Example #1
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
}
Example #2
0
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)
	})
}
Example #3
0
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
}
Example #4
0
func (self *CoordinatorImpl) runQuerySpec(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
	shards := self.clusterConfiguration.GetShards(querySpec)

	shouldAggregateLocally := true
	var processor cluster.QueryProcessor
	var responseChan chan *protocol.Response
	var seriesClosed chan bool
	for _, s := range shards {
		// If the aggregation is done at the shard level, we don't need to
		// do it here at the coordinator level.
		if !s.ShouldAggregateLocally(querySpec) {
			seriesClosed = make(chan bool)
			shouldAggregateLocally = false
			responseChan = make(chan *protocol.Response)

			if querySpec.SelectQuery() != nil {
				processor = engine.NewQueryEngine(querySpec.SelectQuery(), responseChan)
			} else {
				bufferSize := 100
				processor = engine.NewPassthroughEngine(responseChan, bufferSize)
			}
			go func() {
				for {
					res := <-responseChan
					if *res.Type == endStreamResponse || *res.Type == accessDeniedResponse {
						seriesWriter.Close()
						seriesClosed <- true
						return
					}
					if res.Series != nil && len(res.Series.Points) > 0 {
						seriesWriter.Write(res.Series)
					}
				}
			}()
			break
		}
	}

	responses := make([]chan *protocol.Response, 0)
	for _, shard := range shards {
		responseChan := make(chan *protocol.Response, self.config.QueryShardBufferSize)
		go shard.Query(querySpec, responseChan)
		responses = append(responses, responseChan)
	}

	for i, responseChan := range responses {
		log.Debug("READING: shard: ", shards[i].String())
		for {
			response := <-responseChan
			log.Debug("GOT RESPONSE: ", response.Type, response.Series)
			if *response.Type == endStreamResponse || *response.Type == accessDeniedResponse {
				break
			}
			if shouldAggregateLocally {
				log.Debug("WRITING: ", len(response.Series.Points))
				seriesWriter.Write(response.Series)
				log.Debug("WRITING (done)")
				continue
			}

			// if the data wasn't aggregated at the shard level, aggregate
			// the data here
			log.Debug("YIELDING: ", len(response.Series.Points))
			if response.Series != nil {
				for _, p := range response.Series.Points {
					processor.YieldPoint(response.Series.Name, response.Series.Fields, p)
				}
			}
		}
		log.Debug("DONE: shard: ", shards[i].String())
	}
	if !shouldAggregateLocally {
		processor.Close()
		<-seriesClosed
		return nil
	}
	seriesWriter.Close()
	return nil
}