Adding upstream version 2.5.1.

Signed-off-by: Daniel Baumann <daniel@debian.org>

index/scorch/mergeplan/merge_plan.go

@@ -0,0 +1,454 @@
+//  Copyright (c) 2017 Couchbase, Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// 		http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package mergeplan provides a segment merge planning approach that's
+// inspired by Lucene's TieredMergePolicy.java and descriptions like
+// http://blog.mikemccandless.com/2011/02/visualizing-lucenes-segment-merges.html
+package mergeplan
+
+import (
+	"errors"
+	"fmt"
+	"math"
+	"sort"
+	"strings"
+)
+
+// A Segment represents the information that the planner needs to
+// calculate segment merging.
+type Segment interface {
+	// Unique id of the segment -- used for sorting.
+	Id() uint64
+
+	// Full segment size (the size before any logical deletions).
+	FullSize() int64
+
+	// Size of the live data of the segment; i.e., FullSize() minus
+	// any logical deletions.
+	LiveSize() int64
+
+	HasVector() bool
+
+	// Size of the persisted segment file.
+	FileSize() int64
+}
+
+// Plan() will functionally compute a merge plan.  A segment will be
+// assigned to at most a single MergeTask in the output MergePlan.  A
+// segment not assigned to any MergeTask means the segment should
+// remain unmerged.
+func Plan(segments []Segment, o *MergePlanOptions) (*MergePlan, error) {
+	return plan(segments, o)
+}
+
+// A MergePlan is the result of the Plan() API.
+//
+// The planner doesn’t know how or whether these tasks are executed --
+// that’s up to a separate merge execution system, which might execute
+// these tasks concurrently or not, and which might execute all the
+// tasks or not.
+type MergePlan struct {
+	Tasks []*MergeTask
+}
+
+// A MergeTask represents several segments that should be merged
+// together into a single segment.
+type MergeTask struct {
+	Segments []Segment
+}
+
+// The MergePlanOptions is designed to be reusable between planning calls.
+type MergePlanOptions struct {
+	// Max # segments per logarithmic tier, or max width of any
+	// logarithmic “step”.  Smaller values mean more merging but fewer
+	// segments.  Should be >= SegmentsPerMergeTask, else you'll have
+	// too much merging.
+	MaxSegmentsPerTier int
+
+	// Max size of any segment produced after merging.  Actual
+	// merging, however, may produce segment sizes different than the
+	// planner’s predicted sizes.
+	MaxSegmentSize int64
+
+	// Max size (in bytes) of the persisted segment file that contains the
+	// vectors.  This is used to prevent merging of segments that
+	// contain vectors that are too large.
+	MaxSegmentFileSize int64
+
+	// The growth factor for each tier in a staircase of idealized
+	// segments computed by CalcBudget().
+	TierGrowth float64
+
+	// The number of segments in any resulting MergeTask.  e.g.,
+	// len(result.Tasks[ * ].Segments) == SegmentsPerMergeTask.
+	SegmentsPerMergeTask int
+
+	// Small segments are rounded up to this size, i.e., treated as
+	// equal (floor) size for consideration.  This is to prevent lots
+	// of tiny segments from resulting in a long tail in the index.
+	FloorSegmentSize int64
+
+	// Small segments' file size are rounded up to this size to prevent lot
+	// of tiny segments causing a long tail in the index.
+	FloorSegmentFileSize int64
+
+	// Controls how aggressively merges that reclaim more deletions
+	// are favored.  Higher values will more aggressively target
+	// merges that reclaim deletions, but be careful not to go so high
+	// that way too much merging takes place; a value of 3.0 is
+	// probably nearly too high.  A value of 0.0 means deletions don't
+	// impact merge selection.
+	ReclaimDeletesWeight float64
+
+	// Optional, defaults to mergeplan.CalcBudget().
+	CalcBudget func(totalSize int64, firstTierSize int64,
+		o *MergePlanOptions) (budgetNumSegments int)
+
+	// Optional, defaults to mergeplan.ScoreSegments().
+	ScoreSegments func(segments []Segment, o *MergePlanOptions) float64
+
+	// Optional.
+	Logger func(string)
+}
+
+// Returns the higher of the input or FloorSegmentSize.
+func (o *MergePlanOptions) RaiseToFloorSegmentSize(s int64) int64 {
+	if s > o.FloorSegmentSize {
+		return s
+	}
+	return o.FloorSegmentSize
+}
+
+func (o *MergePlanOptions) RaiseToFloorSegmentFileSize(s int64) int64 {
+	if s > o.FloorSegmentFileSize {
+		return s
+	}
+	return o.FloorSegmentFileSize
+}
+
+// MaxSegmentSizeLimit represents the maximum size of a segment,
+// this limit comes with hit-1 optimisation/max encoding limit uint31.
+const MaxSegmentSizeLimit = 1<<31 - 1
+
+// ErrMaxSegmentSizeTooLarge is returned when the size of the segment
+// exceeds the MaxSegmentSizeLimit
+var ErrMaxSegmentSizeTooLarge = errors.New("MaxSegmentSize exceeds the size limit")
+
+// DefaultMergePlanOptions suggests the default options.
+var DefaultMergePlanOptions = MergePlanOptions{
+	MaxSegmentsPerTier:   10,
+	MaxSegmentSize:       5000000,
+	MaxSegmentFileSize:   4000000000, // 4GB
+	TierGrowth:           10.0,
+	SegmentsPerMergeTask: 10,
+	FloorSegmentSize:     2000,
+	ReclaimDeletesWeight: 2.0,
+}
+
+// SingleSegmentMergePlanOptions helps in creating a
+// single segment index.
+var SingleSegmentMergePlanOptions = MergePlanOptions{
+	MaxSegmentsPerTier:   1,
+	MaxSegmentSize:       1 << 30,
+	MaxSegmentFileSize:   1 << 40,
+	TierGrowth:           1.0,
+	SegmentsPerMergeTask: 10,
+	FloorSegmentSize:     1 << 30,
+	ReclaimDeletesWeight: 2.0,
+	FloorSegmentFileSize: 1 << 40,
+}
+
+// -------------------------------------------
+
+func plan(segmentsIn []Segment, o *MergePlanOptions) (*MergePlan, error) {
+	if len(segmentsIn) <= 1 {
+		return nil, nil
+	}
+
+	if o == nil {
+		o = &DefaultMergePlanOptions
+	}
+
+	segments := append([]Segment(nil), segmentsIn...) // Copy.
+
+	sort.Sort(byLiveSizeDescending(segments))
+
+	var minLiveSize int64 = math.MaxInt64
+
+	var eligibles []Segment
+	var eligiblesLiveSize int64
+	var eligiblesFileSize int64
+	var minFileSize int64 = math.MaxInt64
+
+	for _, segment := range segments {
+		if minLiveSize > segment.LiveSize() {
+			minLiveSize = segment.LiveSize()
+		}
+
+		if minFileSize > segment.FileSize() {
+			minFileSize = segment.FileSize()
+		}
+
+		isEligible := segment.LiveSize() < o.MaxSegmentSize/2
+		// An eligible segment (based on #documents) may be too large
+		// and thus need a stricter check based on the file size.
+		// This is particularly important for segments that contain
+		// vectors.
+		if isEligible && segment.HasVector() && o.MaxSegmentFileSize > 0 {
+			isEligible = segment.FileSize() < o.MaxSegmentFileSize/2
+		}
+
+		// Only small-enough segments are eligible.
+		if isEligible {
+			eligibles = append(eligibles, segment)
+			eligiblesLiveSize += segment.LiveSize()
+			eligiblesFileSize += segment.FileSize()
+		}
+	}
+
+	calcBudget := o.CalcBudget
+	if calcBudget == nil {
+		calcBudget = CalcBudget
+	}
+
+	var budgetNumSegments int
+	if o.FloorSegmentFileSize > 0 {
+		minFileSize = o.RaiseToFloorSegmentFileSize(minFileSize)
+		budgetNumSegments = calcBudget(eligiblesFileSize, minFileSize, o)
+
+	} else {
+		minLiveSize = o.RaiseToFloorSegmentSize(minLiveSize)
+		budgetNumSegments = calcBudget(eligiblesLiveSize, minLiveSize, o)
+	}
+
+	scoreSegments := o.ScoreSegments
+	if scoreSegments == nil {
+		scoreSegments = ScoreSegments
+	}
+
+	rv := &MergePlan{}
+
+	var empties []Segment
+	for _, eligible := range eligibles {
+		if eligible.LiveSize() <= 0 {
+			empties = append(empties, eligible)
+		}
+	}
+	if len(empties) > 0 {
+		rv.Tasks = append(rv.Tasks, &MergeTask{Segments: empties})
+		eligibles = removeSegments(eligibles, empties)
+	}
+
+	// While we’re over budget, keep looping, which might produce
+	// another MergeTask.
+	for len(eligibles) > 0 && (len(eligibles)+len(rv.Tasks)) > budgetNumSegments {
+		// Track a current best roster as we examine and score
+		// potential rosters of merges.
+		var bestRoster []Segment
+		var bestRosterScore float64 // Lower score is better.
+
+		for startIdx := 0; startIdx < len(eligibles); startIdx++ {
+			var roster []Segment
+			var rosterLiveSize int64
+			var rosterFileSize int64 // useful for segments with vectors
+
+			for idx := startIdx; idx < len(eligibles) && len(roster) < o.SegmentsPerMergeTask; idx++ {
+				eligible := eligibles[idx]
+
+				if rosterLiveSize+eligible.LiveSize() >= o.MaxSegmentSize {
+					continue
+				}
+
+				if eligible.HasVector() {
+					efs := eligible.FileSize()
+					if rosterFileSize+efs >= o.MaxSegmentFileSize {
+						continue
+					}
+					rosterFileSize += efs
+				}
+
+				roster = append(roster, eligible)
+				rosterLiveSize += eligible.LiveSize()
+			}
+
+			if len(roster) > 0 {
+				rosterScore := scoreSegments(roster, o)
+
+				if len(bestRoster) == 0 || rosterScore < bestRosterScore {
+					bestRoster = roster
+					bestRosterScore = rosterScore
+				}
+			}
+		}
+
+		if len(bestRoster) == 0 {
+			return rv, nil
+		}
+
+		rv.Tasks = append(rv.Tasks, &MergeTask{Segments: bestRoster})
+
+		eligibles = removeSegments(eligibles, bestRoster)
+	}
+
+	return rv, nil
+}
+
+// Compute the number of segments that would be needed to cover the
+// totalSize, by climbing up a logarithmically growing staircase of
+// segment tiers.
+func CalcBudget(totalSize int64, firstTierSize int64, o *MergePlanOptions) (
+	budgetNumSegments int) {
+	tierSize := firstTierSize
+	if tierSize < 1 {
+		tierSize = 1
+	}
+
+	maxSegmentsPerTier := o.MaxSegmentsPerTier
+	if maxSegmentsPerTier < 1 {
+		maxSegmentsPerTier = 1
+	}
+
+	tierGrowth := o.TierGrowth
+	if tierGrowth < 1.0 {
+		tierGrowth = 1.0
+	}
+
+	for totalSize > 0 {
+		segmentsInTier := float64(totalSize) / float64(tierSize)
+		if segmentsInTier < float64(maxSegmentsPerTier) {
+			budgetNumSegments += int(math.Ceil(segmentsInTier))
+			break
+		}
+
+		budgetNumSegments += maxSegmentsPerTier
+		totalSize -= int64(maxSegmentsPerTier) * tierSize
+		tierSize = int64(float64(tierSize) * tierGrowth)
+	}
+
+	return budgetNumSegments
+}
+
+// Of note, removeSegments() keeps the ordering of the results stable.
+func removeSegments(segments []Segment, toRemove []Segment) []Segment {
+	rv := make([]Segment, 0, len(segments)-len(toRemove))
+OUTER:
+	for _, segment := range segments {
+		for _, r := range toRemove {
+			if segment == r {
+				continue OUTER
+			}
+		}
+		rv = append(rv, segment)
+	}
+	return rv
+}
+
+// Smaller result score is better.
+func ScoreSegments(segments []Segment, o *MergePlanOptions) float64 {
+	var totBeforeSize int64
+	var totAfterSize int64
+	var totAfterSizeFloored int64
+
+	for _, segment := range segments {
+		totBeforeSize += segment.FullSize()
+		totAfterSize += segment.LiveSize()
+		totAfterSizeFloored += o.RaiseToFloorSegmentSize(segment.LiveSize())
+	}
+
+	if totBeforeSize <= 0 || totAfterSize <= 0 || totAfterSizeFloored <= 0 {
+		return 0
+	}
+
+	// Roughly guess the "balance" of the segments -- whether the
+	// segments are about the same size.
+	balance :=
+		float64(o.RaiseToFloorSegmentSize(segments[0].LiveSize())) /
+			float64(totAfterSizeFloored)
+
+	// Gently favor smaller merges over bigger ones.  We don't want to
+	// make the exponent too large else we end up with poor merges of
+	// small segments in order to avoid the large merges.
+	score := balance * math.Pow(float64(totAfterSize), 0.05)
+
+	// Strongly favor merges that reclaim deletes.
+	nonDelRatio := float64(totAfterSize) / float64(totBeforeSize)
+
+	score *= math.Pow(nonDelRatio, o.ReclaimDeletesWeight)
+
+	return score
+}
+
+// ------------------------------------------
+
+// ToBarChart returns an ASCII rendering of the segments and the plan.
+// The barMax is the max width of the bars in the bar chart.
+func ToBarChart(prefix string, barMax int, segments []Segment, plan *MergePlan) string {
+	rv := make([]string, 0, len(segments))
+
+	var maxFullSize int64
+	for _, segment := range segments {
+		if maxFullSize < segment.FullSize() {
+			maxFullSize = segment.FullSize()
+		}
+	}
+	if maxFullSize < 0 {
+		maxFullSize = 1
+	}
+
+	for _, segment := range segments {
+		barFull := int(segment.FullSize())
+		barLive := int(segment.LiveSize())
+
+		if maxFullSize > int64(barMax) {
+			barFull = int(float64(barMax) * float64(barFull) / float64(maxFullSize))
+			barLive = int(float64(barMax) * float64(barLive) / float64(maxFullSize))
+		}
+
+		barKind := " "
+		barChar := "."
+
+		if plan != nil {
+		TASK_LOOP:
+			for taski, task := range plan.Tasks {
+				for _, taskSegment := range task.Segments {
+					if taskSegment == segment {
+						barKind = "*"
+						barChar = fmt.Sprintf("%d", taski)
+						break TASK_LOOP
+					}
+				}
+			}
+		}
+
+		bar :=
+			strings.Repeat(barChar, barLive)[0:barLive] +
+				strings.Repeat("x", barFull-barLive)[0:barFull-barLive]
+
+		rv = append(rv, fmt.Sprintf("%s %5d: %5d /%5d - %s %s", prefix,
+			segment.Id(),
+			segment.LiveSize(),
+			segment.FullSize(),
+			barKind, bar))
+	}
+
+	return strings.Join(rv, "\n")
+}
+
+// ValidateMergePlannerOptions validates the merge planner options
+func ValidateMergePlannerOptions(options *MergePlanOptions) error {
+	if options.MaxSegmentSize > MaxSegmentSizeLimit {
+		return ErrMaxSegmentSizeTooLarge
+	}
+	return nil
+}