loki/pkg/logql/evaluator.go

package logql

import (
	"container/heap"
	"context"
	"math"
	"sort"
	"time"

	"github.com/pkg/errors"
	"github.com/prometheus/prometheus/pkg/labels"
	"github.com/prometheus/prometheus/promql"

	"github.com/grafana/loki/pkg/iter"
	"github.com/grafana/loki/pkg/logproto"
)

type QueryRangeType string

var (
	InstantType QueryRangeType = "instant"
	RangeType   QueryRangeType = "range"
)

// Params details the parameters associated with a loki request
type Params interface {
	Query() string
	Start() time.Time
	End() time.Time
	Step() time.Duration
	Limit() uint32
	Direction() logproto.Direction
}

// LiteralParams impls Params
type LiteralParams struct {
	qs         string
	start, end time.Time
	step       time.Duration
	direction  logproto.Direction
	limit      uint32
}

// String impls Params
func (p LiteralParams) Query() string { return p.qs }

// Start impls Params
func (p LiteralParams) Start() time.Time { return p.start }

// End impls Params
func (p LiteralParams) End() time.Time { return p.end }

// Step impls Params
func (p LiteralParams) Step() time.Duration { return p.step }

// Limit impls Params
func (p LiteralParams) Limit() uint32 { return p.limit }

// Direction impls Params
func (p LiteralParams) Direction() logproto.Direction { return p.direction }

// GetRangeType returns whether a query is an instant query or range query
func GetRangeType(q Params) QueryRangeType {
	if q.Start() == q.End() && q.Step() == 0 {
		return InstantType
	}
	return RangeType
}

// Evaluator is an interface for iterating over data at different nodes in the AST
type Evaluator interface {
	// StepEvaluator returns a StepEvaluator for a given SampleExpr. It's explicitly passed another StepEvaluator// in order to enable arbitrary compuation of embedded expressions. This allows more modular & extensible
	// StepEvaluator implementations which can be composed.
	StepEvaluator(ctx context.Context, nextEvaluator Evaluator, expr SampleExpr, p Params) (StepEvaluator, error)
	// Iterator returns the iter.EntryIterator for a given LogSelectorExpr
	Iterator(context.Context, LogSelectorExpr, Params) (iter.EntryIterator, error)
}

// EvaluatorUnsupportedType is a helper for signaling that an evaluator does not support an Expr type
func EvaluatorUnsupportedType(expr Expr, ev Evaluator) error {
	return errors.Errorf("unexpected expr type (%T) for Evaluator type (%T) ", expr, ev)
}

type defaultEvaluator struct {
	maxLookBackPeriod time.Duration
	querier           Querier
}

func (ev *defaultEvaluator) Iterator(ctx context.Context, expr LogSelectorExpr, q Params) (iter.EntryIterator, error) {
	params := SelectParams{
		QueryRequest: &logproto.QueryRequest{
			Start:     q.Start(),
			End:       q.End(),
			Limit:     q.Limit(),
			Direction: q.Direction(),
			Selector:  expr.String(),
		},
	}

	if GetRangeType(q) == InstantType {
		params.Start = params.Start.Add(-ev.maxLookBackPeriod)
	}

	return ev.querier.Select(ctx, params)

}

func (ev *defaultEvaluator) StepEvaluator(
	ctx context.Context,
	nextEv Evaluator,
	expr SampleExpr,
	q Params,
) (StepEvaluator, error) {
	switch e := expr.(type) {
	case *vectorAggregationExpr:
		return vectorAggEvaluator(ctx, nextEv, e, q)
	case *rangeAggregationExpr:
		entryIter, err := ev.querier.Select(ctx, SelectParams{
			&logproto.QueryRequest{
				Start:     q.Start().Add(-e.left.interval),
				End:       q.End(),
				Limit:     0,
				Direction: logproto.FORWARD,
				Selector:  expr.Selector().String(),
			},
		})
		if err != nil {
			return nil, err
		}
		return rangeAggEvaluator(entryIter, e, q)
	case *binOpExpr:
		return binOpStepEvaluator(ctx, nextEv, e, q)
	default:
		return nil, EvaluatorUnsupportedType(e, ev)
	}
}

func vectorAggEvaluator(
	ctx context.Context,
	ev Evaluator,
	expr *vectorAggregationExpr,
	q Params,
) (StepEvaluator, error) {
	nextEvaluator, err := ev.StepEvaluator(ctx, ev, expr.left, q)
	if err != nil {
		return nil, err
	}

	return newStepEvaluator(func() (bool, int64, promql.Vector) {
		next, ts, vec := nextEvaluator.Next()

		if !next {
			return false, 0, promql.Vector{}
		}
		result := map[uint64]*groupedAggregation{}
		if expr.operation == OpTypeTopK || expr.operation == OpTypeBottomK {
			if expr.params < 1 {
				return next, ts, promql.Vector{}
			}

		}
		for _, s := range vec {
			metric := s.Metric

			var (
				groupingKey uint64
			)
			if expr.grouping.without {
				groupingKey, _ = metric.HashWithoutLabels(make([]byte, 0, 1024), expr.grouping.groups...)
			} else {
				groupingKey, _ = metric.HashForLabels(make([]byte, 0, 1024), expr.grouping.groups...)
			}
			group, ok := result[groupingKey]
			// Add a new group if it doesn't exist.
			if !ok {
				var m labels.Labels

				if expr.grouping.without {
					lb := labels.NewBuilder(metric)
					lb.Del(expr.grouping.groups...)
					lb.Del(labels.MetricName)
					m = lb.Labels()
				} else {
					m = make(labels.Labels, 0, len(expr.grouping.groups))
					for _, l := range metric {
						for _, n := range expr.grouping.groups {
							if l.Name == n {
								m = append(m, l)
								break
							}
						}
					}
					sort.Sort(m)
				}
				result[groupingKey] = &groupedAggregation{
					labels:     m,
					value:      s.V,
					mean:       s.V,
					groupCount: 1,
				}

				inputVecLen := len(vec)
				resultSize := expr.params
				if expr.params > inputVecLen {
					resultSize = inputVecLen
				}
				if expr.operation == OpTypeStdvar || expr.operation == OpTypeStddev {
					result[groupingKey].value = 0.0
				} else if expr.operation == OpTypeTopK {
					result[groupingKey].heap = make(vectorByValueHeap, 0, resultSize)
					heap.Push(&result[groupingKey].heap, &promql.Sample{
						Point:  promql.Point{V: s.V},
						Metric: s.Metric,
					})
				} else if expr.operation == OpTypeBottomK {
					result[groupingKey].reverseHeap = make(vectorByReverseValueHeap, 0, resultSize)
					heap.Push(&result[groupingKey].reverseHeap, &promql.Sample{
						Point:  promql.Point{V: s.V},
						Metric: s.Metric,
					})
				}
				continue
			}
			switch expr.operation {
			case OpTypeSum:
				group.value += s.V

			case OpTypeAvg:
				group.groupCount++
				group.mean += (s.V - group.mean) / float64(group.groupCount)

			case OpTypeMax:
				if group.value < s.V || math.IsNaN(group.value) {
					group.value = s.V
				}

			case OpTypeMin:
				if group.value > s.V || math.IsNaN(group.value) {
					group.value = s.V
				}

			case OpTypeCount:
				group.groupCount++

			case OpTypeStddev, OpTypeStdvar:
				group.groupCount++
				delta := s.V - group.mean
				group.mean += delta / float64(group.groupCount)
				group.value += delta * (s.V - group.mean)

			case OpTypeTopK:
				if len(group.heap) < expr.params || group.heap[0].V < s.V || math.IsNaN(group.heap[0].V) {
					if len(group.heap) == expr.params {
						heap.Pop(&group.heap)
					}
					heap.Push(&group.heap, &promql.Sample{
						Point:  promql.Point{V: s.V},
						Metric: s.Metric,
					})
				}

			case OpTypeBottomK:
				if len(group.reverseHeap) < expr.params || group.reverseHeap[0].V > s.V || math.IsNaN(group.reverseHeap[0].V) {
					if len(group.reverseHeap) == expr.params {
						heap.Pop(&group.reverseHeap)
					}
					heap.Push(&group.reverseHeap, &promql.Sample{
						Point:  promql.Point{V: s.V},
						Metric: s.Metric,
					})
				}
			default:
				panic(errors.Errorf("expected aggregation operator but got %q", expr.operation))
			}
		}
		vec = vec[:0]
		for _, aggr := range result {
			switch expr.operation {
			case OpTypeAvg:
				aggr.value = aggr.mean

			case OpTypeCount:
				aggr.value = float64(aggr.groupCount)

			case OpTypeStddev:
				aggr.value = math.Sqrt(aggr.value / float64(aggr.groupCount))

			case OpTypeStdvar:
				aggr.value = aggr.value / float64(aggr.groupCount)

			case OpTypeTopK:
				// The heap keeps the lowest value on top, so reverse it.
				sort.Sort(sort.Reverse(aggr.heap))
				for _, v := range aggr.heap {
					vec = append(vec, promql.Sample{
						Metric: v.Metric,
						Point: promql.Point{
							T: ts,
							V: v.V,
						},
					})
				}
				continue // Bypass default append.

			case OpTypeBottomK:
				// The heap keeps the lowest value on top, so reverse it.
				sort.Sort(sort.Reverse(aggr.reverseHeap))
				for _, v := range aggr.reverseHeap {
					vec = append(vec, promql.Sample{
						Metric: v.Metric,
						Point: promql.Point{
							T: ts,
							V: v.V,
						},
					})
				}
				continue // Bypass default append.
			default:
			}
			vec = append(vec, promql.Sample{
				Metric: aggr.labels,
				Point: promql.Point{
					T: ts,
					V: aggr.value,
				},
			})
		}
		return next, ts, vec

	}, nextEvaluator.Close)
}

func rangeAggEvaluator(
	entryIter iter.EntryIterator,
	expr *rangeAggregationExpr,
	q Params,
) (StepEvaluator, error) {
	vecIter := newRangeVectorIterator(entryIter, expr.left.interval.Nanoseconds(), q.Step().Nanoseconds(),
		q.Start().UnixNano(), q.End().UnixNano())

	var fn RangeVectorAggregator
	switch expr.operation {
	case OpTypeRate:
		fn = rate(expr.left.interval)
	case OpTypeCountOverTime:
		fn = count
	}

	return newStepEvaluator(func() (bool, int64, promql.Vector) {
		next := vecIter.Next()
		if !next {
			return false, 0, promql.Vector{}
		}
		ts, vec := vecIter.At(fn)
		return true, ts, vec

	}, vecIter.Close)
}

// binOpExpr explicly does not handle when both legs are literals as
// it makes the type system simpler and these are reduced in mustNewBinOpExpr
func binOpStepEvaluator(
	ctx context.Context,
	ev Evaluator,
	expr *binOpExpr,
	q Params,
) (StepEvaluator, error) {
	// first check if either side is a literal
	leftLit, lOk := expr.SampleExpr.(*literalExpr)
	rightLit, rOk := expr.RHS.(*literalExpr)

	// match a literal expr with all labels in the other leg
	if lOk {
		rhs, err := ev.StepEvaluator(ctx, ev, expr.RHS, q)
		if err != nil {
			return nil, err
		}
		return literalStepEvaluator(expr.op, leftLit, rhs, false)
	}
	if rOk {
		lhs, err := ev.StepEvaluator(ctx, ev, expr.SampleExpr, q)
		if err != nil {
			return nil, err
		}
		return literalStepEvaluator(expr.op, rightLit, lhs, true)
	}

	// we have two non literal legs
	lhs, err := ev.StepEvaluator(ctx, ev, expr.SampleExpr, q)
	if err != nil {
		return nil, err
	}
	rhs, err := ev.StepEvaluator(ctx, ev, expr.RHS, q)
	if err != nil {
		return nil, err
	}

	return newStepEvaluator(func() (bool, int64, promql.Vector) {
		pairs := map[uint64][2]*promql.Sample{}
		var ts int64

		// populate pairs
		for i, eval := range []StepEvaluator{lhs, rhs} {
			next, timestamp, vec := eval.Next()

			ts = timestamp

			// These should _always_ happen at the same step on each evaluator.
			if !next {
				return next, ts, nil
			}

			for _, sample := range vec {
				// TODO(owen-d): this seems wildly inefficient: we're calculating
				// the hash on each sample & step per evaluator.
				// We seem limited to this approach due to using the StepEvaluator ifc.
				hash := sample.Metric.Hash()
				pair := pairs[hash]
				pair[i] = &promql.Sample{
					Metric: sample.Metric,
					Point:  sample.Point,
				}
				pairs[hash] = pair
			}
		}

		results := make(promql.Vector, 0, len(pairs))
		for _, pair := range pairs {

			// merge
			if merged := mergeBinOp(expr.op, pair[0], pair[1]); merged != nil {
				results = append(results, *merged)
			}
		}

		return true, ts, results
	}, func() (lastError error) {
		for _, ev := range []StepEvaluator{lhs, rhs} {
			if err := ev.Close(); err != nil {
				lastError = err
			}
		}
		return lastError
	})
}

func mergeBinOp(op string, left, right *promql.Sample) *promql.Sample {
	var merger func(left, right *promql.Sample) *promql.Sample

	switch op {
	case OpTypeOr:
		merger = func(left, right *promql.Sample) *promql.Sample {
			// return the left entry found (prefers left hand side)
			if left != nil {
				return left
			}
			return right
		}

	case OpTypeAnd:
		merger = func(left, right *promql.Sample) *promql.Sample {
			// return left sample if there's a second sample for that label set
			if left != nil && right != nil {
				return left
			}
			return nil
		}

	case OpTypeUnless:
		merger = func(left, right *promql.Sample) *promql.Sample {
			// return left sample if there's not a second sample for that label set
			if right == nil {
				return left
			}
			return nil
		}

	case OpTypeAdd:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}
			res := promql.Sample{
				Metric: left.Metric,
				Point:  left.Point,
			}
			res.Point.V += right.Point.V
			return &res
		}

	case OpTypeSub:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}
			res := promql.Sample{
				Metric: left.Metric,
				Point:  left.Point,
			}
			res.Point.V -= right.Point.V
			return &res
		}

	case OpTypeMul:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}
			res := promql.Sample{
				Metric: left.Metric,
				Point:  left.Point,
			}
			res.Point.V *= right.Point.V
			return &res
		}

	case OpTypeDiv:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}
			res := promql.Sample{
				Metric: left.Metric.Copy(),
				Point:  left.Point,
			}

			// guard against divide by zero
			if right.Point.V == 0 {
				res.Point.V = math.NaN()
			} else {
				res.Point.V /= right.Point.V
			}
			return &res
		}

	case OpTypeMod:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}
			res := promql.Sample{
				Metric: left.Metric,
				Point:  left.Point,
			}
			// guard against divide by zero
			if right.Point.V == 0 {
				res.Point.V = math.NaN()
			} else {
				res.Point.V = math.Mod(res.Point.V, right.Point.V)
			}
			return &res
		}

	case OpTypePow:
		merger = func(left, right *promql.Sample) *promql.Sample {
			if left == nil || right == nil {
				return nil
			}

			res := promql.Sample{
				Metric: left.Metric,
				Point:  left.Point,
			}
			res.Point.V = math.Pow(left.Point.V, right.Point.V)
			return &res
		}

	default:
		panic(errors.Errorf("should never happen: unexpected operation: (%s)", op))
	}

	return merger(left, right)

}

// literalStepEvaluator merges a literal with a StepEvaluator. Since order matters in
// non commutative operations, inverted should be true when the literalExpr is not the left argument.
func literalStepEvaluator(
	op string,
	lit *literalExpr,
	eval StepEvaluator,
	inverted bool,
) (StepEvaluator, error) {
	return newStepEvaluator(
		func() (bool, int64, promql.Vector) {
			ok, ts, vec := eval.Next()

			results := make(promql.Vector, 0, len(vec))
			for _, sample := range vec {
				literalPoint := promql.Sample{
					Metric: sample.Metric,
					Point:  promql.Point{T: ts, V: lit.value},
				}

				left, right := &literalPoint, &sample
				if inverted {
					left, right = right, left
				}

				if merged := mergeBinOp(
					op,
					left,
					right,
				); merged != nil {
					results = append(results, *merged)
				}
			}

			return ok, ts, results
		},
		eval.Close,
	)
}