get table " + tableName, e);
console.printError("Cannot get table " + tableName, e.toString());
}
return aggregateStats(hive);
}
@Override
public StageType getType() {
return StageType.STATS;
}
@Override
public String getName() {
return "STATS";
}
private int aggregateStats(Hive db) {
StatsAggregator statsAggregator = null;
int ret = 0;
StatsCollectionContext scc = null;
EnvironmentContext environmentContext = null;
try {
// Stats setup:
final Warehouse wh = new Warehouse(conf);
if (!getWork().getNoStatsAggregator() && !getWork().isNoScanAnalyzeCommand()) {
try {
scc = getContext();
statsAggregator = createStatsAggregator(scc, conf);
} catch (HiveException e) {
if (HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
throw e;
}
console.printError(ErrorMsg.STATS_SKIPPING_BY_ERROR.getErrorCodedMsg(e.toString()));
}
}
List<Partition> partitions = getPartitionsList(db);
boolean atomic = HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_STATS_ATOMIC);
String tableFullName = table.getDbName() + "." + table.getTableName();
if (partitions == null) {
org.apache.hadoop.hive.metastore.api.Table tTable = table.getTTable();
Map<String, String> parameters = tTable.getParameters();
// In the following scenarios, we need to reset the stats to true.
// work.getTableSpecs() != null means analyze command
// work.getLoadTableDesc().getReplace() is true means insert overwrite command
// work.getLoadFileDesc().getDestinationCreateTable().isEmpty() means CTAS etc.
// acidTable will not have accurate stats unless it is set through analyze command.
if (work.getTableSpecs() == null && AcidUtils.isAcidTable(table)) {
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.FALSE);
} else if (work.getTableSpecs() != null
|| (work.getLoadTableDesc() != null && work.getLoadTableDesc().getReplace())
|| (work.getLoadFileDesc() != null && !work.getLoadFileDesc()
.getDestinationCreateTable().isEmpty())) {
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.TRUE);
}
// non-partitioned tables:
if (!existStats(parameters) && atomic) {
return 0;
}
// The collectable stats for the aggregator needs to be cleared.
// For eg. if a file is being loaded, the old number of rows are not valid
if (work.isClearAggregatorStats()) {
// we choose to keep the invalid stats and only change the setting.
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.FALSE);
}
updateQuickStats(wh, parameters, tTable.getSd());
if (StatsSetupConst.areBasicStatsUptoDate(parameters)) {
if (statsAggregator != null) {
String prefix = getAggregationPrefix(table, null);
updateStats(statsAggregator, parameters, prefix, atomic);
}
// write table stats to metastore
if (!getWork().getNoStatsAggregator()) {
environmentContext = new EnvironmentContext();
environmentContext.putToProperties(StatsSetupConst.STATS_GENERATED,
StatsSetupConst.TASK);
}
}
getHive().alterTable(tableFullName, new Table(tTable), environmentContext);
if (conf.getBoolVar(ConfVars.TEZ_EXEC_SUMMARY)) {
console.printInfo("Table " + tableFullName + " stats: [" + toString(parameters) + ']');
}
LOG.info("Table " + tableFullName + " stats: [" + toString(parameters) + ']');
} else {
// Partitioned table:
// Need to get the old stats of the partition
// and update the table stats based on the old and new stats.
List<Partition> updates = new ArrayList<Partition>();
//Get the file status up-front for all partitions. Beneficial in cases of blob storage systems
final Map<String, FileStatus[]> fileStatusMap = new ConcurrentHashMap<String, FileStatus[]>();
int poolSize = conf.getInt(ConfVars.HIVE_MOVE_FILES_THREAD_COUNT.varname, 1);
// In case thread count is set to 0, use single thread.
poolSize = Math.max(poolSize, 1);
final ExecutorService pool = Executors.newFixedThreadPool(poolSize,
new ThreadFactoryBuilder().setDaemon(true)
.setNameFormat("stats-updater-thread-%d")
.build());
final List<Future<Void>> futures = Lists.newLinkedList();
LOG.debug("Getting file stats of all partitions. threadpool size:" + poolSize);
try {
for(final Partition partn : partitions) {
final String partitionName = partn.getName();
final org.apache.hadoop.hive.metastore.api.Partition tPart = partn.getTPartition();
Map<String, String> parameters = tPart.getParameters();
if (!existStats(parameters) && atomic) {
continue;
}
futures.add(pool.submit(new Callable<Void>() {
@Override
public Void call() throws Exception {
FileStatus[] partfileStatus = wh.getFileStatusesForSD(tPart.getSd());
fileStatusMap.put(partitionName, partfileStatus);
return null;
}
}));
}
pool.shutdown();
for(Future<Void> future : futures) {
future.get();
}
} catch (InterruptedException e) {
LOG.debug("Cancelling " + futures.size() + " file stats lookup tasks");
//cancel other futures
for (Future future : futures) {
future.cancel(true);
}
// Fail the query if the stats are supposed to be reliable
if (work.isStatsReliable()) {
ret = 1;
}
} finally {
if (pool != null) {
pool.shutdownNow();
}
LOG.debug("Finished getting file stats of all partitions");
}
for (Partition partn : partitions) {
//
// get the old partition stats
//
org.apache.hadoop.hive.metastore.api.Partition tPart = partn.getTPartition();
Map<String, String> parameters = tPart.getParameters();
if (work.getTableSpecs() == null && AcidUtils.isAcidTable(table)) {
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.FALSE);
} else if (work.getTableSpecs() != null
|| (work.getLoadTableDesc() != null && work.getLoadTableDesc().getReplace())
|| (work.getLoadFileDesc() != null && !work.getLoadFileDesc()
.getDestinationCreateTable().isEmpty())) {
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.TRUE);
}
//only when the stats exist, it is added to fileStatusMap
if (!fileStatusMap.containsKey(partn.getName())) {
continue;
}
// The collectable stats for the aggregator needs to be cleared.
// For eg. if a file is being loaded, the old number of rows are not valid
if (work.isClearAggregatorStats()) {
// we choose to keep the invalid stats and only change the setting.
StatsSetupConst.setBasicStatsState(parameters, StatsSetupConst.FALSE);
}
updateQuickStats(parameters, fileStatusMap.get(partn.getName()));
if (StatsSetupConst.areBasicStatsUptoDate(parameters)) {
if (statsAggregator != null) {
String prefix = getAggregationPrefix(table, partn);
updateStats(statsAggregator, parameters, prefix, atomic);
}
if (!getWork().getNoStatsAggregator()) {
environmentContext = new EnvironmentContext();
environmentContext.putToProperties(StatsSetupConst.STATS_GENERATED,
StatsSetupConst.TASK);
}
}
updates.add(new Partition(table, tPart));
if (conf.getBoolVar(ConfVars.TEZ_EXEC_SUMMARY)) {
console.printInfo("Partition " + tableFullName + partn.getSpec() +
" stats: [" + toString(parameters) + ']');
}
LOG.info("Partition " + tableFullName + partn.getSpec() +
" stats: [" + toString(parameters) + ']');
}
if (!updates.isEmpty()) {
db.alterPartitions(tableFullName, updates, environmentContext);
}
}
} catch (Exception e) {
console.printInfo("[Warning] could not update stats.",
"Failed with exception " + e.getMessage() + "\n"
+ StringUtils.stringifyException(e));
// Fail the query if the stats are supposed to be reliable
if (work.isStatsReliable()) {
ret = 1;
}
} finally {
if (statsAggregator != null) {
statsAggregator.closeConnection(scc);
}
}
// The return value of 0 indicates success,
// anything else indicates failure
return ret;
}
private String getAggregationPrefix(Table table, Partition partition)
throws MetaException {
// prefix is of the form dbName.tblName
String prefix = table.getDbName() + "." + MetaStoreUtils.encodeTableName(table.getTableName());
if (partition != null) {
return Utilities.join(prefix, Warehouse.makePartPath(partition.getSpec()));
}
return prefix;
}
private StatsAggregator createStatsAggregator(StatsCollectionContext scc, HiveConf conf) throws HiveException {
String statsImpl = HiveConf.getVar(conf, HiveConf.ConfVars.HIVESTATSDBCLASS);
StatsFactory factory = StatsFactory.newFactory(statsImpl, conf);
if (factory == null) {
throw new HiveException(ErrorMsg.STATSPUBLISHER_NOT_OBTAINED.getErrorCodedMsg());
}
// initialize stats publishing table for noscan which has only stats task
// the rest of MR task following stats task initializes it in ExecDriver.java
StatsPublisher statsPublisher = factory.getStatsPublisher();
if (!statsPublisher.init(scc)) { // creating stats table if not exists
throw new HiveException(ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
}
// manufacture a StatsAggregator
StatsAggregator statsAggregator = factory.getStatsAggregator();
if (!statsAggregator.connect(scc)) {
throw new HiveException(ErrorMsg.STATSAGGREGATOR_CONNECTION_ERROR.getErrorCodedMsg(statsImpl));
}
return statsAggregator;
}
private StatsCollectionContext getContext() throws HiveException {
StatsCollectionContext scc = new StatsCollectionContext(conf);
Task sourceTask = getWork().getSourceTask();
if (sourceTask == null) {
throw new HiveException(ErrorMsg.STATSAGGREGATOR_SOURCETASK_NULL.getErrorCodedMsg());
}
scc.setTask(sourceTask);
scc.setStatsTmpDir(this.getWork().getStatsTmpDir());
return scc;
}
private boolean existStats(Map<String, String> parameters) {
return parameters.containsKey(StatsSetupConst.ROW_COUNT)
|| parameters.containsKey(StatsSetupConst.NUM_FILES)
|| parameters.containsKey(StatsSetupConst.TOTAL_SIZE)
|| parameters.containsKey(StatsSetupConst.RAW_DATA_SIZE)
|| parameters.containsKey(StatsSetupConst.NUM_PARTITIONS);
}
private void updateStats(StatsAggregator statsAggregator,
Map<String, String> parameters, String prefix, boolean atomic)
throws HiveException {
String aggKey = prefix.endsWith(Path.SEPARATOR) ? prefix : prefix + Path.SEPARATOR;
for (String statType : StatsSetupConst.statsRequireCompute) {
String value = statsAggregator.aggregateStats(aggKey, statType);
if (value != null && !value.isEmpty()) {
long longValue = Long.parseLong(value);
if (work.getLoadTableDesc() != null &&
!work.getLoadTableDesc().getReplace()) {
String originalValue = parameters.get(statType);
if (originalValue != null) {
longValue += Long.parseLong(originalValue); // todo: invalid + valid = invalid
}
}
parameters.put(statType, String.valueOf(longValue));
} else {
if (atomic) {
throw new HiveException(ErrorMsg.STATSAGGREGATOR_MISSED_SOMESTATS, statType);
}
}
}
}
private void updateQuickStats(Warehouse wh, Map<String, String> parameters,
StorageDescriptor desc) throws MetaException {
/**
* calculate fast statistics
*/
FileStatus[] partfileStatus = wh.getFileStatusesForSD(desc);
updateQuickStats(parameters, partfileStatus);
}
private void updateQuickStats(Map<String, String> parameters,
FileStatus[] partfileStatus) throws MetaException {
MetaStoreUtils.populateQuickStats(partfileStatus, parameters);
}
private String toString(Map<String, String> parameters) {
StringBuilder builder = new StringBuilder();
for (String statType : StatsSetupConst.supportedStats) {
String value = parameters.get(statType);
if (value != null) {
if (builder.length() > 0) {
builder.append(", ");
}
builder.append(statType).append('=').append(value);
}
}
return builder.toString();
}
/**
* Get the list of partitions that need to update statistics.
* TODO: we should reuse the Partitions generated at compile time
* since getting the list of partitions is quite expensive.
*
* @return a list of partitions that need to update statistics.
* @throws HiveException
*/
private List<Partition> getPartitionsList(Hive db) throws HiveException {
if (work.getLoadFileDesc() != null) {
return null; //we are in CTAS, so we know there are no partitions
}
List<Partition> list = new ArrayList<Partition>();
if (work.getTableSpecs() != null) {
// ANALYZE command
TableSpec tblSpec = work.getTableSpecs();
table = tblSpec.tableHandle;
if (!table.isPartitioned()) {
return null;
}
// get all partitions that matches with the partition spec
List<Partition> partitions = tblSpec.partitions;
if (partitions != null) {
for (Partition partn : partitions) {
list.add(partn);
}
}
} else if (work.getLoadTableDesc() != null) {
// INSERT OVERWRITE command
LoadTableDesc tbd = work.getLoadTableDesc();
table = db.getTable(tbd.getTable().getTableName());
if (!table.isPartitioned()) {
return null;
}
DynamicPartitionCtx dpCtx = tbd.getDPCtx();
if (dpCtx != null && dpCtx.getNumDPCols() > 0) { // dynamic partitions
// If no dynamic partitions are generated, dpPartSpecs may not be initialized
if (dpPartSpecs != null) {
// load the list of DP partitions and return the list of partition specs
list.addAll(dpPartSpecs);
}
} else { // static partition
Partition partn = db.getPartition(table, tbd.getPartitionSpec(), false);
list.add(partn);
}
}
return list;
}
}