Ejemplo n.º 1
0
func (e *StatementExecutor) createIterators(stmt *influxql.SelectStatement, ctx *influxql.ExecutionContext) ([]influxql.Iterator, *influxql.SelectStatement, error) {
	// It is important to "stamp" this time so that everywhere we evaluate `now()` in the statement is EXACTLY the same `now`
	now := time.Now().UTC()
	opt := influxql.SelectOptions{
		InterruptCh: ctx.InterruptCh,
		NodeID:      ctx.ExecutionOptions.NodeID,
		MaxSeriesN:  e.MaxSelectSeriesN,
	}

	// Replace instances of "now()" with the current time, and check the resultant times.
	nowValuer := influxql.NowValuer{Now: now}
	stmt.Condition = influxql.Reduce(stmt.Condition, &nowValuer)
	// Replace instances of "now()" with the current time in the dimensions.
	for _, d := range stmt.Dimensions {
		d.Expr = influxql.Reduce(d.Expr, &nowValuer)
	}

	var err error
	opt.MinTime, opt.MaxTime, err = influxql.TimeRange(stmt.Condition)
	if err != nil {
		return nil, stmt, err
	}

	if opt.MaxTime.IsZero() {
		// In the case that we're executing a meta query where the user cannot
		// specify a time condition, then we expand the default max time
		// to the maximum possible value, to ensure that data where all points
		// are in the future are returned.
		if influxql.Sources(stmt.Sources).HasSystemSource() {
			opt.MaxTime = time.Unix(0, influxql.MaxTime).UTC()
		} else {
			if interval, err := stmt.GroupByInterval(); err != nil {
				return nil, stmt, err
			} else if interval > 0 {
				opt.MaxTime = now
			} else {
				opt.MaxTime = time.Unix(0, influxql.MaxTime).UTC()
			}
		}
	}
	if opt.MinTime.IsZero() {
		opt.MinTime = time.Unix(0, influxql.MinTime).UTC()
	}

	// Convert DISTINCT into a call.
	stmt.RewriteDistinct()

	// Remove "time" from fields list.
	stmt.RewriteTimeFields()

	// Rewrite any regex conditions that could make use of the index.
	stmt.RewriteRegexConditions()

	// Create an iterator creator based on the shards in the cluster.
	ic, err := e.iteratorCreator(stmt, &opt)
	if err != nil {
		return nil, stmt, err
	}

	// Expand regex sources to their actual source names.
	if stmt.Sources.HasRegex() {
		sources, err := ic.ExpandSources(stmt.Sources)
		if err != nil {
			return nil, stmt, err
		}
		stmt.Sources = sources
	}

	// Rewrite wildcards, if any exist.
	tmp, err := stmt.RewriteFields(ic)
	if err != nil {
		return nil, stmt, err
	}
	stmt = tmp

	if e.MaxSelectBucketsN > 0 && !stmt.IsRawQuery {
		interval, err := stmt.GroupByInterval()
		if err != nil {
			return nil, stmt, err
		}

		if interval > 0 {
			// Determine the start and end time matched to the interval (may not match the actual times).
			min := opt.MinTime.Truncate(interval)
			max := opt.MaxTime.Truncate(interval).Add(interval)

			// Determine the number of buckets by finding the time span and dividing by the interval.
			buckets := int64(max.Sub(min)) / int64(interval)
			if int(buckets) > e.MaxSelectBucketsN {
				return nil, stmt, fmt.Errorf("max-select-buckets limit exceeded: (%d/%d)", buckets, e.MaxSelectBucketsN)
			}
		}
	}

	// Create a set of iterators from a selection.
	itrs, err := influxql.Select(stmt, ic, &opt)
	if err != nil {
		return nil, stmt, err
	}

	if e.MaxSelectPointN > 0 {
		monitor := influxql.PointLimitMonitor(itrs, influxql.DefaultStatsInterval, e.MaxSelectPointN)
		ctx.Query.Monitor(monitor)
	}
	return itrs, stmt, nil
}
Ejemplo n.º 2
0
func (e *StatementExecutor) executeSelectStatement(stmt *influxql.SelectStatement, ctx *influxql.ExecutionContext) error {
	// Handle SHOW TAG VALUES separately so it can be optimized.
	// https://github.com/influxdata/influxdb/issues/6233
	if source, ok := stmt.Sources[0].(*influxql.Measurement); ok && source.Name == "_tags" {
		// Use the optimized version only if we have direct access to the database.
		if store, ok := e.TSDBStore.(LocalTSDBStore); ok {
			return e.executeShowTagValues(stmt, ctx, store)
		}
	}

	// It is important to "stamp" this time so that everywhere we evaluate `now()` in the statement is EXACTLY the same `now`
	now := time.Now().UTC()
	opt := influxql.SelectOptions{InterruptCh: ctx.InterruptCh}

	// Replace instances of "now()" with the current time, and check the resultant times.
	nowValuer := influxql.NowValuer{Now: now}
	stmt.Condition = influxql.Reduce(stmt.Condition, &nowValuer)
	// Replace instances of "now()" with the current time in the dimensions.
	for _, d := range stmt.Dimensions {
		d.Expr = influxql.Reduce(d.Expr, &nowValuer)
	}

	var err error
	opt.MinTime, opt.MaxTime, err = influxql.TimeRange(stmt.Condition)
	if err != nil {
		return err
	}

	if opt.MaxTime.IsZero() {
		// In the case that we're executing a meta query where the user cannot
		// specify a time condition, then we expand the default max time
		// to the maximum possible value, to ensure that data where all points
		// are in the future are returned.
		if influxql.Sources(stmt.Sources).HasSystemSource() {
			opt.MaxTime = time.Unix(0, influxql.MaxTime).UTC()
		} else {
			opt.MaxTime = now
		}
	}
	if opt.MinTime.IsZero() {
		opt.MinTime = time.Unix(0, 0)
	}

	// Convert DISTINCT into a call.
	stmt.RewriteDistinct()

	// Remove "time" from fields list.
	stmt.RewriteTimeFields()

	// Create an iterator creator based on the shards in the cluster.
	ic, err := e.iteratorCreator(stmt, &opt)
	if err != nil {
		return err
	}

	// Expand regex sources to their actual source names.
	if stmt.Sources.HasRegex() {
		sources, err := ic.ExpandSources(stmt.Sources)
		if err != nil {
			return err
		}
		stmt.Sources = sources
	}

	// Rewrite wildcards, if any exist.
	tmp, err := stmt.RewriteFields(ic)
	if err != nil {
		return err
	}
	stmt = tmp

	if e.MaxSelectBucketsN > 0 && !stmt.IsRawQuery {
		interval, err := stmt.GroupByInterval()
		if err != nil {
			return err
		}

		if interval > 0 {
			// Determine the start and end time matched to the interval (may not match the actual times).
			min := opt.MinTime.Truncate(interval)
			max := opt.MaxTime.Truncate(interval).Add(interval)

			// Determine the number of buckets by finding the time span and dividing by the interval.
			buckets := int64(max.Sub(min)) / int64(interval)
			if int(buckets) > e.MaxSelectBucketsN {
				return fmt.Errorf("max select bucket count exceeded: %d buckets", buckets)
			}
		}
	}

	// Create a set of iterators from a selection.
	itrs, err := influxql.Select(stmt, ic, &opt)
	if err != nil {
		return err
	}

	if e.MaxSelectPointN > 0 {
		monitor := influxql.PointLimitMonitor(itrs, influxql.DefaultStatsInterval, e.MaxSelectPointN)
		ctx.Query.Monitor(monitor)
	}

	// Generate a row emitter from the iterator set.
	em := influxql.NewEmitter(itrs, stmt.TimeAscending(), ctx.ChunkSize)
	em.Columns = stmt.ColumnNames()
	em.OmitTime = stmt.OmitTime
	defer em.Close()

	// Calculate initial stats across all iterators.
	stats := influxql.Iterators(itrs).Stats()
	if e.MaxSelectSeriesN > 0 && stats.SeriesN > e.MaxSelectSeriesN {
		return fmt.Errorf("max select series count exceeded: %d series", stats.SeriesN)
	}

	// Emit rows to the results channel.
	var writeN int64
	var emitted bool

	var pointsWriter *BufferedPointsWriter
	if stmt.Target != nil {
		pointsWriter = NewBufferedPointsWriter(e.PointsWriter, stmt.Target.Measurement.Database, stmt.Target.Measurement.RetentionPolicy, 10000)
	}

	for {
		row, err := em.Emit()
		if err != nil {
			return err
		} else if row == nil {
			// Check if the query was interrupted while emitting.
			select {
			case <-ctx.InterruptCh:
				return influxql.ErrQueryInterrupted
			default:
			}
			break
		}

		// Write points back into system for INTO statements.
		if stmt.Target != nil {
			if err := e.writeInto(pointsWriter, stmt, row); err != nil {
				return err
			}
			writeN += int64(len(row.Values))
			continue
		}

		result := &influxql.Result{
			StatementID: ctx.StatementID,
			Series:      []*models.Row{row},
		}

		// Send results or exit if closing.
		select {
		case <-ctx.InterruptCh:
			return influxql.ErrQueryInterrupted
		case ctx.Results <- result:
		}

		emitted = true
	}

	// Flush remaing points and emit write count if an INTO statement.
	if stmt.Target != nil {
		if err := pointsWriter.Flush(); err != nil {
			return err
		}

		var messages []*influxql.Message
		if ctx.ReadOnly {
			messages = append(messages, influxql.ReadOnlyWarning(stmt.String()))
		}

		ctx.Results <- &influxql.Result{
			StatementID: ctx.StatementID,
			Messages:    messages,
			Series: []*models.Row{{
				Name:    "result",
				Columns: []string{"time", "written"},
				Values:  [][]interface{}{{time.Unix(0, 0).UTC(), writeN}},
			}},
		}
		return nil
	}

	// Always emit at least one result.
	if !emitted {
		ctx.Results <- &influxql.Result{
			StatementID: ctx.StatementID,
			Series:      make([]*models.Row, 0),
		}
	}

	return nil
}