func (self *LevelDbDatastore) ExecuteQuery(user common.User, database string, query *parser.Query, yield func(*protocol.Series) error) error {
seriesAndColumns := query.GetReferencedColumns()
hasAccess := true
for series, columns := range seriesAndColumns {
if regex, ok := series.GetCompiledRegex(); ok {
seriesNames := self.getSeriesForDbAndRegex(database, regex)
for _, name := range seriesNames {
if !user.HasReadAccess(name) {
hasAccess = false
continue
}
err := self.executeQueryForSeries(database, name, columns, query, yield)
if err != nil {
return err
}
}
} else {
if !user.HasReadAccess(series.Name) {
hasAccess = false
continue
}
err := self.executeQueryForSeries(database, series.Name, columns, query, yield)
if err != nil {
return err
}
}
}
if !hasAccess {
return fmt.Errorf("You don't have permission to access one or more time series")
}
return nil
}
// Return the number of dropped ticks from filtering. if the series
// had more than one alias, returns the min of all dropped ticks
func (self *LevelDbDatastore) sendBatch(query *parser.Query, series *protocol.Series, yield func(series *protocol.Series) error) (int, error) {
dropped := int(math.MaxInt32)
for _, alias := range query.GetTableAliases(*series.Name) {
_alias := alias
newSeries := &protocol.Series{Name: &_alias, Points: series.Points, Fields: series.Fields}
lengthBeforeFiltering := len(newSeries.Points)
var filteredResult *protocol.Series
if query.GetFromClause().Type == parser.FromClauseInnerJoin {
filteredResult = newSeries
} else {
filteredResult, _ = Filter(query, newSeries)
}
_dropped := lengthBeforeFiltering - len(filteredResult.Points)
if _dropped < dropped {
dropped = _dropped
}
if err := yield(filteredResult); err != nil {
return 0, err
}
}
return dropped, nil
}
func isAggregateQuery(query *parser.Query) bool {
for _, column := range query.GetColumnNames() {
if column.IsFunctionCall() {
return true
}
}
return false
}
func getJoinYield(query *parser.Query, yield func(*protocol.Series) error) func(*protocol.Series) error {
var lastPoint1 *protocol.Point
var lastFields1 []string
var lastPoint2 *protocol.Point
var lastFields2 []string
table1 := query.GetFromClause().Names[0].GetAlias()
table2 := query.GetFromClause().Names[1].GetAlias()
name := table1 + "_join_" + table2
return mergeYield(table1, table2, false, query.Ascending, func(s *protocol.Series) error {
if *s.Name == table1 {
lastPoint1 = s.Points[len(s.Points)-1]
if lastFields1 == nil {
for _, f := range s.Fields {
lastFields1 = append(lastFields1, table1+"."+f)
}
}
}
if *s.Name == table2 {
lastPoint2 = s.Points[len(s.Points)-1]
if lastFields2 == nil {
for _, f := range s.Fields {
lastFields2 = append(lastFields2, table2+"."+f)
}
}
}
if lastPoint1 == nil || lastPoint2 == nil {
return nil
}
newSeries := &protocol.Series{
Name: &name,
Fields: append(lastFields1, lastFields2...),
Points: []*protocol.Point{
&protocol.Point{
Values: append(lastPoint1.Values, lastPoint2.Values...),
Timestamp: lastPoint2.Timestamp,
SequenceNumber: lastPoint2.SequenceNumber,
},
},
}
lastPoint1 = nil
lastPoint2 = nil
filteredSeries, _ := datastore.Filter(query, newSeries)
if len(filteredSeries.Points) > 0 {
return yield(newSeries)
}
return nil
})
}
// distribute query and possibly do the merge/join before yielding the points
func (self *QueryEngine) distributeQuery(user common.User, database string, query *parser.Query, yield func(*protocol.Series) error) (err error) {
// see if this is a merge query
fromClause := query.GetFromClause()
if fromClause.Type == parser.FromClauseMerge {
yield = getMergeYield(fromClause.Names[0].Name.Name, fromClause.Names[1].Name.Name, query.Ascending, yield)
}
if fromClause.Type == parser.FromClauseInnerJoin {
yield = getJoinYield(query, yield)
}
return self.coordinator.DistributeQuery(user, database, query, yield)
}
func NewTimestampAggregator(query *parser.Query, _ *parser.Value) (Aggregator, error) {
duration, err := query.GetGroupByClause().GetGroupByTime()
if err != nil {
return nil, err
}
var durationPtr *int64
if duration != nil {
newDuration := int64(*duration / time.Microsecond)
durationPtr = &newDuration
}
return &TimestampAggregator{
timestamps: make(map[string]map[interface{}]int64),
duration: durationPtr,
}, nil
}
func Filter(query *parser.Query, series *protocol.Series) (*protocol.Series, error) {
if query.GetWhereCondition() == nil {
return series, nil
}
columns := map[string]bool{}
getColumns(query.GetColumnNames(), columns)
points := series.Points
series.Points = nil
for _, point := range points {
ok, err := matches(query.GetWhereCondition(), series.Fields, point)
if err != nil {
return nil, err
}
if ok {
filterColumns(columns, series.Fields, point)
series.Points = append(series.Points, point)
}
}
if !columns["*"] {
newFields := []string{}
for _, f := range series.Fields {
if _, ok := columns[f]; !ok {
continue
}
newFields = append(newFields, f)
}
series.Fields = newFields
}
return series, nil
}
func (self *LevelDbDatastore) executeQueryForSeries(database, series string, columns []string, query *parser.Query, yield func(*protocol.Series) error) error {
startTimeBytes, endTimeBytes := self.byteArraysForStartAndEndTimes(common.TimeToMicroseconds(query.GetStartTime()), common.TimeToMicroseconds(query.GetEndTime()))
fields, err := self.getFieldsForSeries(database, series, columns)
if err != nil {
return err
}
fieldCount := len(fields)
prefixes := make([][]byte, fieldCount, fieldCount)
iterators := make([]*levigo.Iterator, fieldCount, fieldCount)
fieldNames := make([]string, len(fields))
// start the iterators to go through the series data
for i, field := range fields {
fieldNames[i] = field.Name
prefixes[i] = field.Id
iterators[i] = self.db.NewIterator(self.readOptions)
if query.Ascending {
iterators[i].Seek(append(field.Id, startTimeBytes...))
} else {
iterators[i].Seek(append(append(field.Id, endTimeBytes...), MAX_SEQUENCE...))
if iterators[i].Valid() {
iterators[i].Prev()
}
}
}
result := &protocol.Series{Name: &series, Fields: fieldNames, Points: make([]*protocol.Point, 0)}
rawColumnValues := make([]*rawColumnValue, fieldCount, fieldCount)
isValid := true
limit := query.Limit
if limit == 0 {
limit = MAX_POINTS_TO_SCAN
}
resultByteCount := 0
// TODO: clean up, this is super gnarly
// optimize for the case where we're pulling back only a single column or aggregate
for isValid {
isValid = false
latestTimeRaw := make([]byte, 8, 8)
latestSequenceRaw := make([]byte, 8, 8)
point := &protocol.Point{Values: make([]*protocol.FieldValue, fieldCount, fieldCount)}
for i, it := range iterators {
if rawColumnValues[i] == nil && it.Valid() {
k := it.Key()
if len(k) >= 16 {
t := k[8:16]
if bytes.Equal(k[:8], fields[i].Id) && bytes.Compare(t, startTimeBytes) > -1 && bytes.Compare(t, endTimeBytes) < 1 {
v := it.Value()
s := k[16:]
rawColumnValues[i] = &rawColumnValue{time: t, sequence: s, value: v}
timeCompare := bytes.Compare(t, latestTimeRaw)
if timeCompare == 1 {
latestTimeRaw = t
latestSequenceRaw = s
} else if timeCompare == 0 {
if bytes.Compare(s, latestSequenceRaw) == 1 {
latestSequenceRaw = s
}
}
}
}
}
}
for i, iterator := range iterators {
if rawColumnValues[i] != nil && bytes.Equal(rawColumnValues[i].time, latestTimeRaw) && bytes.Equal(rawColumnValues[i].sequence, latestSequenceRaw) {
isValid = true
if query.Ascending {
iterator.Next()
} else {
iterator.Prev()
}
fv := &protocol.FieldValue{}
err := proto.Unmarshal(rawColumnValues[i].value, fv)
if err != nil {
return err
}
resultByteCount += len(rawColumnValues[i].value)
point.Values[i] = fv
var t uint64
binary.Read(bytes.NewBuffer(rawColumnValues[i].time), binary.BigEndian, &t)
time := self.convertUintTimestampToInt64(&t)
var sequence uint64
binary.Read(bytes.NewBuffer(rawColumnValues[i].sequence), binary.BigEndian, &sequence)
seq32 := uint32(sequence)
point.SetTimestampInMicroseconds(time)
point.SequenceNumber = &seq32
rawColumnValues[i] = nil
}
}
if isValid {
limit -= 1
result.Points = append(result.Points, point)
// add byte count for the timestamp and the sequence
resultByteCount += 16
// check if we should send the batch along
if resultByteCount > MAX_SERIES_SIZE {
filteredResult, _ := Filter(query, result)
if err := yield(filteredResult); err != nil {
return err
}
resultByteCount = 0
result = &protocol.Series{Name: &series, Fields: fieldNames, Points: make([]*protocol.Point, 0)}
}
}
if limit < 1 {
break
}
}
filteredResult, _ := Filter(query, result)
if err := yield(filteredResult); err != nil {
return err
}
emptyResult := &protocol.Series{Name: &series, Fields: fieldNames, Points: nil}
return yield(emptyResult)
}
func (self *LevelDbDatastore) executeQueryForSeries(database, series string, columns []string, query *parser.Query, yield func(*protocol.Series) error) error {
startTimeBytes, endTimeBytes := self.byteArraysForStartAndEndTimes(common.TimeToMicroseconds(query.GetStartTime()), common.TimeToMicroseconds(query.GetEndTime()))
fields, err := self.getFieldsForSeries(database, series, columns)
if err != nil {
return err
}
fieldCount := len(fields)
fieldNames, iterators := self.getIterators(fields, startTimeBytes, endTimeBytes, query.Ascending)
// iterators :=
result := &protocol.Series{Name: &series, Fields: fieldNames, Points: make([]*protocol.Point, 0)}
rawColumnValues := make([]*rawColumnValue, fieldCount, fieldCount)
limit := query.Limit
if limit == 0 {
limit = MAX_POINTS_TO_SCAN
}
resultByteCount := 0
// TODO: clean up, this is super gnarly
// optimize for the case where we're pulling back only a single column or aggregate
for {
isValid := false
point := &protocol.Point{Values: make([]*protocol.FieldValue, fieldCount, fieldCount)}
for i, it := range iterators {
if rawColumnValues[i] != nil || !it.Valid() {
continue
}
key := it.Key()
if len(key) < 16 {
continue
}
if !isPointInRange(fields[i].Id, startTimeBytes, endTimeBytes, key) {
continue
}
time := key[8:16]
value := it.Value()
sequenceNumber := key[16:]
rawValue := &rawColumnValue{time: time, sequence: sequenceNumber, value: value}
rawColumnValues[i] = rawValue
}
var pointTimeRaw []byte
var pointSequenceRaw []byte
// choose the highest (or lowest in case of ascending queries) timestamp
// and sequence number. that will become the timestamp and sequence of
// the next point.
for _, value := range rawColumnValues {
if value == nil {
continue
}
pointTimeRaw, pointSequenceRaw = value.updatePointTimeAndSequence(pointTimeRaw,
pointSequenceRaw, query.Ascending)
}
for i, iterator := range iterators {
// if the value is nil, or doesn't match the point's timestamp and sequence number
// then skip it
if rawColumnValues[i] == nil ||
!bytes.Equal(rawColumnValues[i].time, pointTimeRaw) ||
!bytes.Equal(rawColumnValues[i].sequence, pointSequenceRaw) {
continue
}
// if we emitted at lease one column, then we should keep
// trying to get more points
isValid = true
// advance the iterator to read a new value in the next iteration
if query.Ascending {
iterator.Next()
} else {
iterator.Prev()
}
fv := &protocol.FieldValue{}
err := proto.Unmarshal(rawColumnValues[i].value, fv)
if err != nil {
return err
}
resultByteCount += len(rawColumnValues[i].value)
point.Values[i] = fv
var t uint64
binary.Read(bytes.NewBuffer(rawColumnValues[i].time), binary.BigEndian, &t)
time := self.convertUintTimestampToInt64(&t)
var sequence uint64
binary.Read(bytes.NewBuffer(rawColumnValues[i].sequence), binary.BigEndian, &sequence)
seq32 := uint32(sequence)
point.SetTimestampInMicroseconds(time)
point.SequenceNumber = &seq32
rawColumnValues[i] = nil
}
// stop the loop if we ran out of points
if !isValid {
break
}
limit -= 1
result.Points = append(result.Points, point)
// add byte count for the timestamp and the sequence
resultByteCount += 16
// check if we should send the batch along
if resultByteCount > MAX_SERIES_SIZE || limit < 1 {
dropped, err := self.sendBatch(query, result, yield)
if err != nil {
return err
}
limit += dropped
resultByteCount = 0
result = &protocol.Series{Name: &series, Fields: fieldNames, Points: make([]*protocol.Point, 0)}
}
if limit < 1 {
break
}
}
if _, err := self.sendBatch(query, result, yield); err != nil {
return err
}
emptyResult := &protocol.Series{Name: &series, Fields: fieldNames, Points: nil}
_, err = self.sendBatch(query, emptyResult, yield)
return err
}
func (self *QueryEngine) executeCountQueryWithGroupBy(user common.User, database string, query *parser.Query,
yield func(*protocol.Series) error) error {
duration, err := query.GetGroupByClause().GetGroupByTime()
if err != nil {
return err
}
aggregators := []Aggregator{}
for _, value := range query.GetColumnNames() {
if value.IsFunctionCall() {
lowerCaseName := strings.ToLower(value.Name)
initializer := registeredAggregators[lowerCaseName]
if initializer == nil {
return common.NewQueryError(common.InvalidArgument, fmt.Sprintf("Unknown function %s", value.Name))
}
aggregator, err := initializer(query, value)
if err != nil {
return err
}
aggregators = append(aggregators, aggregator)
}
}
timestampAggregator, err := NewTimestampAggregator(query, nil)
if err != nil {
return err
}
groups := make(map[string]map[interface{}]bool)
groupBy := query.GetGroupByClause()
var inverse InverseMapper
err = self.distributeQuery(user, database, query, func(series *protocol.Series) error {
var mapper Mapper
mapper, inverse, err = createValuesToInterface(groupBy, series.Fields)
if err != nil {
return err
}
for _, aggregator := range aggregators {
if err := aggregator.InitializeFieldsMetadata(series); err != nil {
return err
}
}
for _, point := range series.Points {
value := mapper(point)
for _, aggregator := range aggregators {
aggregator.AggregatePoint(*series.Name, value, point)
}
timestampAggregator.AggregatePoint(*series.Name, value, point)
seriesGroups := groups[*series.Name]
if seriesGroups == nil {
seriesGroups = make(map[interface{}]bool)
groups[*series.Name] = seriesGroups
}
seriesGroups[value] = true
}
return nil
})
if err != nil {
return err
}
var sequenceNumber uint32 = 1
fields := []string{}
for _, aggregator := range aggregators {
columnName := aggregator.ColumnName()
fields = append(fields, columnName)
}
for _, value := range groupBy {
if value.IsFunctionCall() {
continue
}
tempName := value.Name
fields = append(fields, tempName)
}
for table, tableGroups := range groups {
tempTable := table
points := []*protocol.Point{}
for groupId, _ := range tableGroups {
timestamp := *timestampAggregator.GetValue(table, groupId)[0].Int64Value
values := [][]*protocol.FieldValue{}
for _, aggregator := range aggregators {
values = append(values, aggregator.GetValue(table, groupId))
}
// do cross product of all the values
values = crossProduct(values)
for _, v := range values {
/* groupPoints := []*protocol.Point{} */
point := &protocol.Point{
SequenceNumber: &sequenceNumber,
Values: v,
}
point.SetTimestampInMicroseconds(timestamp)
// FIXME: this should be looking at the fields slice not the group by clause
// FIXME: we should check whether the selected columns are in the group by clause
for idx, _ := range groupBy {
if duration != nil && idx == 0 {
continue
}
value := inverse(groupId, idx)
switch x := value.(type) {
case string:
point.Values = append(point.Values, &protocol.FieldValue{StringValue: &x})
case bool:
point.Values = append(point.Values, &protocol.FieldValue{BoolValue: &x})
case float64:
point.Values = append(point.Values, &protocol.FieldValue{DoubleValue: &x})
case int64:
point.Values = append(point.Values, &protocol.FieldValue{Int64Value: &x})
}
}
points = append(points, point)
}
}
expectedData := &protocol.Series{
Name: &tempTable,
Fields: fields,
Points: points,
}
yield(expectedData)
}
return nil
}