// createVarRefIterator creates an iterator for a variable reference.
func (e *Engine) createVarRefIterator(opt influxql.IteratorOptions) ([]influxql.Iterator, error) {
ref, _ := opt.Expr.(*influxql.VarRef)
var itrs []influxql.Iterator
if err := func() error {
mms := tsdb.Measurements(e.index.MeasurementsByName(influxql.Sources(opt.Sources).Names()))
// Retrieve the maximum number of fields (without time).
conditionFields := make([]string, len(influxql.ExprNames(opt.Condition)))
for _, mm := range mms {
// Determine tagsets for this measurement based on dimensions and filters.
tagSets, err := mm.TagSets(opt.Dimensions, opt.Condition)
if err != nil {
return err
}
// Calculate tag sets and apply SLIMIT/SOFFSET.
tagSets = influxql.LimitTagSets(tagSets, opt.SLimit, opt.SOffset)
for _, t := range tagSets {
for i, seriesKey := range t.SeriesKeys {
fields := 0
if t.Filters[i] != nil {
// Retrieve non-time fields from this series filter and filter out tags.
for _, f := range influxql.ExprNames(t.Filters[i]) {
if mm.HasField(f) {
conditionFields[fields] = f
fields++
}
}
}
itr, err := e.createVarRefSeriesIterator(ref, mm, seriesKey, t, t.Filters[i], conditionFields[:fields], opt)
if err != nil {
return err
} else if itr == nil {
continue
}
itrs = append(itrs, itr)
}
}
}
return nil
}(); err != nil {
influxql.Iterators(itrs).Close()
return nil, err
}
return itrs, nil
}
// ReadAll reads all points from all iterators.
func (itrs Iterators) ReadAll() [][]influxql.Point {
var a [][]influxql.Point
// Read from every iterator until a nil is encountered.
for {
points := itrs.Next()
if points == nil {
break
}
a = append(a, points)
}
// Close all iterators.
influxql.Iterators(itrs).Close()
return a
}
// createTagKeysIterator returns an iterator for all tag keys across measurements.
func (a Shards) createTagKeysIterator(opt influxql.IteratorOptions) (influxql.Iterator, error) {
itrs := make([]influxql.Iterator, 0, len(a))
if err := func() error {
for _, sh := range a {
itr, err := NewTagKeysIterator(sh, opt)
if err != nil {
return err
}
itrs = append(itrs, itr)
}
return nil
}(); err != nil {
influxql.Iterators(itrs).Close()
return nil, err
}
return influxql.NewMergeIterator(itrs, opt), nil
}
// createVarRefIterator creates an iterator for a variable reference.
func (e *Engine) createVarRefIterator(opt influxql.IteratorOptions) ([]influxql.Iterator, error) {
ref, _ := opt.Expr.(*influxql.VarRef)
var itrs []influxql.Iterator
if err := func() error {
mms := tsdb.Measurements(e.index.MeasurementsByName(influxql.Sources(opt.Sources).Names()))
for _, mm := range mms {
// Determine tagsets for this measurement based on dimensions and filters.
tagSets, err := mm.TagSets(opt.Dimensions, opt.Condition)
if err != nil {
return err
}
// Calculate tag sets and apply SLIMIT/SOFFSET.
tagSets = influxql.LimitTagSets(tagSets, opt.SLimit, opt.SOffset)
for _, t := range tagSets {
inputs, err := e.createTagSetIterators(ref, mm, t, opt)
if err != nil {
return err
}
if len(inputs) > 0 && (opt.Limit > 0 || opt.Offset > 0) {
var itr influxql.Iterator
if opt.MergeSorted() {
itr = influxql.NewSortedMergeIterator(inputs, opt)
} else {
itr = influxql.NewMergeIterator(inputs, opt)
}
itrs = append(itrs, newLimitIterator(itr, opt))
} else {
itrs = append(itrs, inputs...)
}
}
}
return nil
}(); err != nil {
influxql.Iterators(itrs).Close()
return nil, err
}
return itrs, nil
}
// ReadAll reads all points from all iterators.
func (itrs Iterators) ReadAll() ([][]influxql.Point, error) {
var a [][]influxql.Point
// Read from every iterator until a nil is encountered.
for {
points, err := itrs.Next()
if err != nil {
return nil, err
} else if points == nil {
break
}
a = append(a, influxql.Points(points).Clone())
}
// Close all iterators.
influxql.Iterators(itrs).Close()
return a, nil
}
// CreateIterator returns a single combined iterator for the shards.
func (a Shards) CreateIterator(opt influxql.IteratorOptions) (influxql.Iterator, error) {
if influxql.Sources(opt.Sources).HasSystemSource() {
return a.createSystemIterator(opt)
}
// Create iterators for each shard.
// Ensure that they are closed if an error occurs.
itrs := make([]influxql.Iterator, 0, len(a))
if err := func() error {
for _, sh := range a {
itr, err := sh.CreateIterator(opt)
if err != nil {
return err
}
itrs = append(itrs, itr)
}
return nil
}(); err != nil {
influxql.Iterators(itrs).Close()
return nil, err
}
// Merge into a single iterator.
if opt.MergeSorted() {
return influxql.NewSortedMergeIterator(itrs, opt), nil
}
itr := influxql.NewMergeIterator(itrs, opt)
if opt.Expr != nil {
if expr, ok := opt.Expr.(*influxql.Call); ok && expr.Name == "count" {
opt.Expr = &influxql.Call{
Name: "sum",
Args: expr.Args,
}
}
}
return influxql.NewCallIterator(itr, opt), nil
}
// createTagSetIterators creates a set of iterators for a tagset.
func (e *Engine) createTagSetIterators(ref *influxql.VarRef, mm *tsdb.Measurement, t *influxql.TagSet, opt influxql.IteratorOptions) ([]influxql.Iterator, error) {
// Set parallelism by number of logical cpus.
parallelism := runtime.GOMAXPROCS(0)
if parallelism > len(t.SeriesKeys) {
parallelism = len(t.SeriesKeys)
}
// Create series key groupings w/ return error.
groups := make([]struct {
keys []string
filters []influxql.Expr
itrs []influxql.Iterator
err error
}, parallelism)
// Group series keys.
n := len(t.SeriesKeys) / parallelism
for i := 0; i < parallelism; i++ {
group := &groups[i]
if i < parallelism-1 {
group.keys = t.SeriesKeys[i*n : (i+1)*n]
group.filters = t.Filters[i*n : (i+1)*n]
} else {
group.keys = t.SeriesKeys[i*n:]
group.filters = t.Filters[i*n:]
}
group.itrs = make([]influxql.Iterator, 0, len(group.keys))
}
// Read series groups in parallel.
var wg sync.WaitGroup
for i := range groups {
wg.Add(1)
go func(i int) {
defer wg.Done()
groups[i].itrs, groups[i].err = e.createTagSetGroupIterators(ref, mm, groups[i].keys, t, groups[i].filters, opt)
}(i)
}
wg.Wait()
// Determine total number of iterators so we can allocate only once.
var itrN int
for _, group := range groups {
itrN += len(group.itrs)
}
// Combine all iterators together and check for errors.
var err error
itrs := make([]influxql.Iterator, 0, itrN)
for _, group := range groups {
if group.err != nil {
err = group.err
}
itrs = append(itrs, group.itrs...)
}
// If an error occurred, make sure we close all created iterators.
if err != nil {
influxql.Iterators(itrs).Close()
return nil, err
}
return itrs, 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 *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
}
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{InterruptCh: closing}
// 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()
// 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 <-closing:
return influxql.ErrQueryInterrupted
default:
}
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 influxql.ErrQueryInterrupted
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
}
