Esempio n. 1
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
}