// Open opens and initializes the mapper.
func (m *RawMapper) Open() error {
// Ignore if node has the shard but hasn't written to it yet.
if m.shard == nil {
return nil
}
// Rewrite statement.
stmt, err := m.shard.index.RewriteSelectStatement(m.stmt)
if err != nil {
return err
}
m.stmt = stmt
// Set all time-related parameters on the mapper.
m.qmin, m.qmax = influxql.TimeRangeAsEpochNano(m.stmt.Condition)
// Get a read-only transaction.
tx, err := m.shard.engine.Begin(false)
if err != nil {
return err
}
m.tx = tx
// Collect measurements.
mms := Measurements(m.shard.index.MeasurementsByName(m.stmt.SourceNames()))
m.selectFields = mms.SelectFields(m.stmt)
m.selectTags = mms.SelectTags(m.stmt)
m.whereFields = mms.WhereFields(m.stmt)
// Open cursors for each measurement.
for _, mm := range mms {
if err := m.openMeasurement(mm); err != nil {
return err
}
}
// Remove cursors if there are not SELECT fields.
if len(m.selectFields) == 0 {
m.cursors = nil
}
return nil
}
// Open opens the local mapper.
func (lm *SelectMapper) Open() error {
if lm.remote != nil {
return lm.remote.Open()
}
var err error
// Get a read-only transaction.
tx, err := lm.shard.engine.Begin(false)
if err != nil {
return err
}
lm.tx = tx
if s, ok := lm.stmt.(*influxql.SelectStatement); ok {
stmt, err := lm.rewriteSelectStatement(s)
if err != nil {
return err
}
lm.selectStmt = stmt
lm.rawMode = (s.IsRawQuery && !s.HasDistinct()) || s.IsSimpleDerivative()
} else {
return lm.openMeta()
}
// Set all time-related parameters on the mapper.
lm.queryTMin, lm.queryTMax = influxql.TimeRangeAsEpochNano(lm.selectStmt.Condition)
if !lm.rawMode {
if err := lm.initializeMapFunctions(); err != nil {
return err
}
// For GROUP BY time queries, limit the number of data points returned by the limit and offset
d, err := lm.selectStmt.GroupByInterval()
if err != nil {
return err
}
lm.intervalSize = d.Nanoseconds()
if lm.queryTMin == 0 || lm.intervalSize == 0 {
lm.numIntervals = 1
lm.intervalSize = lm.queryTMax - lm.queryTMin
} else {
intervalTop := lm.queryTMax/lm.intervalSize*lm.intervalSize + lm.intervalSize
intervalBottom := lm.queryTMin / lm.intervalSize * lm.intervalSize
lm.numIntervals = int((intervalTop - intervalBottom) / lm.intervalSize)
}
if lm.selectStmt.Limit > 0 || lm.selectStmt.Offset > 0 {
// ensure that the offset isn't higher than the number of points we'd get
if lm.selectStmt.Offset > lm.numIntervals {
return nil
}
// Take the lesser of either the pre computed number of GROUP BY buckets that
// will be in the result or the limit passed in by the user
if lm.selectStmt.Limit < lm.numIntervals {
lm.numIntervals = lm.selectStmt.Limit
}
}
// If we are exceeding our MaxGroupByPoints error out
if lm.numIntervals > MaxGroupByPoints {
return errors.New("too many points in the group by interval. maybe you forgot to specify a where time clause?")
}
// Ensure that the start time for the results is on the start of the window.
lm.queryTMinWindow = lm.queryTMin
if lm.intervalSize > 0 && lm.numIntervals > 1 {
lm.queryTMinWindow = lm.queryTMinWindow / lm.intervalSize * lm.intervalSize
}
}
selectFields := newStringSet()
selectTags := newStringSet()
whereFields := newStringSet()
// Create the TagSet cursors for the Mapper.
for _, src := range lm.selectStmt.Sources {
mm, ok := src.(*influxql.Measurement)
if !ok {
return fmt.Errorf("invalid source type: %#v", src)
}
m := lm.shard.index.Measurement(mm.Name)
if m == nil {
// This shard have never received data for the measurement. No Mapper
// required.
return nil
}
// Validate that ANY GROUP BY is not a field for thie measurement.
if err := m.ValidateGroupBy(lm.selectStmt); err != nil {
return err
}
// Create tagset cursors and determine various field types within SELECT statement.
tsf, err := createTagSetsAndFields(m, lm.selectStmt)
if err != nil {
return err
}
tagSets := tsf.tagSets
selectFields.add(tsf.selectFields...)
selectTags.add(tsf.selectTags...)
whereFields.add(tsf.whereFields...)
// If we only have tags in our select clause we just return
if len(selectFields) == 0 && len(selectTags) > 0 {
return fmt.Errorf("statement must have at least one field in select clause")
}
// Validate that any GROUP BY is not on a field
if err := m.ValidateGroupBy(lm.selectStmt); err != nil {
return err
}
// SLIMIT and SOFFSET the unique series
if lm.selectStmt.SLimit > 0 || lm.selectStmt.SOffset > 0 {
if lm.selectStmt.SOffset > len(tagSets) {
tagSets = nil
} else {
if lm.selectStmt.SOffset+lm.selectStmt.SLimit > len(tagSets) {
lm.selectStmt.SLimit = len(tagSets) - lm.selectStmt.SOffset
}
tagSets = tagSets[lm.selectStmt.SOffset : lm.selectStmt.SOffset+lm.selectStmt.SLimit]
}
}
// Create all cursors for reading the data from this shard.
for _, t := range tagSets {
cursors := []*seriesCursor{}
for i, key := range t.SeriesKeys {
c := lm.tx.Cursor(key)
if c == nil {
// No data exists for this key.
continue
}
seriesTags := lm.shard.index.TagsForSeries(key)
cm := newSeriesCursor(c, t.Filters[i], seriesTags)
cursors = append(cursors, cm)
}
tsc := newTagSetCursor(m.Name, t.Tags, cursors, lm.shard.FieldCodec(m.Name))
if lm.rawMode {
tsc.pointHeap = newPointHeap()
//Prime the buffers.
for i := 0; i < len(tsc.cursors); i++ {
k, v := tsc.cursors[i].SeekTo(lm.queryTMin)
if k == -1 {
continue
}
p := &pointHeapItem{
timestamp: k,
value: v,
cursor: tsc.cursors[i],
}
heap.Push(tsc.pointHeap, p)
}
}
lm.cursors = append(lm.cursors, tsc)
}
sort.Sort(tagSetCursors(lm.cursors))
}
lm.selectFields = selectFields.list()
lm.selectTags = selectTags.list()
lm.whereFields = whereFields.list()
// If the query does not aggregate, then at least 1 SELECT field should be present.
if lm.rawMode && len(lm.selectFields) == 0 {
// None of the SELECT fields exist in this data. Wipe out all tagset cursors.
lm.cursors = nil
}
return nil
}
// Open opens and initializes the mapper.
func (m *AggregateMapper) Open() error {
// Ignore if node has the shard but hasn't written to it yet.
if m.shard == nil {
return nil
}
// Rewrite statement.
stmt, err := m.shard.index.RewriteSelectStatement(m.stmt)
if err != nil {
return err
}
m.stmt = stmt
// Set all time-related parameters on the mapper.
m.qmin, m.qmax = influxql.TimeRangeAsEpochNano(m.stmt.Condition)
if err := m.initializeMapFunctions(); err != nil {
return err
}
// For GROUP BY time queries, limit the number of data points returned by the limit and offset
d, err := m.stmt.GroupByInterval()
if err != nil {
return err
}
m.intervalSize = d.Nanoseconds()
if m.qmin == 0 || m.intervalSize == 0 {
m.intervalN = 1
m.intervalSize = m.qmax - m.qmin
} else {
intervalTop := m.qmax/m.intervalSize*m.intervalSize + m.intervalSize
intervalBottom := m.qmin / m.intervalSize * m.intervalSize
m.intervalN = int((intervalTop - intervalBottom) / m.intervalSize)
}
if m.stmt.Limit > 0 || m.stmt.Offset > 0 {
// ensure that the offset isn't higher than the number of points we'd get
if m.stmt.Offset > m.intervalN {
return nil
}
// Take the lesser of either the pre computed number of GROUP BY buckets that
// will be in the result or the limit passed in by the user
if m.stmt.Limit < m.intervalN {
m.intervalN = m.stmt.Limit
}
}
// If we are exceeding our MaxGroupByPoints error out
if m.intervalN > MaxGroupByPoints {
return errors.New("too many points in the group by interval. maybe you forgot to specify a where time clause?")
}
// Ensure that the start time for the results is on the start of the window.
m.qminWindow = m.qmin
if m.intervalSize > 0 && m.intervalN > 1 {
m.qminWindow = m.qminWindow / m.intervalSize * m.intervalSize
}
// Get a read-only transaction.
tx, err := m.shard.engine.Begin(false)
if err != nil {
return err
}
m.tx = tx
// Collect measurements.
mms := Measurements(m.shard.index.MeasurementsByName(m.stmt.SourceNames()))
m.selectFields = mms.SelectFields(m.stmt)
m.selectTags = mms.SelectTags(m.stmt)
m.whereFields = mms.WhereFields(m.stmt)
// Open cursors for each measurement.
for _, mm := range mms {
if err := m.openMeasurement(mm); err != nil {
return err
}
}
return nil
}
