blob: 193ce6b3d7935af87e2df387e946d3bd4ef7cf17 [file] [log] [blame]
/*
* 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.lens.cube.parse;
import static java.util.Comparator.naturalOrder;
import static org.apache.lens.cube.parse.CandidateTablePruneCause.CandidateTablePruneCode;
import static org.apache.lens.cube.parse.CandidateTablePruneCause.SkipUpdatePeriodCode;
import static org.apache.lens.cube.parse.CandidateTablePruneCause.timeDimNotSupported;
import static org.apache.lens.cube.parse.StorageUtil.getFallbackRange;
import static org.apache.lens.cube.parse.StorageUtil.processCubeColForDataCompleteness;
import static org.apache.lens.cube.parse.StorageUtil.processExpressionsForCompleteness;
import java.text.DateFormat;
import java.text.SimpleDateFormat;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.Date;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.LinkedHashMap;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.OptionalDouble;
import java.util.Set;
import java.util.TimeZone;
import java.util.TreeSet;
import java.util.stream.Stream;
import org.apache.lens.cube.metadata.*;
import org.apache.lens.server.api.error.LensException;
import org.apache.lens.server.api.metastore.DataCompletenessChecker;
import org.apache.commons.lang.StringUtils;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.session.SessionState;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Sets;
import lombok.Getter;
import lombok.Setter;
import lombok.extern.slf4j.Slf4j;
/**
* Represents a fact on a storage table and the dimensions it needs to be joined with to answer the query
*/
@Slf4j
public class StorageCandidate implements Candidate, CandidateTable {
// TODO union : Put comments on member variables.
@Getter
private final CubeQueryContext cubeQueryContext;
private final String processTimePartCol;
private final String completenessPartCol;
private final float completenessThreshold;
/**
* Name of this storage candidate = storageName_factName
*/
@Getter
@Setter
private String name;
/**
* This is the storage table specific name. It is used while generating query from this candidate
*/
@Setter
private String resolvedName;
/**
* Valid update periods populated by Phase 1.
*/
@Getter
private TreeSet<UpdatePeriod> validUpdatePeriods = new TreeSet<>();
/**
* These are the update periods that finally participate in partitions.
* @see #getParticipatingPartitions()
*/
@Getter
private TreeSet<UpdatePeriod> participatingUpdatePeriods = new TreeSet<>();
@Getter
@Setter
Map<String, SkipUpdatePeriodCode> updatePeriodRejectionCause;
@Getter
Set<Dimension> queriedDims = Sets.newHashSet();
private Collection<StorageCandidate> periodSpecificStorageCandidates;
/**
* This map holds Tags (A tag refers to one or more measures) that have incomplete (below configured threshold) data.
* Value is a map of date string and %completeness.
*/
@Getter
private Map<String, Map<String, Float>> dataCompletenessMap = new HashMap<>();
private SimpleDateFormat partWhereClauseFormat = null;
/**
* Participating fact, storage and dimensions for this StorageCandidate
*/
@Getter
private FactTable fact;
@Getter
private String storageName;
@Getter
private String storageTable;
@Getter
private Map<TimeRange, Set<FactPartition>> rangeToPartitions = new LinkedHashMap<>();
@Getter
private Map<TimeRange, String> rangeToExtraWhereFallBack = new LinkedHashMap<>();
@Getter
private Set<Integer> answerableMeasurePhraseIndices = Sets.newHashSet();
@Getter
@Setter
private String fromString;
@Getter
private CubeInterface cube;
@Getter
private Date startTime;
@Getter
private Date endTime;
/**
* Cached fact columns
*/
private Collection<String> factColumns;
/**
* Non existing partitions
*/
@Getter
private Set<String> nonExistingPartitions = new HashSet<>();
/**
* This will be true if this storage candidate has multiple storage tables (one per update period)
* https://issues.apache.org/jira/browse/LENS-1386
*/
@Getter
private boolean isStorageTblsAtUpdatePeriodLevel;
@Getter
private int numQueriedParts = 0;
public StorageCandidate(StorageCandidate sc) throws LensException {
this(sc.getCube(), sc.getFact(), sc.getStorageName(), sc.getCubeQueryContext());
this.validUpdatePeriods.addAll(sc.getValidUpdatePeriods());
this.fromString = sc.fromString;
this.factColumns = sc.factColumns;
this.answerableMeasurePhraseIndices.addAll(sc.answerableMeasurePhraseIndices);
for (Map.Entry<TimeRange, Set<FactPartition>> entry : sc.getRangeToPartitions().entrySet()) {
rangeToPartitions.put(entry.getKey(), new LinkedHashSet<>(entry.getValue()));
}
this.rangeToExtraWhereFallBack = sc.rangeToExtraWhereFallBack;
}
public StorageCandidate(CubeInterface cube, FactTable fact, String storageName, CubeQueryContext cubeQueryContext)
throws LensException {
this.cube = cube;
this.fact = fact;
this.cubeQueryContext = cubeQueryContext;
if ((getCube() == null) || (fact == null) || (storageName == null)) {
throw new IllegalArgumentException("Cube,fact and storageName should be non null");
}
this.storageName = storageName;
this.storageTable = MetastoreUtil.getFactOrDimtableStorageTableName(fact.getSourceFactName(), storageName);
this.name = getFact().getName();
this.processTimePartCol = getConf().get(CubeQueryConfUtil.PROCESS_TIME_PART_COL);
String formatStr = getConf().get(CubeQueryConfUtil.PART_WHERE_CLAUSE_DATE_FORMAT);
if (formatStr != null) {
this.partWhereClauseFormat = new SimpleDateFormat(formatStr);
}
completenessPartCol = getConf().get(CubeQueryConfUtil.COMPLETENESS_CHECK_PART_COL);
completenessThreshold = getConf()
.getFloat(CubeQueryConfUtil.COMPLETENESS_THRESHOLD, CubeQueryConfUtil.DEFAULT_COMPLETENESS_THRESHOLD);
Set<String> storageTblNames = getCubeMetastoreClient().getStorageTables(fact, storageName);
isStorageTblsAtUpdatePeriodLevel = storageTblNames.size() > 1
|| !storageTblNames.iterator().next().equalsIgnoreCase(storageTable);
setStorageStartAndEndDate();
}
String getTimeRangeWhereClasue(TimeRangeWriter rangeWriter, TimeRange range)
throws LensException {
String rangeWhere = rangeWriter.getTimeRangeWhereClause(
getCubeQueryContext(), getCubeQueryContext().getAliasForTableName(getCube().getName()),
getRangeToPartitions().get(range));
String fallback = getRangeToExtraWhereFallBack().get(range);
if (StringUtils.isNotBlank(fallback)){
rangeWhere = "((" + rangeWhere + ") and (" + fallback + "))";
}
return rangeWhere;
}
/**
* Sets Storage candidates start and end time based on underlying storage-tables
*
* CASE 1
* If has Storage has single storage table*
* Storage start time = max(storage start time , fact start time)
* Storage end time = min(storage end time , fact start time)
*
* CASE 2
* If the Storage has multiple Storage Tables (one per update period)*
* update Period start Time = Max(update start time, fact start time)
* update Period end Time = Min(update end time, fact end time)
* Stoarge start and end time is derived form the underlying update period start and end times.
* Storage start time = min(update1 start time ,...., updateN start time)
* Storage end time = max(update1 end time ,...., updateN end time)
*
* Note in Case 2 its assumed that the time range supported by different update periods are either
* overlapping(Example 2) or form a non overlapping but continuous chain(Example 1) as illustrated
* in examples below
*
* Example 1
* A Storage has 2 Non Oevralpping but continuous Update Periods.
* MONTHLY with start time as now.month -13 months and end time as now.month -2months and
* DAILY with start time as now.day and end time as now.month -2months
* Then this Sorage will have an implied start time as now.month -13 month and end time as now.day
*
* Example 2
* A Storage has 2 Overlapping Update Periods.
* MONTHLY with start time as now.month -13 months and end time as now.month -1months and
* DAILY with start time as now.day and end time as now.month -2months
* Then this Sorage will have an implied start time as now.month -13 month and end time as now.day
*
* @throws LensException
*/
void setStorageStartAndEndDate() throws LensException {
if (this.startTime != null && !this.isStorageTblsAtUpdatePeriodLevel) {
//If the times are already set and are not dependent of update period, no point setting times again.
return;
}
List<Date> startDates = new ArrayList<>();
List<Date> endDates = new ArrayList<>();
for (String storageTablePrefix : getValidStorageTableNames()) {
startDates.add(getCubeMetastoreClient().getStorageTableStartDate(storageTablePrefix, fact));
endDates.add(getCubeMetastoreClient().getStorageTableEndDate(storageTablePrefix, fact));
}
this.startTime = Collections.min(startDates);
this.endTime = Collections.max(endDates);
}
private Set<String> getValidStorageTableNames() throws LensException {
if (!validUpdatePeriods.isEmpty()) {
// In this case skip invalid update periods and get storage tables only for valid ones.
Set<String> uniqueStorageTables = new HashSet<>();
for (UpdatePeriod updatePeriod : validUpdatePeriods) {
uniqueStorageTables.add(
getCubeMetastoreClient().getStorageTableName(fact.getSourceFactName(), storageName, updatePeriod)
);
}
return uniqueStorageTables;
} else {
//Get all storage tables.
return getCubeMetastoreClient().getStorageTables(fact, storageName);
}
}
public void addAnswerableMeasurePhraseIndices(int index) {
answerableMeasurePhraseIndices.add(index);
}
@Override
public Candidate explode() throws LensException {
if (splitAtUpdatePeriodLevelIfReq().size() > 1) {
return new UnionCandidate(splitAtUpdatePeriodLevelIfReq(), getCubeQueryContext());
} else {
return splitAtUpdatePeriodLevelIfReq().iterator().next();
}
}
@Override
public String getStorageString(String alias) {
return storageName + " " + alias;
}
@Override
public AbstractCubeTable getBaseTable() {
return (AbstractCubeTable) cube;
}
@Override
public StorageCandidate copy() throws LensException {
return new StorageCandidate(this);
}
@Override
public boolean isPhraseAnswerable(QueriedPhraseContext phrase) throws LensException {
return phrase.isEvaluable(this);
}
@Override
public AbstractCubeTable getTable() {
return (AbstractCubeTable) fact;
}
public Optional<Date> getColumnStartTime(String column) {
Date startTime = null;
Map<String, String> propertiesMap = this.getFact().getSourceFactProperties();
for (String key : propertiesMap.keySet()) {
if (key.contains(MetastoreConstants.FACT_COL_START_TIME_PFX)) {
String propCol = StringUtils.substringAfter(key, MetastoreConstants.FACT_COL_START_TIME_PFX);
if (column.equals(propCol)) {
startTime = MetastoreUtil.getDateFromProperty(propertiesMap.get(key), false, true);
}
}
}
return Optional.ofNullable(startTime);
}
@Override
public Optional<Date> getColumnEndTime(String column) {
Date endTime = null;
Map<String, String> propertiesMap = this.getFact().getSourceFactProperties();
for (String key : propertiesMap.keySet()) {
if (key.contains(MetastoreConstants.FACT_COL_END_TIME_PFX)) {
String propCol = StringUtils.substringAfter(key, MetastoreConstants.FACT_COL_END_TIME_PFX);
if (column.equals(propCol)) {
endTime = MetastoreUtil.getDateFromProperty(propertiesMap.get(key), false, true);
}
}
}
return Optional.ofNullable(endTime);
}
@Override
public Collection<String> getColumns() {
if (factColumns == null) {
factColumns = fact.getValidColumns();
if (factColumns == null) {
factColumns = fact.getAllFieldNames();
}
}
return factColumns;
}
@Override
public OptionalDouble getCost() {
return OptionalDouble.of(fact.weight());
}
@Override
public boolean contains(Candidate candidate) {
return this.equals(candidate);
}
@Override
public Collection<Candidate> getChildren() {
return null;
}
private void updatePartitionStorage(FactPartition part) throws LensException {
try {
if (getCubeMetastoreClient().factPartitionExists(fact, part, storageTable)) {
part.getStorageTables().add(storageTable);
part.setFound(true);
}
} catch (HiveException e) {
log.warn("Hive exception while getting storage table partition", e);
}
}
/**
* Gets FactPartitions for the given fact using the following logic
*
* 1. Find the max update interval that will be used for the query. Lets assume time
* range is 15 Sep to 15 Dec and the fact has two storage with update periods as MONTHLY,DAILY,HOURLY.
* In this case the data for [15 sep - 1 oct)U[1 Dec - 15 Dec) will be answered by DAILY partitions
* and [1 oct - 1Dec) will be answered by MONTHLY partitions. The max interavl for this query will be MONTHLY.
*
* 2.Prune Storgaes that do not fall in the queries time range.
* {@link org.apache.lens.cube.metadata.CubeMetastoreClient#isStorageTableCandidateForRange(String, Date, Date)}
*
* 3. Iterate over max interavl . In out case it will give two months Oct and Nov. Find partitions for
* these two months.Check validity of FactPartitions for Oct and Nov
* via {@link #updatePartitionStorage(FactPartition)}.
* If the partition is missing, try getting partitions for the time range form other update periods (DAILY,HOURLY).
* This is achieved by calling getPartitions() recursively but passing only 2 update periods (DAILY,HOURLY)
*
* 4.If the monthly partitions are found, check for lookahead partitions and call getPartitions recursively for the
* remaining time intervals i.e, [15 sep - 1 oct) and [1 Dec - 15 Dec)
*
* TODO union : Move this into util.
*/
private boolean getPartitions(Date fromDate, Date toDate, String partCol, Set<FactPartition> partitions,
TreeSet<UpdatePeriod> updatePeriods, boolean addNonExistingParts, boolean failOnPartialData,
PartitionRangesForPartitionColumns missingPartitions) throws LensException {
if (fromDate.equals(toDate) || fromDate.after(toDate)) {
return true;
}
if (updatePeriods == null || updatePeriods.isEmpty()) {
return false;
}
UpdatePeriod maxInterval = CubeFactTable.maxIntervalInRange(fromDate, toDate, updatePeriods);
if (maxInterval == null) {
log.info("No max interval for range: {} to {}", fromDate, toDate);
return false;
}
if (maxInterval == UpdatePeriod.CONTINUOUS
&& cubeQueryContext.getRangeWriter().getClass().equals(BetweenTimeRangeWriter.class)) {
FactPartition part = new FactPartition(partCol, fromDate, maxInterval, null, partWhereClauseFormat);
partitions.add(part);
part.getStorageTables().add(storageTable);
part = new FactPartition(partCol, toDate, maxInterval, null, partWhereClauseFormat);
partitions.add(part);
part.getStorageTables().add(storageTable);
this.participatingUpdatePeriods.add(maxInterval);
log.info("Added continuous fact partition for storage table {}", storageName);
return true;
}
if (!getCubeMetastoreClient().partColExists(this.getFact(), storageName, partCol)) {
log.info("{} does not exist in {}", partCol, name);
return false;
}
Date maxIntervalStorageTblStartDate = getStorageTableStartDate(maxInterval);
Date maxIntervalStorageTblEndDate = getStorageTableEndDate(maxInterval);
TreeSet<UpdatePeriod> remainingIntervals = new TreeSet<>(updatePeriods);
remainingIntervals.remove(maxInterval);
if (!isCandidatePartiallyValidForTimeRange(
maxIntervalStorageTblStartDate, maxIntervalStorageTblEndDate, fromDate, toDate)) {
//Check the time range in remainingIntervals as maxInterval is not useful
return getPartitions(fromDate, toDate, partCol, partitions, remainingIntervals,
addNonExistingParts, failOnPartialData, missingPartitions);
}
Date ceilFromDate = DateUtil.getCeilDate(fromDate.after(maxIntervalStorageTblStartDate)
? fromDate : maxIntervalStorageTblStartDate, maxInterval);
Date floorToDate = DateUtil.getFloorDate(toDate.before(maxIntervalStorageTblEndDate)
? toDate : maxIntervalStorageTblEndDate, maxInterval);
if (ceilFromDate.equals(floorToDate) || floorToDate.before(ceilFromDate)) {
return getPartitions(fromDate, toDate, partCol, partitions, remainingIntervals,
addNonExistingParts, failOnPartialData, missingPartitions);
}
int lookAheadNumParts = getConf()
.getInt(CubeQueryConfUtil.getLookAheadPTPartsKey(maxInterval), CubeQueryConfUtil.DEFAULT_LOOK_AHEAD_PT_PARTS);
int lookAheadNumTimeParts = getConf()
.getInt(CubeQueryConfUtil.getLookAheadTimePartsKey(maxInterval), CubeQueryConfUtil.DEFAULT_LOOK_AHEAD_TIME_PARTS);
TimeRange.Iterable.Iterator iter = TimeRange.iterable(ceilFromDate, floorToDate, maxInterval, 1).iterator();
// add partitions from ceilFrom to floorTo
while (iter.hasNext()) {
Date dt = iter.next();
Date nextDt = iter.peekNext();
FactPartition part = new FactPartition(partCol, dt, maxInterval, null, partWhereClauseFormat);
updatePartitionStorage(part);
log.debug("Storage tables containing Partition {} are: {}", part, part.getStorageTables());
if (part.isFound()) {
log.debug("Adding existing partition {}", part);
partitions.add(part);
this.participatingUpdatePeriods.add(maxInterval);
log.debug("Looking for look ahead process time partitions for {}", part);
if (processTimePartCol == null) {
log.debug("processTimePartCol is null");
} else if (partCol.equals(processTimePartCol)) {
log.debug("part column is process time col");
} else if (updatePeriods.first().equals(maxInterval)) {
log.debug("Update period is the least update period");
} else if ((iter.getNumIters() - iter.getCounter()) > lookAheadNumTimeParts) {
// see if this is the part of the last-n look ahead partitions
log.debug("Not a look ahead partition");
} else {
log.debug("Looking for look ahead process time partitions for {}", part);
// check if finer partitions are required
// final partitions are required if no partitions from
// look-ahead
// process time are present
TimeRange.Iterable.Iterator processTimeIter = TimeRange.iterable(nextDt, lookAheadNumParts, maxInterval, 1)
.iterator();
TimeRange.Iterable.Iterator timeIter = TimeRange.iterable(nextDt, lookAheadNumTimeParts, maxInterval, 1)
.iterator();
while (processTimeIter.hasNext()) {
Date pdt = processTimeIter.next();
Date nextPdt = processTimeIter.peekNext();
FactPartition currFactPartition;
boolean allProcessTimePartitionsFound = true;
while (timeIter.hasNext()){
Date date = timeIter.next();
currFactPartition = new FactPartition(processTimePartCol, date, maxInterval, null,
partWhereClauseFormat);
updatePartitionStorage(currFactPartition);
if (!currFactPartition.isFound()) {
log.debug("Looked ahead process time partition {} is not found : " + currFactPartition);
allProcessTimePartitionsFound = false;
break;
}
}
if (allProcessTimePartitionsFound) {
log.debug("Finer parts not required for look-ahead partition :{}", part);
} else {
TreeSet<UpdatePeriod> newset = new TreeSet<UpdatePeriod>();
newset.addAll(updatePeriods);
newset.remove(maxInterval);
log.debug("newset of update periods:{}", newset);
if (!newset.isEmpty()) {
// Get partitions for look ahead process time
log.debug("Looking for process time partitions between {} and {}", pdt, nextPdt);
Set<FactPartition> processTimeParts = getPartitions(
TimeRange.builder().fromDate(pdt).toDate(nextPdt).partitionColumn(processTimePartCol).build(),
newset, true, failOnPartialData, missingPartitions);
log.debug("Look ahead partitions: {}", processTimeParts);
TimeRange timeRange = TimeRange.builder().fromDate(dt).toDate(nextDt).build();
for (FactPartition pPart : processTimeParts) {
log.debug("Looking for finer partitions in pPart: {}", pPart);
for (Date date : timeRange.iterable(pPart.getPeriod(), 1)) {
FactPartition innerPart = new FactPartition(partCol, date, pPart.getPeriod(), pPart,
partWhereClauseFormat);
updatePartitionStorage(innerPart);
innerPart.setFound(pPart.isFound());
if (innerPart.isFound() || !failOnPartialData) {
this.participatingUpdatePeriods.add(maxInterval);
partitions.add(innerPart);
}
}
log.debug("added all sub partitions blindly in pPart: {}", pPart);
}
}
}
}
}
} else {
log.info("Partition:{} does not exist in any storage table", part);
if (!getPartitions(dt, nextDt, partCol, partitions, remainingIntervals, false, failOnPartialData,
missingPartitions)) {
log.debug("Adding non existing partition {}", part);
if (addNonExistingParts) {
// Add non existing partitions for all cases of whether we populate all non existing or not.
missingPartitions.add(part);
if (!failOnPartialData) {
this.participatingUpdatePeriods.add(maxInterval);
partitions.add(part);
part.getStorageTables().add(storageTable);
}
} else {
log.info("No finer granualar partitions exist for {}", part);
return false;
}
} else {
log.debug("Finer granualar partitions added for {}", part);
}
}
}
return getPartitions(fromDate, ceilFromDate, partCol, partitions, remainingIntervals,
addNonExistingParts, failOnPartialData, missingPartitions)
&& getPartitions(floorToDate, toDate, partCol, partitions, remainingIntervals,
addNonExistingParts, failOnPartialData, missingPartitions);
}
private boolean isCandidatePartiallyValidForTimeRange(Date startDate, Date endDate, Date fromDate, Date toDate) {
return Stream.of(startDate, fromDate).max(naturalOrder()).orElse(startDate)
.before(Stream.of(endDate, toDate).min(naturalOrder()).orElse(endDate));
}
@Override
public boolean evaluateCompleteness(TimeRange timeRange, TimeRange queriedTimeRange, boolean failOnPartialData)
throws LensException {
// Check the measure tags.
if (!evaluateMeasuresCompleteness(timeRange)) {
log.info("Storage candidate:{} has partitions with incomplete data: {} for given ranges: {}", this,
dataCompletenessMap, cubeQueryContext.getTimeRanges());
if (failOnPartialData) {
return false;
}
}
PartitionRangesForPartitionColumns missingParts = new PartitionRangesForPartitionColumns();
PruneCauses<Candidate> storagePruningMsgs = cubeQueryContext.getStoragePruningMsgs();
Set<String> unsupportedTimeDims = Sets.newHashSet();
Set<String> partColsQueried = Sets.newHashSet();
partColsQueried.add(timeRange.getPartitionColumn());
StringBuilder extraWhereClauseFallback = new StringBuilder();
Set<FactPartition> rangeParts = getPartitions(timeRange, validUpdatePeriods, true, failOnPartialData, missingParts);
String partCol = timeRange.getPartitionColumn();
boolean partColNotSupported = rangeParts.isEmpty();
if (storagePruningMsgs.containsKey(this)) {
List<CandidateTablePruneCause> causes = storagePruningMsgs.get(this);
// Find the PART_COL_DOES_NOT_EXISTS
for (CandidateTablePruneCause cause : causes) {
if (cause.getCause().equals(CandidateTablePruneCode.PART_COL_DOES_NOT_EXIST)) {
partColNotSupported &= cause.getNonExistantPartCols().contains(partCol);
}
}
} else {
partColNotSupported = false;
}
TimeRange prevRange = timeRange;
String sep = "";
while (rangeParts.isEmpty()) {
String timeDim = cubeQueryContext.getBaseCube().getTimeDimOfPartitionColumn(partCol);
if (getFact() instanceof CubeFactTable) {
if (partColNotSupported && !((CubeFactTable) getFact()).hasColumn(timeDim)) {
unsupportedTimeDims.add(
cubeQueryContext.getBaseCube().getTimeDimOfPartitionColumn(timeRange.getPartitionColumn())
);
break;
}
}
TimeRange fallBackRange = getFallbackRange(prevRange, this.getFact().getSourceFactName(), cubeQueryContext);
log.info("No partitions for range:{}. fallback range: {}", timeRange, fallBackRange);
if (fallBackRange == null) {
break;
}
partColsQueried.add(fallBackRange.getPartitionColumn());
rangeParts = getPartitions(fallBackRange, validUpdatePeriods, true, failOnPartialData, missingParts);
extraWhereClauseFallback.append(sep).append(
prevRange.toTimeDimWhereClause(cubeQueryContext.getAliasForTableName(cubeQueryContext.getCube()), timeDim)
);
sep = " AND ";
prevRange = fallBackRange;
partCol = prevRange.getPartitionColumn();
if (!rangeParts.isEmpty()) {
break;
}
}
// Add all the partitions. participatingPartitions contains all the partitions for previous time ranges also.
rangeToPartitions.put(queriedTimeRange, rangeParts);
numQueriedParts += rangeParts.size();
if (!unsupportedTimeDims.isEmpty()) {
log.info("Not considering storage candidate:{} as it doesn't support time dimensions: {}", this,
unsupportedTimeDims);
cubeQueryContext.addStoragePruningMsg(this, timeDimNotSupported(unsupportedTimeDims));
return false;
}
Set<String> nonExistingParts = missingParts.toSet(partColsQueried);
// TODO union : Relook at this.
nonExistingPartitions.addAll(nonExistingParts);
if (rangeParts.size() == 0 || (failOnPartialData && !nonExistingParts.isEmpty())) {
log.info("Not considering storage candidate:{} as no partitions for fallback range:{}", this, timeRange);
return false;
}
String extraWhere = extraWhereClauseFallback.toString();
if (!StringUtils.isEmpty(extraWhere)) {
rangeToExtraWhereFallBack.put(queriedTimeRange, extraWhere);
}
return true;
}
@Override
public Set<FactPartition> getParticipatingPartitions() {
Set<FactPartition> allPartitions = new HashSet<>(numQueriedParts);
for (Set<FactPartition> rangePartitions : rangeToPartitions.values()) {
allPartitions.addAll(rangePartitions);
}
return allPartitions;
}
private boolean evaluateMeasuresCompleteness(TimeRange timeRange) throws LensException {
if (getCubeMetastoreClient() == null || !getCubeMetastoreClient().isDataCompletenessCheckEnabled()) {
log.info("Skipping availability check for the fact table: {} as dataCompleteness check is not enabled", fact);
return true;
}
String factDataCompletenessTag = fact.getDataCompletenessTag();
if (factDataCompletenessTag == null) {
log.info("Not checking completeness for the fact table:{} as the dataCompletenessTag is not set", fact);
return true;
}
Set<String> measureTag = new HashSet<>();
Map<String, String> tagToMeasureOrExprMap = new HashMap<>();
processExpressionsForCompleteness(cubeQueryContext, measureTag, tagToMeasureOrExprMap);
Set<String> measures = cubeQueryContext.getQueriedMsrs();
if (measures == null) {
measures = new HashSet<>();
}
for (String measure : measures) {
processCubeColForDataCompleteness(cubeQueryContext, measure, measure, measureTag, tagToMeasureOrExprMap);
}
//Checking if dataCompletenessTag is set for the fact
if (measureTag.isEmpty()) {
log.info("No Queried measures with the dataCompletenessTag, hence skipping the availability check");
return true;
}
// default completenessTag will be true
boolean isDataComplete = true;
DataCompletenessChecker completenessChecker = getCubeMetastoreClient().getCompletenessChecker();
DateFormat formatter = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");
formatter.setTimeZone(TimeZone.getTimeZone("UTC"));
if (!timeRange.getPartitionColumn().equals(completenessPartCol)) {
log.info("Completeness check not available for partCol:{}", timeRange.getPartitionColumn());
return true;
}
Date from = timeRange.getFromDate();
Date to = timeRange.getToDate();
Map<String, Map<Date, Float>> completenessMap = completenessChecker
.getCompleteness(factDataCompletenessTag, from, to, measureTag);
if (completenessMap != null && !completenessMap.isEmpty()) {
for (Map.Entry<String, Map<Date, Float>> measureCompleteness : completenessMap.entrySet()) {
String tag = measureCompleteness.getKey();
for (Map.Entry<Date, Float> completenessResult : measureCompleteness.getValue().entrySet()) {
if (completenessResult.getValue() < completenessThreshold) {
log.info("Completeness for the measure_tag {} is {}, threshold: {}, for the hour {}", tag,
completenessResult.getValue(), completenessThreshold, formatter.format(completenessResult.getKey()));
String measureorExprFromTag = tagToMeasureOrExprMap.get(tag);
dataCompletenessMap.computeIfAbsent(measureorExprFromTag, k -> new HashMap<>())
.put(formatter.format(completenessResult.getKey()), completenessResult.getValue());
// set completeness to false if availability for measure is below threshold
isDataComplete = false;
}
}
}
}
return isDataComplete;
}
private Set<FactPartition> getPartitions(TimeRange timeRange, TreeSet<UpdatePeriod> updatePeriods,
boolean addNonExistingParts, boolean failOnPartialData, PartitionRangesForPartitionColumns missingParts)
throws LensException {
Set<FactPartition> partitions = new TreeSet<>();
if (timeRange != null && timeRange.isCoverableBy(updatePeriods)) {
getPartitions(timeRange.getFromDate(), timeRange.getToDate(), timeRange.getPartitionColumn(),
partitions, updatePeriods, addNonExistingParts, failOnPartialData, missingParts);
}
return partitions;
}
@Override
public boolean isExpressionEvaluable(ExpressionResolver.ExpressionContext expr) {
return expr.isEvaluable(this);
}
@Override
public boolean isExpressionEvaluable(String expr) {
return isExpressionEvaluable(
getCubeQueryContext().getExprCtx().getExpressionContext(
expr, getCubeQueryContext().getAliasForTableName(getBaseTable().getName()))
);
}
@Override
public boolean isDimAttributeEvaluable(String dim) throws LensException {
return getCubeQueryContext().getDeNormCtx()
.addRefUsage(getCubeQueryContext(), this, dim, getCubeQueryContext().getCube().getName());
}
@Override
public boolean equals(Object obj) {
if (super.equals(obj)) {
return true;
}
if (obj == null || !(obj instanceof StorageCandidate)) {
return false;
}
StorageCandidate storageCandidateObj = (StorageCandidate) obj;
//Assuming that same instance of cube and fact will be used across StorageCandidate s and hence relying directly
//on == check for these.
return (this.cube == storageCandidateObj.cube && this.fact == storageCandidateObj.fact && this.storageTable
.equals(storageCandidateObj.storageTable));
}
@Override
public int hashCode() {
return this.storageTable.hashCode();
}
@Override
public String toString() {
return getResolvedName();
}
void addValidUpdatePeriod(UpdatePeriod updatePeriod) {
this.validUpdatePeriods.add(updatePeriod);
}
public String getAliasForTable(String alias) {
String database = SessionState.get().getCurrentDatabase();
String ret;
if (alias == null || alias.isEmpty()) {
ret = getResolvedName();
} else {
ret = getResolvedName() + " " + alias;
}
if (StringUtils.isNotBlank(database) && !"default".equalsIgnoreCase(database)) {
ret = database + "." + ret;
}
return ret;
}
boolean isUpdatePeriodUseful(UpdatePeriod updatePeriod) {
return getCubeQueryContext().getTimeRanges().stream()
.anyMatch(timeRange -> isUpdatePeriodUseful(timeRange, updatePeriod));
}
/**
* Is the update period useful for this time range. e.g. for a time range of hours and days, monthly
* and yearly update periods are useless. DAILY and HOURLY are useful. It further checks if the update
* period answers the range at least partially based on start and end times configured at update period
* level or at storage or fact level.
* @param timeRange The time range
* @param updatePeriod Update period
* @return Whether it's useless
*/
private boolean isUpdatePeriodUseful(TimeRange timeRange, UpdatePeriod updatePeriod) {
try {
if (!timeRange.truncate(getStorageTableStartDate(updatePeriod),
getStorageTableEndDate(updatePeriod)).isValid()) {
return false;
}
Date storageTblStartDate = getStorageTableStartDate(updatePeriod);
Date storageTblEndDate = getStorageTableEndDate(updatePeriod);
TimeRange.builder() //TODO date calculation to move to util method and resued
.fromDate(timeRange.getFromDate().after(storageTblStartDate) ? timeRange.getFromDate() : storageTblStartDate)
.toDate(timeRange.getToDate().before(storageTblEndDate) ? timeRange.getToDate() : storageTblEndDate)
.partitionColumn(timeRange.getPartitionColumn())
.build()
.truncate(updatePeriod);
return true;
} catch (LensException e) {
return false;
}
}
/**
* Is time range coverable based on valid update periods of this storage candidate
*
* @param timeRange
* @return
* @throws LensException
*/
public boolean isTimeRangeCoverable(TimeRange timeRange) throws LensException {
return isTimeRangeCoverable(timeRange.getFromDate(), timeRange.getToDate(), validUpdatePeriods);
}
/*
* Is the time range coverable by given update periods.
* Extracts the max update period, then extracts maximum amount of range from the middle that this update
* period can cover. Then recurses on the remaining ranges on the left and right side of the extracted chunk
* using one less update period.
*
* @param timeRangeStart
* @param timeRangeEnd
* @param intervals Update periods to check
* @return Whether time range is coverable by provided update periods or not.
*/
private boolean isTimeRangeCoverable(Date timeRangeStart, Date timeRangeEnd,
Set<UpdatePeriod> intervals) throws LensException {
if (timeRangeStart.equals(timeRangeEnd) || timeRangeStart.after(timeRangeEnd)) {
return true;
}
if (intervals == null || intervals.isEmpty()) {
return false;
}
UpdatePeriod maxInterval = CubeFactTable.maxIntervalInRange(timeRangeStart, timeRangeEnd, intervals);
if (maxInterval == null) {
return false;
}
if (maxInterval == UpdatePeriod.CONTINUOUS
&& getCubeQueryContext().getRangeWriter().getClass().equals(BetweenTimeRangeWriter.class)) {
return true;
}
Date maxIntervalStorageTableStartDate = getStorageTableStartDate(maxInterval);
Date maxIntervalStorageTableEndDate = getStorageTableEndDate(maxInterval);
Set<UpdatePeriod> remainingIntervals = Sets.difference(intervals, Sets.newHashSet(maxInterval));
if (!isCandidatePartiallyValidForTimeRange(
maxIntervalStorageTableStartDate, maxIntervalStorageTableEndDate, timeRangeStart, timeRangeEnd)) {
//Check the time range in remainingIntervals as maxInterval is not useful
return isTimeRangeCoverable(timeRangeStart, timeRangeEnd, remainingIntervals);
}
Date ceilFromDate = DateUtil.getCeilDate(timeRangeStart.after(maxIntervalStorageTableStartDate)
? timeRangeStart : maxIntervalStorageTableStartDate, maxInterval);
Date floorToDate = DateUtil.getFloorDate(timeRangeEnd.before(maxIntervalStorageTableEndDate)
? timeRangeEnd : maxIntervalStorageTableEndDate, maxInterval);
if (ceilFromDate.equals(floorToDate) || floorToDate.before(ceilFromDate)) {
return isTimeRangeCoverable(timeRangeStart, timeRangeEnd, remainingIntervals);
}
//ceilFromDate to floorToDate time range is covered by maxInterval (though there may be holes.. but that's ok)
//Check the remaining part of time range in remainingIntervals
return isTimeRangeCoverable(timeRangeStart, ceilFromDate, remainingIntervals)
&& isTimeRangeCoverable(floorToDate, timeRangeEnd, remainingIntervals);
}
private Date getStorageTableStartDate(UpdatePeriod interval) throws LensException {
if (!isStorageTblsAtUpdatePeriodLevel) {
//In this case the start time and end time is at Storage Level and will be same for all update periods.
return this.startTime;
}
return getCubeMetastoreClient().getStorageTableStartDate(
getCubeMetastoreClient().getStorageTableName(fact.getSourceFactName(), storageName, interval), fact);
}
private Date getStorageTableEndDate(UpdatePeriod interval) throws LensException {
if (!isStorageTblsAtUpdatePeriodLevel) {
//In this case the start time and end time is at Storage Level and will be same for all update periods.
return this.endTime;
}
return getCubeMetastoreClient().getStorageTableEndDate(
getCubeMetastoreClient().getStorageTableName(fact.getSourceFactName(), storageName, interval), fact);
}
public String getResolvedName() {
if (resolvedName == null) {
return storageTable;
}
return resolvedName;
}
/**
* Splits the Storage Candidates into multiple Storage Candidates if storage candidate has multiple
* storage tables (one per update period)
*
* @return
* @throws LensException
*/
public Collection<StorageCandidate> splitAtUpdatePeriodLevelIfReq() throws LensException {
if (!isStorageTblsAtUpdatePeriodLevel) {
return Lists.newArrayList(this); // No need to explode in this case
}
return getPeriodSpecificStorageCandidates();
}
private Collection<StorageCandidate> getPeriodSpecificStorageCandidates() throws LensException {
if (periodSpecificStorageCandidates == null) {
List<StorageCandidate> periodSpecificScList = new ArrayList<>(participatingUpdatePeriods.size());
StorageCandidate updatePeriodSpecificSc;
for (UpdatePeriod period : participatingUpdatePeriods) {
updatePeriodSpecificSc = copy();
updatePeriodSpecificSc.setResolvedName(getCubeMetastoreClient().getStorageTableName(fact.getSourceFactName(),
storageName, period));
updatePeriodSpecificSc.isStorageTblsAtUpdatePeriodLevel = false;
updatePeriodSpecificSc.truncatePartitions(period);
periodSpecificScList.add(updatePeriodSpecificSc);
}
periodSpecificStorageCandidates = periodSpecificScList;
}
return periodSpecificStorageCandidates;
}
/**
* Truncates partitions in {@link #rangeToPartitions} such that only partitions belonging to
* the passed undatePeriod are retained.
* @param updatePeriod
*/
private void truncatePartitions(UpdatePeriod updatePeriod) {
Iterator<Map.Entry<TimeRange, Set<FactPartition>>> rangeItr = rangeToPartitions.entrySet().iterator();
while (rangeItr.hasNext()) {
Map.Entry<TimeRange, Set<FactPartition>> rangeEntry = rangeItr.next();
rangeEntry.getValue().removeIf(factPartition -> !factPartition.getPeriod().equals(updatePeriod));
rangeEntry.getValue().forEach(factPartition -> {
factPartition.getStorageTables().remove(storageTable);
factPartition.getStorageTables().add(resolvedName);
});
if (rangeEntry.getValue().isEmpty()) {
rangeItr.remove();
}
}
}
@Override
public StorageCandidateHQLContext toQueryWriterContext(Map<Dimension, CandidateDim> dimsToQuery,
CubeQueryContext rootCubeQueryContext) throws LensException {
DefaultQueryAST ast = DefaultQueryAST.fromStorageCandidate(null, getCubeQueryContext());
ast.copyFrom(getCubeQueryContext());
return new StorageCandidateHQLContext(this, Maps.newHashMap(dimsToQuery), ast, rootCubeQueryContext);
}
@Override
public Set<Integer> decideMeasurePhrasesToAnswer(Set<Integer> measureIndices) {
answerableMeasurePhraseIndices.retainAll(measureIndices);
return answerableMeasurePhraseIndices;
}
}