statistics

package
v2.1.0-alpha+incompatible Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jun 4, 2018 License: Apache-2.0 Imports: 39 Imported by: 1,365

Documentation

Index

Constants

View Source 
const (
	// StatsOwnerKey is the stats owner path that is saved to etcd.
	StatsOwnerKey = "/tidb/stats/owner"
	// StatsPrompt is the prompt for stats owner manager.
	StatsPrompt = "stats"
)

Variables

View Source 
var (
	// MaxNumberOfRanges is the max number of ranges before split to collect feedback.
	MaxNumberOfRanges = 20
	// FeedbackProbability is the probability to collect the feedback.
	FeedbackProbability = 0.0
)
View Source 
var AutoAnalyzeMinCnt int64 = 1000

AutoAnalyzeMinCnt means if the count of table is less than this value, we needn't do auto analyze.

View Source 
var MaxQueryFeedbackCount = 1 << 10

MaxQueryFeedbackCount is the max number of feedback that cache in memory.

Functions

func CMSketchToProto 

func CMSketchToProto(c *CMSketch) *tipb.CMSketch

CMSketchToProto converts CMSketch to its protobuf representation.

func FMSketchToProto 

func FMSketchToProto(s *FMSketch) *tipb.FMSketch

FMSketchToProto converts FMSketch to its protobuf representation.

func HistogramEqual 

func HistogramEqual(a, b *Histogram, ignoreID bool) bool

HistogramEqual tests if two histograms are equal.

func HistogramToProto 

func HistogramToProto(hg *Histogram) *tipb.Histogram

HistogramToProto converts Histogram to its protobuf representation. Note that when this is used, the lower/upper bound in the bucket must be BytesDatum.

func SampleCollectorToProto 

func SampleCollectorToProto(c *SampleCollector) *tipb.SampleCollector

SampleCollectorToProto converts SampleCollector to its protobuf representation.

func SaveMetaToStorage 

func SaveMetaToStorage(sctx sessionctx.Context, tableID, count, modifyCount int64) error

SaveMetaToStorage will save stats_meta to storage.

func SaveStatsToStorage 

func SaveStatsToStorage(sctx sessionctx.Context, tableID int64, count int64, isIndex int, hg *Histogram, cms *CMSketch) error

SaveStatsToStorage saves the stats to storage.

func ValueToString 

func ValueToString(value *types.Datum, idxCols int) (string, error)

ValueToString converts a possible encoded value to a formatted string. If the value is encoded, then idxCols equals to number of origin values, else idxCols is 0.

Types

type AnalyzeResult 

type AnalyzeResult struct {
	TableID int64
	Hist    []*Histogram
	Cms     []*CMSketch
	Count   int64
	IsIndex int
	Err     error
}

AnalyzeResult is used to represent analyze result.

type Bucket 

type Bucket struct {
	Count  int64
	Repeat int64
}

Bucket store the bucket count and repeat.

type BucketFeedback 

type BucketFeedback struct {
	// contains filtered or unexported fields
}

BucketFeedback stands for all the feedback for a bucket.

type CMSketch 

type CMSketch struct {
	// contains filtered or unexported fields
}

CMSketch is used to estimate point queries. Refer: https://en.wikipedia.org/wiki/Count-min_sketch

func CMSketchFromProto 

func CMSketchFromProto(protoSketch *tipb.CMSketch) *CMSketch

CMSketchFromProto converts CMSketch from its protobuf representation.

func NewCMSketch 

func NewCMSketch(d, w int32) *CMSketch

NewCMSketch returns a new CM sketch.

func (*CMSketch) Equal 

func (c *CMSketch) Equal(rc *CMSketch) bool

Equal tests if two CM Sketch equal, it is only used for test.

func (*CMSketch) InsertBytes 

func (c *CMSketch) InsertBytes(bytes []byte)

InsertBytes inserts the bytes value into the CM Sketch.

func (*CMSketch) MergeCMSketch 

func (c *CMSketch) MergeCMSketch(rc *CMSketch) error

MergeCMSketch merges two CM Sketch.

func (*CMSketch) TotalCount 

func (c *CMSketch) TotalCount() uint64

TotalCount returns the count, it is only used for test.

type Column 

type Column struct {
	Histogram
	*CMSketch
	Count int64
	Info  *model.ColumnInfo
}

Column represents a column histogram.

func (*Column) AvgColSize 

func (c *Column) AvgColSize(count int64) float64

AvgColSize is the average column size of the histogram.

func (*Column) String 

func (c *Column) String() string

type FMSketch 

type FMSketch struct {
	// contains filtered or unexported fields
}

FMSketch is used to count the number of distinct elements in a set.

func FMSketchFromProto 

func FMSketchFromProto(protoSketch *tipb.FMSketch) *FMSketch

FMSketchFromProto converts FMSketch from its protobuf representation.

func NewFMSketch 

func NewFMSketch(maxSize int) *FMSketch

NewFMSketch returns a new FM sketch.

func (*FMSketch) InsertValue 

func (s *FMSketch) InsertValue(sc *stmtctx.StatementContext, value types.Datum) error

InsertValue inserts a value into the FM sketch.

func (*FMSketch) NDV 

func (s *FMSketch) NDV() int64

NDV returns the ndv of the sketch.

type Handle 

type Handle struct {

	// LastVersion is the latest update version before last lease. Exported for test.
	LastVersion uint64
	// PrevLastVersion is the latest update version before two lease. Exported for test.
	// We need this because for two tables, the smaller version may write later than the one with larger version.
	// We can read the version with lastTwoVersion if the diff between commit time and version is less than one lease.
	// PrevLastVersion will be assigned by LastVersion every time Update is called.
	PrevLastVersion uint64

	Lease time.Duration
	// contains filtered or unexported fields
}

Handle can update stats info periodically.

func NewHandle 

func NewHandle(ctx sessionctx.Context, lease time.Duration) *Handle

NewHandle creates a Handle for update stats.

func (*Handle) AnalyzeResultCh 

func (h *Handle) AnalyzeResultCh() chan *AnalyzeResult

AnalyzeResultCh returns analyze result channel in handle.

func (*Handle) Clear 

func (h *Handle) Clear()

Clear the statsCache, only for test.

func (*Handle) DDLEventCh 

func (h *Handle) DDLEventCh() chan *util.Event

DDLEventCh returns ddl events channel in handle.

func (*Handle) DeleteTableStatsFromKV 

func (h *Handle) DeleteTableStatsFromKV(id int64) error

DeleteTableStatsFromKV deletes table statistics from kv.

func (*Handle) DumpStatsDeltaToKV 

func (h *Handle) DumpStatsDeltaToKV() error

DumpStatsDeltaToKV sweeps the whole list and updates the global map. Then we dumps every table that held in map to KV.

func (*Handle) DumpStatsFeedbackToKV 

func (h *Handle) DumpStatsFeedbackToKV() error

DumpStatsFeedbackToKV dumps the stats feedback to KV.

func (*Handle) DumpStatsToJSON 

func (h *Handle) DumpStatsToJSON(dbName string, tableInfo *model.TableInfo) (*JSONTable, error)

DumpStatsToJSON dumps statistic to json.

func (*Handle) GCStats 

func (h *Handle) GCStats(is infoschema.InfoSchema, ddlLease time.Duration) error

GCStats will garbage collect the useless stats info. For dropped tables, we will first update their version so that other tidb could know that table is deleted.

func (*Handle) GetQueryFeedback 

func (h *Handle) GetQueryFeedback() []*QueryFeedback

GetQueryFeedback gets the query feedback. It is only use in test.

func (*Handle) GetTableStats 

func (h *Handle) GetTableStats(tblInfo *model.TableInfo) *Table

GetTableStats retrieves the statistics table from cache, and the cache will be updated by a goroutine.

func (*Handle) HandleAutoAnalyze 

func (h *Handle) HandleAutoAnalyze(is infoschema.InfoSchema) error

HandleAutoAnalyze analyzes the newly created table or index.

func (*Handle) HandleDDLEvent 

func (h *Handle) HandleDDLEvent(t *util.Event) error

HandleDDLEvent begins to process a ddl task.

func (*Handle) HandleUpdateStats 

func (h *Handle) HandleUpdateStats(is infoschema.InfoSchema) error

HandleUpdateStats update the stats using feedback.

func (*Handle) InitStats added in v1.0.5 

func (h *Handle) InitStats(is infoschema.InfoSchema) error

InitStats will init the stats cache using full load strategy.

func (*Handle) LoadMetaCh 

func (h *Handle) LoadMetaCh() chan *LoadMeta

LoadMetaCh returns loaded statistic meta channel in handle.

func (*Handle) LoadNeededHistograms 

func (h *Handle) LoadNeededHistograms() error

LoadNeededHistograms will load histograms for those needed columns.

func (*Handle) LoadStatsFromJSON 

func (h *Handle) LoadStatsFromJSON(is infoschema.InfoSchema, jsonTbl *JSONTable) error

LoadStatsFromJSON will load statistic from JSONTable, and save it to the storage.

func (*Handle) LoadStatsFromJSONToTable 

func (h *Handle) LoadStatsFromJSONToTable(tableInfo *model.TableInfo, jsonTbl *JSONTable) (*Table, error)

LoadStatsFromJSONToTable load statistic from JSONTable and return the Table of statistic.

func (*Handle) NewSessionStatsCollector 

func (h *Handle) NewSessionStatsCollector() *SessionStatsCollector

NewSessionStatsCollector allocates a stats collector for a session.

func (*Handle) Update 

func (h *Handle) Update(is infoschema.InfoSchema) error

Update reads stats meta from store and updates the stats map.

func (*Handle) UpdateTableStats 

func (h *Handle) UpdateTableStats(tables []*Table, deletedIDs []int64)

UpdateTableStats updates the statistics table cache using copy on write.

type Histogram 

type Histogram struct {
	ID        int64 // Column ID.
	NDV       int64 // Number of distinct values.
	NullCount int64 // Number of null values.
	// LastUpdateVersion is the version that this histogram updated last time.
	LastUpdateVersion uint64

	// Histogram elements.
	//
	// A bucket bound is the smallest and greatest values stored in the bucket. The lower and upper bound
	// are stored in one column.
	//
	// A bucket count is the number of items stored in all previous buckets and the current bucket.
	// Bucket counts are always in increasing order.
	//
	// A bucket repeat is the number of repeats of the bucket value, it can be used to find popular values.
	Bounds  *chunk.Chunk
	Buckets []Bucket

	// TotColSize is the total column size for the histogram.
	TotColSize int64
	// contains filtered or unexported fields
}

Histogram represents statistics for a column or index.

func BuildColumn 

func BuildColumn(ctx sessionctx.Context, numBuckets, id int64, collector *SampleCollector, tp *types.FieldType) (*Histogram, error)

BuildColumn builds histogram from samples for column.

func HistogramFromProto 

func HistogramFromProto(protoHg *tipb.Histogram) *Histogram

HistogramFromProto converts Histogram from its protobuf representation. Note that we will set BytesDatum for the lower/upper bound in the bucket, the decode will be after all histograms merged.

func MergeHistograms 

func MergeHistograms(sc *stmtctx.StatementContext, lh *Histogram, rh *Histogram, bucketSize int) (*Histogram, error)

MergeHistograms merges two histograms.

func NewHistogram 

func NewHistogram(id, ndv, nullCount int64, version uint64, tp *types.FieldType, bucketSize int, totColSize int64) *Histogram

NewHistogram creates a new histogram.

func UpdateHistogram 

func UpdateHistogram(h *Histogram, feedback *QueryFeedback) *Histogram

UpdateHistogram updates the histogram according buckets.

func (*Histogram) AppendBucket 

func (hg *Histogram) AppendBucket(lower *types.Datum, upper *types.Datum, count, repeat int64)

AppendBucket appends a bucket into `hg`.

func (*Histogram) ConvertTo 

func (hg *Histogram) ConvertTo(sc *stmtctx.StatementContext, tp *types.FieldType) (*Histogram, error)

ConvertTo converts the histogram bucket values into `tp`.

func (*Histogram) DecodeTo 

func (hg *Histogram) DecodeTo(tp *types.FieldType, timeZone *time.Location) error

DecodeTo decodes the histogram bucket values into `tp`.

func (*Histogram) GetLower 

func (hg *Histogram) GetLower(idx int) *types.Datum

GetLower gets the lower bound of bucket `idx`.

func (*Histogram) GetUpper 

func (hg *Histogram) GetUpper(idx int) *types.Datum

GetUpper gets the upper bound of bucket `idx`.

func (*Histogram) Len 

func (hg *Histogram) Len() int

Len is the number of buckets in the histogram.

func (*Histogram) PreCalculateScalar 

func (hg *Histogram) PreCalculateScalar()

PreCalculateScalar converts the lower and upper to scalar. When the datum type is KindString or KindBytes, we also calculate their common prefix length, because when a value falls between lower and upper, the common prefix of lower and upper equals to the common prefix of the lower, upper and the value. For some simple types like `Int64`, we do not convert it because we can directly infer the scalar value.

func (*Histogram) SplitRange 

func (hg *Histogram) SplitRange(ranges []*ranger.Range) []*ranger.Range

SplitRange splits the range according to the histogram upper bound. Note that we treat last bucket's upper bound as inf, so all the split ranges will totally fall in one of the (-inf, u(0)], (u(0), u(1)],...(u(n-3), u(n-2)], (u(n-2), +inf), where n is the number of buckets, u(i) is the i-th bucket's upper bound.

func (*Histogram) ToString 

func (hg *Histogram) ToString(idxCols int) string

ToString gets the string representation for the histogram.

type Index 

type Index struct {
	Histogram
	*CMSketch
	Info *model.IndexInfo
}

Index represents an index histogram.

func (*Index) String 

func (idx *Index) String() string

type JSONTable 

type JSONTable struct {
	DatabaseName string                 `json:"database_name"`
	TableName    string                 `json:"table_name"`
	Columns      map[string]*jsonColumn `json:"columns"`
	Indices      map[string]*jsonColumn `json:"indices"`
	Count        int64                  `json:"count"`
	ModifyCount  int64                  `json:"modify_count"`
}

JSONTable is used for dumping statistics.

type LoadMeta 

type LoadMeta struct {
	TableID     int64
	Count       int64
	ModifyCount int64
}

LoadMeta is the statistic meta loaded from json file.

type QueryFeedback 

type QueryFeedback struct {
	// contains filtered or unexported fields
}

QueryFeedback is used to represent the query feedback info. It contains the query's scan ranges and number of rows in each range.

func NewQueryFeedback 

func NewQueryFeedback(tableID int64, hist *Histogram, expected int64, desc bool) *QueryFeedback

NewQueryFeedback returns a new query feedback.

func (*QueryFeedback) Actual 

func (q *QueryFeedback) Actual() int64

Actual gets the actual row count.

func (*QueryFeedback) CollectFeedback 

func (q *QueryFeedback) CollectFeedback(numOfRanges int) bool

CollectFeedback decides whether to collect the feedback. It returns false when: 1: the histogram is nil or has no buckets; 2: the number of scan ranges exceeds the limit because it may affect the performance; 3: it does not pass the probabilistic sampler.

func (*QueryFeedback) Equal 

func (q *QueryFeedback) Equal(rq *QueryFeedback) bool

Equal tests if two query feedback equal, it is only used in test.

func (*QueryFeedback) Hist 

func (q *QueryFeedback) Hist() *Histogram

Hist gets the histogram.

func (*QueryFeedback) Invalidate 

func (q *QueryFeedback) Invalidate()

Invalidate is used to invalidate the query feedback.

func (*QueryFeedback) StoreRanges 

func (q *QueryFeedback) StoreRanges(ranges []*ranger.Range)

StoreRanges stores the ranges for update.

func (*QueryFeedback) Update 

func (q *QueryFeedback) Update(startKey kv.Key, counts []int64)

Update updates the query feedback. `startKey` is the start scan key of the partial result, used to find the range for update. `counts` is the scan counts of each range, used to update the feedback count info.

type SampleBuilder 

type SampleBuilder struct {
	Sc              *stmtctx.StatementContext
	RecordSet       ast.RecordSet
	ColLen          int // ColLen is the number of columns need to be sampled.
	PkBuilder       *SortedBuilder
	MaxBucketSize   int64
	MaxSampleSize   int64
	MaxFMSketchSize int64
	CMSketchDepth   int32
	CMSketchWidth   int32
}

SampleBuilder is used to build samples for columns. Also, if primary key is handle, it will directly build histogram for it.

func (SampleBuilder) CollectColumnStats 

func (s SampleBuilder) CollectColumnStats() ([]*SampleCollector, *SortedBuilder, error)

CollectColumnStats collects sample from the result set using Reservoir Sampling algorithm, and estimates NDVs using FM Sketch during the collecting process. It returns the sample collectors which contain total count, null count, distinct values count and CM Sketch. It also returns the statistic builder for PK which contains the histogram. See https://en.wikipedia.org/wiki/Reservoir_sampling

type SampleCollector 

type SampleCollector struct {
	Samples []types.Datum

	IsMerger      bool
	NullCount     int64
	Count         int64 // Count is the number of non-null rows.
	MaxSampleSize int64
	FMSketch      *FMSketch
	CMSketch      *CMSketch
	TotalSize     int64 // TotalSize is the total size of column.
	// contains filtered or unexported fields
}

SampleCollector will collect Samples and calculate the count and ndv of an attribute.

func SampleCollectorFromProto 

func SampleCollectorFromProto(collector *tipb.SampleCollector) *SampleCollector

SampleCollectorFromProto converts SampleCollector from its protobuf representation.

func (*SampleCollector) MergeSampleCollector 

func (c *SampleCollector) MergeSampleCollector(sc *stmtctx.StatementContext, rc *SampleCollector)

MergeSampleCollector merges two sample collectors.

type SessionStatsCollector 

type SessionStatsCollector struct {
	sync.Mutex
	// contains filtered or unexported fields
}

SessionStatsCollector is a list item that holds the delta mapper. If you want to write or read mapper, you must lock it.

func (*SessionStatsCollector) Delete 

func (s *SessionStatsCollector) Delete()

Delete only sets the deleted flag true, it will be deleted from list when DumpStatsDeltaToKV is called.

func (*SessionStatsCollector) StoreQueryFeedback 

func (s *SessionStatsCollector) StoreQueryFeedback(feedback interface{})

StoreQueryFeedback will merges the feedback into stats collector.

func (*SessionStatsCollector) Update 

func (s *SessionStatsCollector) Update(id int64, delta int64, count int64, colSize *map[int64]int64)

Update will updates the delta and count for one table id.

type SortedBuilder 

type SortedBuilder struct {
	Count int64
	// contains filtered or unexported fields
}

SortedBuilder is used to build histograms for PK and index.

func NewSortedBuilder 

func NewSortedBuilder(sc *stmtctx.StatementContext, numBuckets, id int64, tp *types.FieldType) *SortedBuilder

NewSortedBuilder creates a new SortedBuilder.

func (*SortedBuilder) Hist 

func (b *SortedBuilder) Hist() *Histogram

Hist returns the histogram built by SortedBuilder.

func (*SortedBuilder) Iterate 

func (b *SortedBuilder) Iterate(data types.Datum) error

Iterate updates the histogram incrementally.

type Table 

type Table struct {
	TableID     int64
	Columns     map[int64]*Column
	Indices     map[int64]*Index
	Count       int64 // Total row count in a table.
	ModifyCount int64 // Total modify count in a table.
	Version     uint64
	Pseudo      bool
	PKIsHandle  bool
}

Table represents statistics for a table.

func PseudoTable 

func PseudoTable(tblInfo *model.TableInfo) *Table

PseudoTable creates a pseudo table statistics.

func (*Table) ColumnBetweenRowCount 

func (t *Table) ColumnBetweenRowCount(sc *stmtctx.StatementContext, a, b types.Datum, colID int64) float64

ColumnBetweenRowCount estimates the row count where column greater or equal to a and less than b.

func (*Table) ColumnEqualRowCount 

func (t *Table) ColumnEqualRowCount(sc *stmtctx.StatementContext, value types.Datum, colID int64) (float64, error)

ColumnEqualRowCount estimates the row count where the column equals to value.

func (*Table) ColumnGreaterRowCount 

func (t *Table) ColumnGreaterRowCount(sc *stmtctx.StatementContext, value types.Datum, colID int64) float64

ColumnGreaterRowCount estimates the row count where the column greater than value.

func (*Table) ColumnIsInvalid 

func (t *Table) ColumnIsInvalid(sc *stmtctx.StatementContext, colID int64) bool

ColumnIsInvalid checks if this column is invalid. If this column has histogram but not loaded yet, then we mark it as need histogram.

func (*Table) ColumnLessRowCount 

func (t *Table) ColumnLessRowCount(sc *stmtctx.StatementContext, value types.Datum, colID int64) float64

ColumnLessRowCount estimates the row count where the column less than value.

func (*Table) GetRowCountByColumnRanges 

func (t *Table) GetRowCountByColumnRanges(sc *stmtctx.StatementContext, colID int64, colRanges []*ranger.Range) (float64, error)

GetRowCountByColumnRanges estimates the row count by a slice of Range.

func (*Table) GetRowCountByIndexRanges 

func (t *Table) GetRowCountByIndexRanges(sc *stmtctx.StatementContext, idxID int64, indexRanges []*ranger.Range) (float64, error)

GetRowCountByIndexRanges estimates the row count by a slice of Range.

func (*Table) GetRowCountByIntColumnRanges 

func (t *Table) GetRowCountByIntColumnRanges(sc *stmtctx.StatementContext, colID int64, intRanges []*ranger.Range) (float64, error)

GetRowCountByIntColumnRanges estimates the row count by a slice of IntColumnRange.

func (*Table) Selectivity 

func (t *Table) Selectivity(ctx sessionctx.Context, exprs []expression.Expression) (float64, error)

Selectivity is a function calculate the selectivity of the expressions. The definition of selectivity is (row count after filter / row count before filter). And exprs must be CNF now, in other words, `exprs[0] and exprs[1] and ... and exprs[len - 1]` should be held when you call this. TODO: support expressions that the top layer is a DNF. Currently the time complexity is o(n^2).

func (*Table) String 

func (t *Table) String() string

String implements Stringer interface.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.