// PlanSelect creates an execution plan for the given SelectStatement and returns an Executor.
func (q *QueryExecutor) PlanSelect(stmt *influxql.SelectStatement, chunkSize int) (Executor, 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{}
// Replace instances of "now()" with the current time, and check the resultant times.
stmt.Condition = influxql.Reduce(stmt.Condition, &influxql.NowValuer{Now: now})
opt.MinTime, opt.MaxTime = influxql.TimeRange(stmt.Condition)
if opt.MaxTime.IsZero() {
opt.MaxTime = now
}
if opt.MinTime.IsZero() {
opt.MinTime = time.Unix(0, 0)
}
// Expand regex sources to their actual source names.
sources, err := q.Store.ExpandSources(stmt.Sources)
if err != nil {
return nil, err
}
stmt.Sources = sources
// Convert DISTINCT into a call.
stmt.RewriteDistinct()
// Remove "time" from fields list.
stmt.RewriteTimeFields()
// Filter only shards that contain date range.
shardIDs, err := q.MetaClient.ShardIDsByTimeRange(stmt.Sources, opt.MinTime, opt.MaxTime)
if err != nil {
return nil, err
}
shards := q.Store.Shards(shardIDs)
// Rewrite wildcards, if any exist.
tmp, err := stmt.RewriteWildcards(Shards(shards))
if err != nil {
return nil, err
}
stmt = tmp
// Create a set of iterators from a selection.
itrs, err := influxql.Select(stmt, Shards(shards), &opt)
if err != nil {
return nil, err
}
// Generate a row emitter from the iterator set.
em := influxql.NewEmitter(itrs, stmt.TimeAscending())
em.Columns = stmt.ColumnNames()
em.OmitTime = stmt.OmitTime
// Wrap emitter in an adapter to conform to the Executor interface.
return (*emitterExecutor)(em), nil
}
func (e *StatementExecutor) executeSelectStatement(stmt *influxql.SelectStatement, ctx *influxql.ExecutionContext) error {
itrs, stmt, err := e.createIterators(stmt, ctx)
if err != nil {
return err
}
// 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()
// 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, partial, 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},
Partial: partial,
}
// Send results or exit if closing.
if err := ctx.Send(result); err != nil {
return err
}
emitted = true
}
// Flush remaining 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()))
}
return ctx.Send(&influxql.Result{
StatementID: ctx.StatementID,
Messages: messages,
Series: []*models.Row{{
Name: "result",
Columns: []string{"time", "written"},
Values: [][]interface{}{{time.Unix(0, 0).UTC(), writeN}},
}},
})
}
// Always emit at least one result.
if !emitted {
return ctx.Send(&influxql.Result{
StatementID: ctx.StatementID,
Series: make([]*models.Row, 0),
})
}
return nil
}
func (e *StatementExecutor) executeShowTagValues(stmt *influxql.SelectStatement, ctx *influxql.ExecutionContext, store LocalTSDBStore) error {
if stmt.Condition == nil {
return errors.New("a condition is required")
}
source := stmt.Sources[0].(*influxql.Measurement)
index := store.DatabaseIndex(source.Database)
if index == nil {
ctx.Results <- &influxql.Result{StatementID: ctx.StatementID, Series: make([]*models.Row, 0)}
return nil
}
measurementExpr := influxql.CloneExpr(stmt.Condition)
measurementExpr = influxql.Reduce(influxql.RewriteExpr(measurementExpr, func(e influxql.Expr) influxql.Expr {
switch e := e.(type) {
case *influxql.BinaryExpr:
switch e.Op {
case influxql.EQ, influxql.NEQ, influxql.EQREGEX, influxql.NEQREGEX:
tag, ok := e.LHS.(*influxql.VarRef)
if !ok || tag.Val != "_name" {
return nil
}
}
}
return e
}), nil)
mms, ok, err := index.MeasurementsByExpr(measurementExpr)
if err != nil {
return err
} else if !ok {
mms = index.Measurements()
sort.Sort(mms)
}
// If there are no measurements, return immediately.
if len(mms) == 0 {
ctx.Results <- &influxql.Result{StatementID: ctx.StatementID, Series: make([]*models.Row, 0)}
return nil
}
filterExpr := influxql.CloneExpr(stmt.Condition)
filterExpr = influxql.Reduce(influxql.RewriteExpr(filterExpr, func(e influxql.Expr) influxql.Expr {
switch e := e.(type) {
case *influxql.BinaryExpr:
switch e.Op {
case influxql.EQ, influxql.NEQ, influxql.EQREGEX, influxql.NEQREGEX:
tag, ok := e.LHS.(*influxql.VarRef)
if !ok || strings.HasPrefix(tag.Val, "_") {
return nil
}
}
}
return e
}), nil)
var emitted bool
columns := stmt.ColumnNames()
for _, mm := range mms {
ids, err := mm.SeriesIDsAllOrByExpr(filterExpr)
if err != nil {
return err
}
ss := mm.SeriesByIDSlice(ids)
// Determine a list of keys from condition.
keySet, ok, err := mm.TagKeysByExpr(stmt.Condition)
if err != nil {
return err
}
// Loop over all keys for each series.
m := make(map[keyValue]struct{}, len(ss))
for _, series := range ss {
for key, value := range series.Tags {
if !ok {
// nop
} else if _, exists := keySet[key]; !exists {
continue
}
m[keyValue{key, value}] = struct{}{}
}
}
// Move to next series if no key/values match.
if len(m) == 0 {
continue
}
// Sort key/value set.
a := make([]keyValue, 0, len(m))
for kv := range m {
a = append(a, kv)
}
sort.Sort(keyValues(a))
// Convert to result values.
slab := make([]interface{}, len(a)*2)
values := make([][]interface{}, len(a))
for i, elem := range a {
slab[i*2], slab[i*2+1] = elem.key, elem.value
values[i] = slab[i*2 : i*2+2]
}
// Send result to client.
ctx.Results <- &influxql.Result{
StatementID: ctx.StatementID,
Series: []*models.Row{&models.Row{
Name: mm.Name,
Columns: columns,
Values: values,
}},
}
emitted = true
}
// Always emit at least one row.
if !emitted {
ctx.Results <- &influxql.Result{StatementID: ctx.StatementID, Series: make([]*models.Row, 0)}
}
return nil
}
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
}
func (e *StatementExecutor) executeSelectStatement(stmt *influxql.SelectStatement, ctx *influxql.ExecutionContext) 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}
// Replace instances of "now()" with the current time, and check the resultant times.
stmt.Condition = influxql.Reduce(stmt.Condition, &influxql.NowValuer{Now: now})
var err error
opt.MinTime, opt.MaxTime, err = influxql.TimeRange(stmt.Condition)
if err != nil {
return err
}
if opt.MaxTime.IsZero() {
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.RewriteWildcards(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
for {
row := em.Emit()
if row == nil {
// Check if the query was interrupted while emitting.
select {
case <-ctx.InterruptCh:
return influxql.ErrQueryInterrupted
default:
}
break
}
result := &influxql.Result{
StatementID: ctx.StatementID,
Series: []*models.Row{row},
}
// Write points back into system for INTO statements.
if stmt.Target != nil {
if err := e.writeInto(stmt, row); err != nil {
return err
}
writeN += int64(len(row.Values))
continue
}
// Send results or exit if closing.
select {
case <-ctx.InterruptCh:
return influxql.ErrQueryInterrupted
case ctx.Results <- result:
}
emitted = true
}
// Emit write count if an INTO statement.
if stmt.Target != nil {
ctx.Results <- &influxql.Result{
StatementID: ctx.StatementID,
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
}
func (e *QueryExecutor) executeSelectStatement(stmt *influxql.SelectStatement, chunkSize, statementID int, results chan *influxql.Result, closing <-chan struct{}) 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{}
// Replace instances of "now()" with the current time, and check the resultant times.
stmt.Condition = influxql.Reduce(stmt.Condition, &influxql.NowValuer{Now: now})
opt.MinTime, opt.MaxTime = influxql.TimeRange(stmt.Condition)
if opt.MaxTime.IsZero() {
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.RewriteWildcards(ic)
if err != nil {
return err
}
stmt = tmp
// Create a set of iterators from a selection.
itrs, err := influxql.Select(stmt, ic, &opt)
if err != nil {
return err
}
// Generate a row emitter from the iterator set.
em := influxql.NewEmitter(itrs, stmt.TimeAscending())
em.Columns = stmt.ColumnNames()
em.OmitTime = stmt.OmitTime
defer em.Close()
// Emit rows to the results channel.
var writeN int64
var emitted bool
for {
row := em.Emit()
if row == nil {
break
}
result := &influxql.Result{
StatementID: statementID,
Series: []*models.Row{row},
}
// Write points back into system for INTO statements.
if stmt.Target != nil {
if err := e.writeInto(stmt, row); err != nil {
return err
}
writeN += int64(len(row.Values))
continue
}
// Send results or exit if closing.
select {
case <-closing:
return nil
case results <- result:
}
emitted = true
}
// Emit write count if an INTO statement.
if stmt.Target != nil {
results <- &influxql.Result{
StatementID: statementID,
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 {
results <- &influxql.Result{
StatementID: statementID,
Series: make([]*models.Row, 0),
}
}
return nil
}
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)
}
}
itrs, stmt, err := e.createIterators(stmt, ctx)
if err != nil {
return err
}
// 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
}