/**
* 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.optimizer;
import java.io.FileNotFoundException;
import java.io.IOException;
import java.util.ArrayList;
import java.util.HashSet;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.atomic.AtomicLong;
import com.google.common.collect.Lists;
import com.google.common.util.concurrent.ThreadFactoryBuilder;
import org.apache.hadoop.fs.LocatedFileStatus;
import org.apache.hadoop.fs.RemoteIterator;
import org.apache.hadoop.hive.common.StatsSetupConst;
import org.apache.hadoop.hive.ql.stats.StatsUtils;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.ql.exec.CommonJoinOperator;
import org.apache.hadoop.hive.ql.exec.FetchTask;
import org.apache.hadoop.hive.ql.exec.FileSinkOperator;
import org.apache.hadoop.hive.ql.exec.FilterOperator;
import org.apache.hadoop.hive.ql.exec.LimitOperator;
import org.apache.hadoop.hive.ql.exec.ListSinkOperator;
import org.apache.hadoop.hive.ql.exec.Operator;
import org.apache.hadoop.hive.ql.exec.OperatorFactory;
import org.apache.hadoop.hive.ql.exec.ReduceSinkOperator;
import org.apache.hadoop.hive.ql.exec.ScriptOperator;
import org.apache.hadoop.hive.ql.exec.SelectOperator;
import org.apache.hadoop.hive.ql.exec.TableScanOperator;
import org.apache.hadoop.hive.ql.exec.TaskFactory;
import org.apache.hadoop.hive.ql.exec.Utilities;
import org.apache.hadoop.hive.ql.hooks.ReadEntity;
import org.apache.hadoop.hive.ql.io.ContentSummaryInputFormat;
import org.apache.hadoop.hive.ql.io.HiveInputFormat;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.metadata.HiveStorageHandler;
import org.apache.hadoop.hive.ql.metadata.InputEstimator;
import org.apache.hadoop.hive.ql.metadata.Partition;
import org.apache.hadoop.hive.ql.metadata.Table;
import org.apache.hadoop.hive.ql.optimizer.ppr.PartitionPruner;
import org.apache.hadoop.hive.ql.parse.ParseContext;
import org.apache.hadoop.hive.ql.parse.PrunedPartitionList;
import org.apache.hadoop.hive.ql.parse.SemanticException;
import org.apache.hadoop.hive.ql.parse.SplitSample;
import org.apache.hadoop.hive.ql.plan.ExprNodeColumnDesc;
import org.apache.hadoop.hive.ql.plan.ExprNodeConstantDesc;
import org.apache.hadoop.hive.ql.plan.ExprNodeDesc;
import org.apache.hadoop.hive.ql.plan.ExprNodeGenericFuncDesc;
import org.apache.hadoop.hive.ql.plan.FetchWork;
import org.apache.hadoop.hive.ql.plan.ListSinkDesc;
import org.apache.hadoop.hive.ql.plan.OperatorDesc;
import org.apache.hadoop.hive.ql.plan.PartitionDesc;
import org.apache.hadoop.hive.ql.plan.PlanUtils;
import org.apache.hadoop.hive.ql.plan.SelectDesc;
import org.apache.hadoop.hive.ql.plan.TableDesc;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDF;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToBinary;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToChar;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToDate;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToDecimal;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToUnixTimeStamp;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToUtcTimestamp;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDFToVarchar;
import org.apache.hadoop.mapred.InputFormat;
import org.apache.hadoop.mapred.JobConf;
/**
* Tries to convert simple fetch query to single fetch task, which fetches rows directly
* from location of table/partition.
*/
public class SimpleFetchOptimizer extends Transform {
private final Logger LOG = LoggerFactory.getLogger(SimpleFetchOptimizer.class.getName());
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
Map<String, TableScanOperator> topOps = pctx.getTopOps();
if (pctx.getQueryProperties().isQuery() && !pctx.getQueryProperties().isAnalyzeCommand()
&& topOps.size() == 1) {
// no join, no groupby, no distinct, no lateral view, no subq,
// no CTAS or insert, not analyze command, and single sourced.
String alias = (String) pctx.getTopOps().keySet().toArray()[0];
TableScanOperator topOp = pctx.getTopOps().values().iterator().next();
try {
FetchTask fetchTask = optimize(pctx, alias, topOp);
if (fetchTask != null) {
pctx.setFetchTask(fetchTask);
}
} catch (Exception e) {
// Has to use full name to make sure it does not conflict with
// org.apache.commons.lang.StringUtils
LOG.error(org.apache.hadoop.util.StringUtils.stringifyException(e));
if (e instanceof SemanticException) {
throw (SemanticException) e;
}
throw new SemanticException(e.getMessage(), e);
}
}
return pctx;
}
// returns non-null FetchTask instance when succeeded
@SuppressWarnings("unchecked")
private FetchTask optimize(ParseContext pctx, String alias, TableScanOperator source)
throws Exception {
String mode = HiveConf.getVar(
pctx.getConf(), HiveConf.ConfVars.HIVEFETCHTASKCONVERSION);
boolean aggressive = "more".equals(mode);
final int limit = pctx.getQueryProperties().getOuterQueryLimit();
// limit = 0 means that we do not need any task.
if (limit == 0) {
return null;
}
FetchData fetch = checkTree(aggressive, pctx, alias, source);
if (fetch != null && checkThreshold(fetch, limit, pctx)) {
FetchWork fetchWork = fetch.convertToWork();
FetchTask fetchTask = (FetchTask) TaskFactory.get(fetchWork, pctx.getConf());
fetchWork.setSink(fetch.completed(pctx, fetchWork));
fetchWork.setSource(source);
fetchWork.setLimit(limit);
return fetchTask;
}
return null;
}
private boolean checkThreshold(FetchData data, int limit, ParseContext pctx) throws Exception {
if (limit > 0) {
if (data.hasOnlyPruningFilter()) {
/* partitioned table + query has only pruning filters */
return true;
} else if (data.isPartitioned() == false && data.isFiltered() == false) {
/* unpartitioned table + no filters */
return true;
}
/* fall through */
}
long threshold = HiveConf.getLongVar(pctx.getConf(),
HiveConf.ConfVars.HIVEFETCHTASKCONVERSIONTHRESHOLD);
if (threshold < 0) {
return true;
}
Operator child = data.scanOp.getChildOperators().get(0);
if(child instanceof SelectOperator) {
// select *, constant and casts can be allowed without a threshold check
if (checkExpressions((SelectOperator)child)) {
return true;
}
}
return data.isDataLengthWithInThreshold(pctx, threshold);
}
// all we can handle is LimitOperator, FilterOperator SelectOperator and final FS
//
// for non-aggressive mode (minimal)
// 1. sampling is not allowed
// 2. for partitioned table, all filters should be targeted to partition column
// 3. SelectOperator should use only simple cast/column access
private FetchData checkTree(boolean aggressive, ParseContext pctx, String alias,
TableScanOperator ts) throws HiveException {
SplitSample splitSample = pctx.getNameToSplitSample().get(alias);
if (!aggressive && splitSample != null) {
return null;
}
if (!aggressive && ts.getConf().getTableSample() != null) {
return null;
}
Table table = ts.getConf().getTableMetadata();
if (table == null) {
return null;
}
ReadEntity parent = PlanUtils.getParentViewInfo(alias, pctx.getViewAliasToInput());
if (!table.isPartitioned()) {
FetchData fetch = new FetchData(ts, parent, table, splitSample);
return checkOperators(fetch, aggressive, false);
}
boolean bypassFilter = false;
if (HiveConf.getBoolVar(pctx.getConf(), HiveConf.ConfVars.HIVEOPTPPD)) {
ExprNodeDesc pruner = pctx.getOpToPartPruner().get(ts);
if (PartitionPruner.onlyContainsPartnCols(table, pruner)) {
bypassFilter = !pctx.getPrunedPartitions(alias, ts).hasUnknownPartitions();
}
}
if (!aggressive && !bypassFilter) {
return null;
}
PrunedPartitionList partitions = pctx.getPrunedPartitions(alias, ts);
FetchData fetch = new FetchData(ts, parent, table, partitions, splitSample, bypassFilter);
return checkOperators(fetch, aggressive, bypassFilter);
}
private FetchData checkOperators(FetchData fetch, boolean aggressive, boolean bypassFilter) {
if (aggressive) {
return isConvertible(fetch) ? fetch : null;
}
return checkOperators(fetch, fetch.scanOp, bypassFilter);
}
private FetchData checkOperators(FetchData fetch, TableScanOperator ts, boolean bypassFilter) {
if (ts.getChildOperators().size() != 1) {
return null;
}
Operator<?> op = ts.getChildOperators().get(0);
for (; ; op = op.getChildOperators().get(0)) {
if (op instanceof SelectOperator) {
if (!checkExpressions((SelectOperator) op)) {
return null;
}
continue;
}
if (!(op instanceof LimitOperator || (op instanceof FilterOperator && bypassFilter))) {
break;
}
if (op.getChildOperators() == null || op.getChildOperators().size() != 1) {
return null;
}
if (op instanceof FilterOperator) {
fetch.setFiltered(true);
}
}
if (op instanceof FileSinkOperator) {
fetch.fileSink = op;
return fetch;
}
return null;
}
private boolean checkExpressions(SelectOperator op) {
SelectDesc desc = op.getConf();
if (desc.isSelectStar() || desc.isSelStarNoCompute()) {
return true;
}
for (ExprNodeDesc expr : desc.getColList()) {
if (!checkExpression(expr)) {
return false;
}
}
return true;
}
private boolean checkExpression(ExprNodeDesc expr) {
if (expr instanceof ExprNodeConstantDesc ||
expr instanceof ExprNodeColumnDesc) {
return true;
}
if (expr instanceof ExprNodeGenericFuncDesc) {
GenericUDF udf = ((ExprNodeGenericFuncDesc) expr).getGenericUDF();
if (udf instanceof GenericUDFToBinary || udf instanceof GenericUDFToChar
|| udf instanceof GenericUDFToDate || udf instanceof GenericUDFToDecimal
|| udf instanceof GenericUDFToUnixTimeStamp || udf instanceof GenericUDFToUtcTimestamp
|| udf instanceof GenericUDFToVarchar) {
return expr.getChildren().size() == 1 && checkExpression(expr.getChildren().get(0));
}
}
return false;
}
private boolean isConvertible(FetchData fetch) {
return isConvertible(fetch, fetch.scanOp, new HashSet<Operator<?>>());
}
private boolean isConvertible(FetchData fetch, Operator<?> operator, Set<Operator<?>> traversed) {
if (operator instanceof ReduceSinkOperator || operator instanceof CommonJoinOperator
|| operator instanceof ScriptOperator) {
return false;
}
if (operator instanceof FilterOperator) {
fetch.setFiltered(true);
}
if (!traversed.add(operator)) {
return true;
}
if (operator.getNumChild() == 0) {
if (operator instanceof FileSinkOperator) {
fetch.fileSink = operator;
return true;
}
return false;
}
for (Operator<?> child : operator.getChildOperators()) {
if (!traversed.containsAll(child.getParentOperators())){
continue;
}
if (!isConvertible(fetch, child, traversed)) {
return false;
}
}
return true;
}
enum Status {
PASS,
FAIL,
UNAVAILABLE
}
private class FetchData {
// source table scan
private final TableScanOperator scanOp;
private final ReadEntity parent;
private final Table table;
private final SplitSample splitSample;
private final PrunedPartitionList partsList;
private final Set<ReadEntity> inputs = new LinkedHashSet<ReadEntity>();
private final boolean onlyPruningFilter;
// this is always non-null when conversion is completed
private Operator<?> fileSink;
private boolean filtered;
private FetchData(TableScanOperator scanOp, ReadEntity parent, Table table, SplitSample splitSample) {
this.scanOp = scanOp;
this.parent = parent;
this.table = table;
this.partsList = null;
this.splitSample = splitSample;
this.onlyPruningFilter = false;
}
private FetchData(TableScanOperator scanOp, ReadEntity parent, Table table, PrunedPartitionList partsList,
SplitSample splitSample, boolean bypassFilter) {
this.scanOp = scanOp;
this.parent = parent;
this.table = table;
this.partsList = partsList;
this.splitSample = splitSample;
this.onlyPruningFilter = bypassFilter;
}
/*
* all filters were executed during partition pruning
*/
public final boolean hasOnlyPruningFilter() {
return this.onlyPruningFilter;
}
public final boolean isPartitioned() {
return this.table.isPartitioned();
}
/* there are filter operators in the pipeline */
public final boolean isFiltered() {
return this.filtered;
}
public final void setFiltered(boolean filtered) {
this.filtered = filtered;
}
private FetchWork convertToWork() throws HiveException {
inputs.clear();
Utilities.addSchemaEvolutionToTableScanOperator(table, scanOp);
TableDesc tableDesc = Utilities.getTableDesc(table);
if (!table.isPartitioned()) {
inputs.add(new ReadEntity(table, parent, !table.isView() && parent == null));
FetchWork work = new FetchWork(table.getPath(), tableDesc);
PlanUtils.configureInputJobPropertiesForStorageHandler(work.getTblDesc());
work.setSplitSample(splitSample);
return work;
}
List<Path> listP = new ArrayList<Path>();
List<PartitionDesc> partP = new ArrayList<PartitionDesc>();
for (Partition partition : partsList.getNotDeniedPartns()) {
inputs.add(new ReadEntity(partition, parent, parent == null));
listP.add(partition.getDataLocation());
partP.add(Utilities.getPartitionDescFromTableDesc(tableDesc, partition, true));
}
Table sourceTable = partsList.getSourceTable();
inputs.add(new ReadEntity(sourceTable, parent, parent == null));
TableDesc table = Utilities.getTableDesc(sourceTable);
FetchWork work = new FetchWork(listP, partP, table);
if (!work.getPartDesc().isEmpty()) {
PartitionDesc part0 = work.getPartDesc().get(0);
PlanUtils.configureInputJobPropertiesForStorageHandler(part0.getTableDesc());
work.setSplitSample(splitSample);
}
return work;
}
// this optimizer is for replacing FS to temp+fetching from temp with
// single direct fetching, which means FS is not needed any more when conversion completed.
// rows forwarded will be received by ListSinkOperator, which is replacing FS
private ListSinkOperator completed(ParseContext pctx, FetchWork work) {
for (ReadEntity input : inputs) {
PlanUtils.addInput(pctx.getSemanticInputs(), input);
}
return replaceFSwithLS(fileSink, work.getSerializationNullFormat());
}
private boolean isDataLengthWithInThreshold(ParseContext pctx, final long threshold)
throws Exception {
if (splitSample != null && splitSample.getTotalLength() != null) {
if (LOG.isDebugEnabled()) {
LOG.debug("Threshold " + splitSample.getTotalLength() + " exceeded for pseudoMR mode");
}
return (threshold - splitSample.getTotalLength()) > 0;
}
Status status = checkThresholdWithMetastoreStats(table, partsList, threshold);
if (status.equals(Status.PASS)) {
return true;
} else if (status.equals(Status.FAIL)) {
return false;
} else {
LOG.info("Cannot fetch stats from metastore for table: {}. Falling back to filesystem scan..",
table.getCompleteName());
// metastore stats is unavailable, fallback to old way
final JobConf jobConf = new JobConf(pctx.getConf());
Utilities.setColumnNameList(jobConf, scanOp, true);
Utilities.setColumnTypeList(jobConf, scanOp, true);
HiveStorageHandler handler = table.getStorageHandler();
if (handler instanceof InputEstimator) {
InputEstimator estimator = (InputEstimator) handler;
TableDesc tableDesc = Utilities.getTableDesc(table);
PlanUtils.configureInputJobPropertiesForStorageHandler(tableDesc);
Utilities.copyTableJobPropertiesToConf(tableDesc, jobConf);
long len = estimator.estimate(jobConf, scanOp, threshold).getTotalLength();
if (LOG.isDebugEnabled()) {
LOG.debug("Threshold " + len + " exceeded for pseudoMR mode");
}
return (threshold - len) > 0;
}
if (table.isNonNative()) {
return true; // nothing can be done
}
if (!table.isPartitioned()) {
long len = getPathLength(jobConf, table.getPath(), table.getInputFormatClass(), threshold);
if (LOG.isDebugEnabled()) {
LOG.debug("Threshold " + len + " exceeded for pseudoMR mode");
}
return (threshold - len) > 0;
}
final AtomicLong total = new AtomicLong(0);
//TODO: use common thread pool later?
int threadCount = HiveConf.getIntVar(pctx.getConf(),
HiveConf.ConfVars.HIVE_STATS_GATHER_NUM_THREADS);
final ExecutorService pool = (threadCount > 0) ?
Executors.newFixedThreadPool(threadCount,
new ThreadFactoryBuilder()
.setDaemon(true)
.setNameFormat("SimpleFetchOptimizer-FileLength-%d").build()) : null;
try {
List<Future> futures = Lists.newLinkedList();
for (final Partition partition : partsList.getNotDeniedPartns()) {
final Path path = partition.getDataLocation();
if (pool != null) {
futures.add(pool.submit(new Callable<Long>() {
@Override
public Long call() throws Exception {
long len = getPathLength(jobConf, path, partition.getInputFormatClass(), threshold);
LOG.trace(path + ", length=" + len);
return total.addAndGet(len);
}
}));
} else {
total.addAndGet(getPathLength(jobConf, path, partition.getInputFormatClass(), threshold));
}
}
if (pool != null) {
pool.shutdown();
for (Future<Long> future : futures) {
long totalLen = future.get();
if ((threshold - totalLen) <= 0) {
// early exit, as getting file lengths can be expensive in object stores.
return false;
}
}
}
return (threshold - total.get()) >= 0;
} finally {
LOG.info("Data set size=" + total.get() + ", threshold=" + threshold);
if (pool != null) {
pool.shutdownNow();
}
}
}
}
// This method gets the basic stats from metastore for table/partitions. This will make use of the statistics from
// AnnotateWithStatistics optimizer when available. If execution engine is tez or spark, AnnotateWithStatistics
// optimization is applied only during physical compilation because of DPP changing the stats. In such case, we
// we will get the basic stats from metastore. When statistics is absent in metastore we will use the fallback of
// scanning the filesystem to get file lengths.
private Status checkThresholdWithMetastoreStats(final Table table, final PrunedPartitionList partsList,
final long threshold) {
if (table != null && !table.isPartitioned()) {
long dataSize = StatsUtils.getTotalSize(table);
if (dataSize <= 0) {
LOG.warn("Cannot determine basic stats for table: {} from metastore. Falling back.", table.getCompleteName());
return Status.UNAVAILABLE;
}
return (threshold - dataSize) >= 0 ? Status.PASS : Status.FAIL;
} else if (table != null && table.isPartitioned() && partsList != null) {
List<Long> dataSizes = StatsUtils.getBasicStatForPartitions(table, partsList.getNotDeniedPartns(),
StatsSetupConst.TOTAL_SIZE);
long totalDataSize = StatsUtils.getSumIgnoreNegatives(dataSizes);
if (totalDataSize <= 0) {
LOG.warn("Cannot determine basic stats for partitioned table: {} from metastore. Falling back.",
table.getCompleteName());
return Status.UNAVAILABLE;
}
return (threshold - totalDataSize) >= 0 ? Status.PASS : Status.FAIL;
}
return Status.UNAVAILABLE;
}
private long getPathLength(JobConf conf, Path path,
Class<? extends InputFormat> clazz, long threshold)
throws IOException {
if (ContentSummaryInputFormat.class.isAssignableFrom(clazz)) {
InputFormat input = HiveInputFormat.getInputFormatFromCache(clazz, conf);
return ((ContentSummaryInputFormat)input).getContentSummary(path, conf).getLength();
} else {
FileSystem fs = path.getFileSystem(conf);
try {
long length = 0;
RemoteIterator<LocatedFileStatus> results = fs.listFiles(path, true);
// No need to iterate more, when threshold is reached
// (beneficial especially for object stores)
while (length <= threshold && results.hasNext()) {
length += results.next().getLen();
}
LOG.trace("length=" + length + ", threshold=" + threshold);
return length;
} catch (FileNotFoundException e) {
return 0;
}
}
}
}
public static ListSinkOperator replaceFSwithLS(Operator<?> fileSink, String nullFormat) {
ListSinkDesc desc = new ListSinkDesc(nullFormat);
ListSinkOperator sink = (ListSinkOperator) OperatorFactory.get(
fileSink.getCompilationOpContext(), desc);
sink.setParentOperators(new ArrayList<Operator<? extends OperatorDesc>>());
Operator<? extends OperatorDesc> parent = fileSink.getParentOperators().get(0);
sink.getParentOperators().add(parent);
parent.replaceChild(fileSink, sink);
fileSink.setParentOperators(null);
return sink;
}
}