예제 #1
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *ShardData) QueryResponseBufferSize(querySpec *parser.QuerySpec, batchPointSize int) int {
	groupByTime := querySpec.GetGroupByInterval()
	if groupByTime == nil {
		// If the group by time is nil, we shouldn't have to use a buffer since the shards should be queried sequentially.
		// However, set this to something high just to be safe.
		log.Debug("BUFFER SIZE: 1000")
		return 1000
	}

	tickCount := int(self.shardNanoseconds / uint64(*groupByTime))
	if tickCount < 10 {
		tickCount = 100
	} else if tickCount > 1000 {
		// cap this because each response should have up to this number of points in it.
		tickCount = tickCount / batchPointSize

		// but make sure it's at least 1k
		if tickCount < 1000 {
			tickCount = 1000
		}
	}
	columnCount := querySpec.GetGroupByColumnCount()
	if columnCount > 1 {
		// we don't really know the cardinality for any column up front. This is a just a multiplier so we'll see how this goes.
		// each response can have many points, so having a buffer of the ticks * 100 should be safe, but we'll see.
		tickCount = tickCount * 100
	}
	log.Debug("BUFFER SIZE: %d", tickCount)
	return tickCount
}
예제 #2
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *Shard) Query(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	if querySpec.IsListSeriesQuery() {
		return self.executeListSeriesQuery(querySpec, processor)
	} else if querySpec.IsDeleteFromSeriesQuery() {
		return self.executeDeleteQuery(querySpec, processor)
	}

	seriesAndColumns := querySpec.SelectQuery().GetReferencedColumns()

	if !self.hasReadAccess(querySpec) {
		return errors.New("User does not have access to one or more of the series requested.")
	}

	for series, columns := range seriesAndColumns {
		if regex, ok := series.GetCompiledRegex(); ok {
			seriesNames := self.metaStore.GetSeriesForDatabaseAndRegex(querySpec.Database(), regex)
			for _, name := range seriesNames {
				if !querySpec.HasReadAccess(name) {
					continue
				}
				err := self.executeQueryForSeries(querySpec, name, columns, processor)
				if err != nil {
					return err
				}
			}
		} else {
			err := self.executeQueryForSeries(querySpec, series.Name, columns, processor)
			if err != nil {
				return err
			}
		}
	}
	return nil
}
예제 #3
0
func (self *LevelDbShard) executeListSeriesQuery(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	it := self.db.NewIterator(self.readOptions)
	defer it.Close()

	database := querySpec.Database()
	seekKey := append(DATABASE_SERIES_INDEX_PREFIX, []byte(querySpec.Database()+"~")...)
	it.Seek(seekKey)
	dbNameStart := len(DATABASE_SERIES_INDEX_PREFIX)
	for it = it; it.Valid(); it.Next() {
		key := it.Key()
		if len(key) < dbNameStart || !bytes.Equal(key[:dbNameStart], DATABASE_SERIES_INDEX_PREFIX) {
			break
		}
		dbSeries := string(key[dbNameStart:])
		parts := strings.Split(dbSeries, "~")
		if len(parts) > 1 {
			if parts[0] != database {
				break
			}
			name := parts[1]
			shouldContinue := processor.YieldPoint(&name, nil, nil)
			if !shouldContinue {
				return nil
			}
		}
	}
	return nil
}
예제 #4
0
func (self *CoordinatorImpl) runDeleteQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	db := querySpec.Database()
	if !querySpec.User().IsDbAdmin(db) {
		return common.NewAuthorizationError("Insufficient permission to write to %s", db)
	}
	querySpec.RunAgainstAllServersInShard = true
	return self.runQuerySpec(querySpec, seriesWriter)
}
예제 #5
0
func (self *CoordinatorImpl) runDeleteQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	user := querySpec.User()
	db := querySpec.Database()
	if ok, err := self.permissions.AuthorizeDeleteQuery(user, db); !ok {
		return err
	}
	querySpec.RunAgainstAllServersInShard = true
	return self.runQuerySpec(querySpec, seriesWriter)
}
예제 #6
0
파일: shard.go 프로젝트: hanshenu/influxdb
func (self *Shard) executeDropSeriesQuery(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	database := querySpec.Database()
	series := querySpec.Query().DropSeriesQuery.GetTableName()
	err := self.dropSeries(database, series)
	if err != nil {
		return err
	}
	self.db.Compact()
	return nil
}
예제 #7
0
파일: shard.go 프로젝트: hanshenu/influxdb
func (self *Shard) hasReadAccess(querySpec *parser.QuerySpec) bool {
	for series := range querySpec.SeriesValuesAndColumns() {
		if _, isRegex := series.GetCompiledRegex(); !isRegex {
			if !querySpec.HasReadAccess(series.Name) {
				return false
			}
		}
	}
	return true
}
예제 #8
0
func (self *CoordinatorImpl) runQuerySpec(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	shards, processor, seriesClosed, err := self.getShardsAndProcessor(querySpec, seriesWriter)
	if err != nil {
		return err
	}

	defer func() {
		if processor != nil {
			processor.Close()
			<-seriesClosed
		} else {
			seriesWriter.Close()
		}
	}()

	shardConcurrentLimit := self.config.ConcurrentShardQueryLimit
	if self.shouldQuerySequentially(shards, querySpec) {
		log.Debug("Querying shards sequentially")
		shardConcurrentLimit = 1
	}
	log.Debug("Shard concurrent limit: ", shardConcurrentLimit)

	errors := make(chan error, shardConcurrentLimit)
	for i := 0; i < shardConcurrentLimit; i++ {
		errors <- nil
	}
	responseChannels := make(chan (<-chan *protocol.Response), shardConcurrentLimit)

	go self.readFromResposneChannels(processor, seriesWriter, querySpec.IsExplainQuery(), errors, responseChannels)

	err = self.queryShards(querySpec, shards, errors, responseChannels)

	// make sure we read the rest of the errors and responses
	for _err := range errors {
		if err == nil {
			err = _err
		}
	}

	for responsechan := range responseChannels {
		for response := range responsechan {
			if response.GetType() != endStreamResponse {
				continue
			}
			if response.ErrorMessage != nil && err == nil {
				err = common.NewQueryError(common.InvalidArgument, *response.ErrorMessage)
			}
			break
		}
	}
	return err
}
예제 #9
0
func (self *CoordinatorImpl) checkPermission(user common.User, querySpec *parser.QuerySpec) error {
	// if this isn't a regex query do the permission check here
	fromClause := querySpec.SelectQuery().GetFromClause()

	for _, n := range fromClause.Names {
		if _, ok := n.Name.GetCompiledRegex(); ok {
			break
		} else if name := n.Name.Name; !user.HasReadAccess(name) {
			return fmt.Errorf("User doesn't have read access to %s", name)
		}
	}
	return nil
}
예제 #10
0
func (self *CoordinatorImpl) runListSeriesQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	series := self.clusterConfiguration.MetaStore.GetSeriesForDatabase(querySpec.Database())
	name := "list_series_result"
	fields := []string{"name"}
	points := make([]*protocol.Point, len(series), len(series))

	for i, s := range series {
		fieldValues := []*protocol.FieldValue{{StringValue: proto.String(s)}}
		points[i] = &protocol.Point{Values: fieldValues}
	}

	seriesResult := &protocol.Series{Name: &name, Fields: fields, Points: points}
	seriesWriter.Write(seriesResult)
	seriesWriter.Close()
	return nil
}
예제 #11
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *Shard) fetchSinglePoint(querySpec *parser.QuerySpec, series string, fields []*metastore.Field) (*protocol.Series, error) {
	query := querySpec.SelectQuery()
	fieldCount := len(fields)
	fieldNames := make([]string, 0, fieldCount)
	point := &protocol.Point{Values: make([]*protocol.FieldValue, 0, fieldCount)}
	timestamp := common.TimeToMicroseconds(query.GetStartTime())
	sequenceNumber, err := query.GetSinglePointQuerySequenceNumber()
	if err != nil {
		return nil, err
	}

	timeAndSequenceBuffer := bytes.NewBuffer(make([]byte, 0, 16))
	binary.Write(timeAndSequenceBuffer, binary.BigEndian, self.convertTimestampToUint(&timestamp))
	binary.Write(timeAndSequenceBuffer, binary.BigEndian, sequenceNumber)
	sequenceNumber_uint64 := uint64(sequenceNumber)
	point.SequenceNumber = &sequenceNumber_uint64
	point.SetTimestampInMicroseconds(timestamp)

	timeAndSequenceBytes := timeAndSequenceBuffer.Bytes()
	for _, field := range fields {
		pointKeyBuff := bytes.NewBuffer(make([]byte, 0, 24))
		pointKeyBuff.Write(field.IdAsBytes())
		pointKeyBuff.Write(timeAndSequenceBytes)

		if data, err := self.db.Get(pointKeyBuff.Bytes()); err != nil {
			return nil, err
		} else {
			fieldValue := &protocol.FieldValue{}
			err := proto.Unmarshal(data, fieldValue)
			if err != nil {
				return nil, err
			}
			if data != nil {
				fieldNames = append(fieldNames, field.Name)
				point.Values = append(point.Values, fieldValue)
			}
		}
	}

	result := &protocol.Series{Name: &series, Fields: fieldNames, Points: []*protocol.Point{point}}

	return result, nil
}
예제 #12
0
func (self *CoordinatorImpl) runDropSeriesQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	user := querySpec.User()
	db := querySpec.Database()
	series := querySpec.Query().DropSeriesQuery.GetTableName()
	if !user.IsClusterAdmin() && !user.IsDbAdmin(db) && !user.HasWriteAccess(series) {
		return common.NewAuthorizationError("Insufficient permissions to drop series")
	}
	querySpec.RunAgainstAllServersInShard = true
	return self.runQuerySpec(querySpec, seriesWriter)
}
예제 #13
0
func (self *CoordinatorImpl) runDropSeriesQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	user := querySpec.User()
	db := querySpec.Database()
	series := querySpec.Query().DropSeriesQuery.GetTableName()
	if ok, err := self.permissions.AuthorizeDropSeries(user, db, series); !ok {
		return err
	}
	querySpec.RunAgainstAllServersInShard = true
	return self.runQuerySpec(querySpec, seriesWriter)
}
예제 #14
0
func (self *CoordinatorImpl) shouldQuerySequentially(shards []*cluster.ShardData, querySpec *parser.QuerySpec) bool {
	// if the query isn't a select, then it doesn't matter
	if querySpec.SelectQuery != nil {
		return false
	}

	// if the query is a regex, we can't predic the number of responses
	// we get back
	if querySpec.IsRegex() {
		return true
	}
	groupByClause := querySpec.SelectQuery().GetGroupByClause()
	// if there's no group by clause, then we're returning raw points
	// with some math done on them, thus we can't predict the number of
	// points
	if groupByClause == nil {
		return true
	}
	// if there's a group by clause but no group by interval, we can't
	// predict the cardinality of the columns used in the group by
	// interval, thus we can't predict the number of responses returned
	// from the shard
	if querySpec.GetGroupByInterval() == nil {
		return true
	}
	// if there's a group by time and other columns, then the previous
	// logic holds
	if len(groupByClause.Elems) > 1 {
		return true
	}

	// parallel querying only if we're querying a single series, with
	// group by time only
	return false
}
예제 #15
0
func (self *ClusterConfiguration) GetShards(querySpec *parser.QuerySpec) []*ShardData {
	self.shardsByIdLock.RLock()
	defer self.shardsByIdLock.RUnlock()

	shouldQueryShortTerm, shouldQueryLongTerm := querySpec.ShouldQueryShortTermAndLongTerm()

	if shouldQueryLongTerm && shouldQueryShortTerm {
		shards := make([]*ShardData, 0)
		shards = append(shards, self.getShardRange(querySpec, self.shortTermShards)...)
		shards = append(shards, self.getShardRange(querySpec, self.longTermShards)...)
		if querySpec.IsAscending() {
			SortShardsByTimeAscending(shards)
		} else {
			SortShardsByTimeDescending(shards)
		}
		return shards
	}

	var shards []*ShardData
	if shouldQueryLongTerm {
		shards = self.getShardRange(querySpec, self.longTermShards)
	} else {
		shards = self.getShardRange(querySpec, self.shortTermShards)
	}
	if querySpec.IsAscending() {
		newShards := append([]*ShardData{}, shards...)
		SortShardsByTimeAscending(newShards)
		return newShards
	}
	return shards
}
예제 #16
0
func (self *ClusterConfiguration) GetShards(querySpec *parser.QuerySpec) []*ShardData {
	self.shardsByIdLock.RLock()
	defer self.shardsByIdLock.RUnlock()

	if querySpec.IsDropSeriesQuery() {
		seriesName := querySpec.Query().DropSeriesQuery.GetTableName()
		if seriesName[0] < FIRST_LOWER_CASE_CHARACTER {
			return self.longTermShards
		}
		return self.shortTermShards
	}

	shouldQueryShortTerm, shouldQueryLongTerm := querySpec.ShouldQueryShortTermAndLongTerm()

	if shouldQueryLongTerm && shouldQueryShortTerm {
		shards := make([]*ShardData, 0)
		shards = append(shards, self.getShardRange(querySpec, self.shortTermShards)...)
		shards = append(shards, self.getShardRange(querySpec, self.longTermShards)...)
		if querySpec.IsAscending() {
			SortShardsByTimeAscending(shards)
		} else {
			SortShardsByTimeDescending(shards)
		}
		return shards
	}

	var shards []*ShardData
	if shouldQueryLongTerm {
		shards = self.getShardRange(querySpec, self.longTermShards)
	} else {
		shards = self.getShardRange(querySpec, self.shortTermShards)
	}
	if querySpec.IsAscending() {
		newShards := append([]*ShardData{}, shards...)
		SortShardsByTimeAscending(newShards)
		return newShards
	}
	return shards
}
예제 #17
0
func (self *CoordinatorImpl) runQuerySpec(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	shards, processor, seriesClosed, err := self.getShardsAndProcessor(querySpec, seriesWriter)
	if err != nil {
		return err
	}

	shardConcurrentLimit := self.config.ConcurrentShardQueryLimit
	if self.shouldQuerySequentially(shards, querySpec) {
		log.Debug("Querying shards sequentially")
		shardConcurrentLimit = 1
	}
	log.Debug("Shard concurrent limit: ", shardConcurrentLimit)
	for i := 0; i < len(shards); i += shardConcurrentLimit {
		responses := make([]<-chan *protocol.Response, 0, shardConcurrentLimit)

		for j := 0; j < shardConcurrentLimit && i+j < len(shards); j++ {
			shard := shards[i+j]
			responseChan := make(chan *protocol.Response, shard.QueryResponseBufferSize(querySpec, self.config.LevelDbPointBatchSize))
			// We query shards for data and stream them to query processor
			log.Debug("QUERYING: shard: ", i+j, shard.String())
			go shard.Query(querySpec, responseChan)
			responses = append(responses, responseChan)
		}

		err := self.readFromResposneChannels(processor, seriesWriter, querySpec.IsExplainQuery(), responses)
		if err != nil {
			log.Error("Reading responses from channels returned an error: %s", err)
			return err
		}
	}

	if processor != nil {
		processor.Close()
		<-seriesClosed
		return err
	}
	seriesWriter.Close()
	return err
}
예제 #18
0
func (self *LevelDbShard) executeDeleteQuery(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	query := querySpec.DeleteQuery()
	series := query.GetFromClause()
	database := querySpec.Database()
	if series.Type != parser.FromClauseArray {
		return fmt.Errorf("Merge and Inner joins can't be used with a delete query", series.Type)
	}

	for _, name := range series.Names {
		var err error
		if regex, ok := name.Name.GetCompiledRegex(); ok {
			err = self.deleteRangeOfRegex(database, regex, query.GetStartTime(), query.GetEndTime())
		} else {
			err = self.deleteRangeOfSeries(database, name.Name.Name, query.GetStartTime(), query.GetEndTime())
		}

		if err != nil {
			return err
		}
	}
	return nil
}
예제 #19
0
func (self *CoordinatorImpl) shouldQuerySequentially(shards []*cluster.ShardData, querySpec *parser.QuerySpec) bool {
	// if the query isn't a select, then it doesn't matter
	if querySpec.SelectQuery() == nil {
		return false
	}

	// if the query is a regex, we can't predic the number of responses
	// we get back
	if querySpec.IsRegex() {
		return true
	}
	groupByClause := querySpec.SelectQuery().GetGroupByClause()
	// if there's no group by clause, then we're returning raw points
	// with some math done on them, thus we can't predict the number of
	// points
	if groupByClause == nil {
		return true
	}
	// if there's a group by clause but no group by interval, we can't
	// predict the cardinality of the columns used in the group by
	// interval, thus we can't predict the number of responses returned
	// from the shard
	if querySpec.GetGroupByInterval() == nil {
		return true
	}
	// if there's a group by time and other columns, then the previous
	// logic holds
	if len(groupByClause.Elems) > 1 {
		return true
	}

	if !self.shouldAggregateLocally(shards, querySpec) {
		return true
	}

	for _, shard := range shards {
		bufferSize := shard.QueryResponseBufferSize(querySpec, self.config.LevelDbPointBatchSize)
		// if the number of repsonses is too big, do a sequential querying
		if bufferSize > self.config.ClusterMaxResponseBufferSize {
			return true
		}
	}

	// parallel querying only if we're querying a single series, with
	// group by time only
	return false
}
예제 #20
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *ShardData) ShouldAggregateLocally(querySpec *parser.QuerySpec) bool {
	f := querySpec.GetFromClause()
	if f != nil && (f.Type == parser.FromClauseInnerJoin || f.Type == parser.FromClauseMerge) {
		return false
	}

	if self.durationIsSplit && querySpec.ReadsFromMultipleSeries() {
		return false
	}
	groupByInterval := querySpec.GetGroupByInterval()
	if groupByInterval == nil {
		if querySpec.HasAggregates() {
			return false
		}
		return true
	}
	return self.shardDuration%*groupByInterval == 0
}
예제 #21
0
func (self *CoordinatorImpl) getShardsAndProcessor(querySpec *parser.QuerySpec, writer SeriesWriter) ([]*cluster.ShardData, cluster.QueryProcessor, chan bool, error) {
	shards := self.clusterConfiguration.GetShards(querySpec)
	shouldAggregateLocally := self.shouldAggregateLocally(shards, querySpec)

	var err error
	var processor cluster.QueryProcessor

	responseChan := make(chan *protocol.Response)
	seriesClosed := make(chan bool)

	selectQuery := querySpec.SelectQuery()
	if selectQuery != nil {
		if !shouldAggregateLocally {
			// if we should aggregate in the coordinator (i.e. aggregation
			// isn't happening locally at the shard level), create an engine
			processor, err = engine.NewQueryEngine(querySpec.SelectQuery(), responseChan)
		} else {
			// if we have a query with limit, then create an engine, or we can
			// make the passthrough limit aware
			processor = engine.NewPassthroughEngineWithLimit(responseChan, 100, selectQuery.Limit)
		}
	} else if !shouldAggregateLocally {
		processor = engine.NewPassthroughEngine(responseChan, 100)
	}

	if err != nil {
		return nil, nil, nil, err
	}

	if processor == nil {
		return shards, nil, nil, nil
	}

	go func() {
		for {
			response := <-responseChan

			if *response.Type == endStreamResponse || *response.Type == accessDeniedResponse {
				writer.Close()
				seriesClosed <- true
				return
			}
			if !(*response.Type == queryResponse && querySpec.IsExplainQuery()) {
				if response.Series != nil && len(response.Series.Points) > 0 {
					writer.Write(response.Series)
				}
			}
		}
	}()

	return shards, processor, seriesClosed, nil
}
예제 #22
0
파일: shard.go 프로젝트: j0ni/influxdb
func (self *ShardData) ShouldAggregateLocally(querySpec *parser.QuerySpec) bool {
	if self.durationIsSplit && querySpec.ReadsFromMultipleSeries() {
		return false
	}
	groupByInterval := querySpec.GetGroupByInterval()
	if groupByInterval == nil {
		if querySpec.HasAggregates() {
			return false
		}
		return true
	}
	return self.shardDuration%*groupByInterval == 0
}
예제 #23
0
func (self *CoordinatorImpl) runDropSeriesQuery(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	user := querySpec.User()
	db := querySpec.Database()
	series := querySpec.Query().DropSeriesQuery.GetTableName()
	if ok, err := self.permissions.AuthorizeDropSeries(user, db, series); !ok {
		return err
	}
	defer seriesWriter.Close()
	fmt.Println("DROP series")
	err := self.raftServer.DropSeries(db, series)
	if err != nil {
		return err
	}
	fmt.Println("DROP returning nil")
	return nil
}
예제 #24
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *ShardData) createRequest(querySpec *parser.QuerySpec) *p.Request {
	queryString := querySpec.GetQueryString()
	user := querySpec.User()
	userName := user.GetName()
	database := querySpec.Database()
	isDbUser := !user.IsClusterAdmin()

	return &p.Request{
		Type:     &queryRequest,
		ShardId:  &self.id,
		Query:    &queryString,
		UserName: &userName,
		Database: &database,
		IsDbUser: &isDbUser,
	}
}
예제 #25
0
func (self *LevelDbShard) executeDropSeriesQuery(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	database := querySpec.Database()
	series := querySpec.Query().DropSeriesQuery.GetTableName()
	return self.dropSeries(database, series)
}
예제 #26
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *ShardData) Query(querySpec *parser.QuerySpec, response chan *p.Response) {
	log.Debug("QUERY: shard %d, query '%s'", self.Id(), querySpec.GetQueryString())
	defer common.RecoverFunc(querySpec.Database(), querySpec.GetQueryString(), func(err interface{}) {
		response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(fmt.Sprintf("%s", err))}
	})

	// This is only for queries that are deletes or drops. They need to be sent everywhere as opposed to just the local or one of the remote shards.
	// But this boolean should only be set to true on the server that receives the initial query.
	if querySpec.RunAgainstAllServersInShard {
		if querySpec.IsDeleteFromSeriesQuery() {
			self.logAndHandleDeleteQuery(querySpec, response)
		} else if querySpec.IsDropSeriesQuery() {
			self.logAndHandleDropSeriesQuery(querySpec, response)
		}
	}

	if self.IsLocal {
		var processor QueryProcessor
		var err error

		if querySpec.IsListSeriesQuery() {
			processor = engine.NewListSeriesEngine(response)
		} else if querySpec.IsDeleteFromSeriesQuery() || querySpec.IsDropSeriesQuery() || querySpec.IsSinglePointQuery() {
			maxDeleteResults := 10000
			processor = engine.NewPassthroughEngine(response, maxDeleteResults)
		} else {
			query := querySpec.SelectQuery()
			if self.ShouldAggregateLocally(querySpec) {
				log.Debug("creating a query engine")
				processor, err = engine.NewQueryEngine(query, response)
				if err != nil {
					response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
					log.Error("Error while creating engine: %s", err)
					return
				}
				processor.SetShardInfo(int(self.Id()), self.IsLocal)
			} else if query.HasAggregates() {
				maxPointsToBufferBeforeSending := 1000
				log.Debug("creating a passthrough engine")
				processor = engine.NewPassthroughEngine(response, maxPointsToBufferBeforeSending)
			} else {
				maxPointsToBufferBeforeSending := 1000
				log.Debug("creating a passthrough engine with limit")
				processor = engine.NewPassthroughEngineWithLimit(response, maxPointsToBufferBeforeSending, query.Limit)
			}

			if query.GetFromClause().Type != parser.FromClauseInnerJoin {
				// Joins do their own filtering since we need to get all
				// points before filtering. This is due to the fact that some
				// where expressions will be difficult to compute before the
				// points are joined together, think where clause with
				// left.column = 'something' or right.column =
				// 'something_else'. We can't filter the individual series
				// separately. The filtering happens in merge.go:55

				processor = engine.NewFilteringEngine(query, processor)
			}
		}
		shard, err := self.store.GetOrCreateShard(self.id)
		if err != nil {
			response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
			log.Error("Error while getting shards: %s", err)
			return
		}
		defer self.store.ReturnShard(self.id)
		err = shard.Query(querySpec, processor)
		// if we call Close() in case of an error it will mask the error
		if err != nil {
			response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
			return
		}
		processor.Close()
		response <- &p.Response{Type: &endStreamResponse}
		return
	}

	if server := self.randomHealthyServer(); server != nil {
		log.Debug("Querying server %d for shard %d", server.GetId(), self.Id())
		request := self.createRequest(querySpec)
		server.MakeRequest(request, response)
		return
	}

	message := fmt.Sprintf("No servers up to query shard %d", self.id)
	response <- &p.Response{Type: &endStreamResponse, ErrorMessage: &message}
	log.Error(message)
}
예제 #27
0
파일: shard.go 프로젝트: hanshenu/influxdb
func (self *Shard) executeListSeriesQuery(querySpec *parser.QuerySpec, processor cluster.QueryProcessor) error {
	return self.yieldSeriesNamesForDb(querySpec.Database(), func(_name string) bool {
		name := _name
		return processor.YieldPoint(&name, nil, nil)
	})
}
예제 #28
0
파일: shard.go 프로젝트: qq101/influxdb
func (self *ShardData) Query(querySpec *parser.QuerySpec, response chan *p.Response) {
	// This is only for queries that are deletes or drops. They need to be sent everywhere as opposed to just the local or one of the remote shards.
	// But this boolean should only be set to true on the server that receives the initial query.
	if querySpec.RunAgainstAllServersInShard {
		if querySpec.IsDeleteFromSeriesQuery() {
			self.logAndHandleDeleteQuery(querySpec, response)
		} else if querySpec.IsDropSeriesQuery() {
			self.logAndHandleDropSeriesQuery(querySpec, response)
		}
	}

	if self.IsLocal {
		var processor QueryProcessor
		var err error

		if querySpec.IsListSeriesQuery() {
			processor = engine.NewListSeriesEngine(response)
		} else if querySpec.IsDeleteFromSeriesQuery() || querySpec.IsDropSeriesQuery() || querySpec.IsSinglePointQuery() {
			maxDeleteResults := 10000
			processor = engine.NewPassthroughEngine(response, maxDeleteResults)
		} else {
			query := querySpec.SelectQuery()
			if self.ShouldAggregateLocally(querySpec) {
				log.Debug("creating a query engine\n")
				processor, err = engine.NewQueryEngine(query, response)
				if err != nil {
					response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
					log.Error("Error while creating engine: %s", err)
					return
				}
				processor.SetShardInfo(int(self.Id()), self.IsLocal)
			} else if query.HasAggregates() {
				maxPointsToBufferBeforeSending := 1000
				log.Debug("creating a passthrough engine\n")
				processor = engine.NewPassthroughEngine(response, maxPointsToBufferBeforeSending)
			} else {
				maxPointsToBufferBeforeSending := 1000
				log.Debug("creating a passthrough engine with limit\n")
				processor = engine.NewPassthroughEngineWithLimit(response, maxPointsToBufferBeforeSending, query.Limit)
			}
			processor = engine.NewFilteringEngine(query, processor)
		}
		shard, err := self.store.GetOrCreateShard(self.id)
		if err != nil {
			response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
			log.Error("Error while getting shards: %s", err)
			return
		}
		defer self.store.ReturnShard(self.id)
		err = shard.Query(querySpec, processor)
		processor.Close()
		if err != nil {
			response <- &p.Response{Type: &endStreamResponse, ErrorMessage: p.String(err.Error())}
		}
		response <- &p.Response{Type: &endStreamResponse}
		return
	}

	healthyServers := make([]*ClusterServer, 0, len(self.clusterServers))
	for _, s := range self.clusterServers {
		if !s.IsUp() {
			continue
		}
		healthyServers = append(healthyServers, s)
	}
	healthyCount := len(healthyServers)
	if healthyCount == 0 {
		message := fmt.Sprintf("No servers up to query shard %d", self.id)
		response <- &p.Response{Type: &endStreamResponse, ErrorMessage: &message}
		log.Error(message)
		return
	}
	randServerIndex := int(time.Now().UnixNano() % int64(healthyCount))
	server := healthyServers[randServerIndex]
	request := self.createRequest(querySpec)

	server.MakeRequest(request, response)
}
예제 #29
0
파일: shard.go 프로젝트: jhermann/influxdb
func (self *ShardData) logAndHandleDeleteQuery(querySpec *parser.QuerySpec, response chan *p.Response) {
	queryString := querySpec.GetQueryStringWithTimeCondition()
	request := self.createRequest(querySpec)
	request.Query = &queryString
	self.LogAndHandleDestructiveQuery(querySpec, request, response, false)
}
예제 #30
0
파일: shard.go 프로젝트: hanshenu/influxdb
func (self *Shard) executeQueryForSeries(querySpec *parser.QuerySpec, seriesName string, columns []string, processor cluster.QueryProcessor) error {
	startTimeBytes := self.byteArrayForTime(querySpec.GetStartTime())
	endTimeBytes := self.byteArrayForTime(querySpec.GetEndTime())

	fields, err := self.getFieldsForSeries(querySpec.Database(), seriesName, columns)
	if err != nil {
		// because a db is distributed across the cluster, it's possible we don't have the series indexed here. ignore
		switch err := err.(type) {
		case FieldLookupError:
			log.Debug("Cannot find fields %v", columns)
			return nil
		default:
			log.Error("Error looking up fields for %s: %s", seriesName, err)
			return fmt.Errorf("Error looking up fields for %s: %s", seriesName, err)
		}
	}

	fieldCount := len(fields)
	rawColumnValues := make([]rawColumnValue, fieldCount, fieldCount)
	query := querySpec.SelectQuery()

	aliases := query.GetTableAliases(seriesName)
	if querySpec.IsSinglePointQuery() {
		series, err := self.fetchSinglePoint(querySpec, seriesName, fields)
		if err != nil {
			log.Error("Error reading a single point: %s", err)
			return err
		}
		if len(series.Points) > 0 {
			processor.YieldPoint(series.Name, series.Fields, series.Points[0])
		}
		return nil
	}

	fieldNames, iterators := self.getIterators(fields, startTimeBytes, endTimeBytes, query.Ascending)
	defer func() {
		for _, it := range iterators {
			it.Close()
		}
	}()

	seriesOutgoing := &protocol.Series{Name: protocol.String(seriesName), Fields: fieldNames, Points: make([]*protocol.Point, 0, self.pointBatchSize)}

	// TODO: clean up, this is super gnarly
	// optimize for the case where we're pulling back only a single column or aggregate
	buffer := bytes.NewBuffer(nil)
	valueBuffer := proto.NewBuffer(nil)
	for {
		isValid := false
		point := &protocol.Point{Values: make([]*protocol.FieldValue, fieldCount, fieldCount)}

		for i, it := range iterators {
			if rawColumnValues[i].value != nil || !it.Valid() {
				if err := it.Error(); err != nil {
					return err
				}
				continue
			}

			key := it.Key()
			if len(key) < 16 {
				continue
			}

			if !isPointInRange(fields[i].Id, startTimeBytes, endTimeBytes, key) {
				continue
			}

			value := it.Value()
			sequenceNumber := key[16:]

			rawTime := key[8:16]
			rawColumnValues[i] = rawColumnValue{time: rawTime, sequence: sequenceNumber, value: value}
		}

		var pointTimeRaw []byte
		var pointSequenceRaw []byte
		// choose the highest (or lowest in case of ascending queries) timestamp
		// and sequence number. that will become the timestamp and sequence of
		// the next point.
		for _, value := range rawColumnValues {
			if value.value == nil {
				continue
			}

			pointTimeRaw, pointSequenceRaw = value.updatePointTimeAndSequence(pointTimeRaw,
				pointSequenceRaw, query.Ascending)
		}

		for i, iterator := range iterators {
			// if the value is nil or doesn't match the point's timestamp and sequence number
			// then skip it
			if rawColumnValues[i].value == nil ||
				!bytes.Equal(rawColumnValues[i].time, pointTimeRaw) ||
				!bytes.Equal(rawColumnValues[i].sequence, pointSequenceRaw) {

				point.Values[i] = &protocol.FieldValue{IsNull: &TRUE}
				continue
			}

			// if we emitted at lease one column, then we should keep
			// trying to get more points
			isValid = true

			// advance the iterator to read a new value in the next iteration
			if query.Ascending {
				iterator.Next()
			} else {
				iterator.Prev()
			}

			fv := &protocol.FieldValue{}
			valueBuffer.SetBuf(rawColumnValues[i].value)
			err := valueBuffer.Unmarshal(fv)
			if err != nil {
				log.Error("Error while running query: %s", err)
				return err
			}
			point.Values[i] = fv
			rawColumnValues[i].value = nil
		}

		var sequence uint64
		var t uint64

		// set the point sequence number and timestamp
		buffer.Reset()
		buffer.Write(pointSequenceRaw)
		binary.Read(buffer, binary.BigEndian, &sequence)
		buffer.Reset()
		buffer.Write(pointTimeRaw)
		binary.Read(buffer, binary.BigEndian, &t)

		time := self.convertUintTimestampToInt64(&t)
		point.SetTimestampInMicroseconds(time)
		point.SequenceNumber = &sequence

		// stop the loop if we ran out of points
		if !isValid {
			break
		}

		shouldContinue := true

		seriesOutgoing.Points = append(seriesOutgoing.Points, point)

		if len(seriesOutgoing.Points) >= self.pointBatchSize {
			for _, alias := range aliases {
				series := &protocol.Series{
					Name:   proto.String(alias),
					Fields: fieldNames,
					Points: seriesOutgoing.Points,
				}
				if !processor.YieldSeries(series) {
					log.Info("Stopping processing")
					shouldContinue = false
				}
			}
			seriesOutgoing = &protocol.Series{Name: protocol.String(seriesName), Fields: fieldNames, Points: make([]*protocol.Point, 0, self.pointBatchSize)}
		}

		if !shouldContinue {
			break
		}
	}

	//Yield remaining data
	for _, alias := range aliases {
		log.Debug("Final Flush %s", alias)
		series := &protocol.Series{Name: protocol.String(alias), Fields: seriesOutgoing.Fields, Points: seriesOutgoing.Points}
		if !processor.YieldSeries(series) {
			log.Debug("Cancelled...")
		}
	}

	log.Debug("Finished running query %s", query.GetQueryString())
	return nil
}