func NewTopOrBottomAggregator(name string, v *parser.Value, isTop bool, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) != 2 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, fmt.Sprintf("function %s() requires at exactly 2 arguments", name))
}
if v.Elems[1].Type != parser.ValueInt {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("function %s() second parameter expect int", name))
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
limit, err := strconv.ParseInt(v.Elems[1].Name, 10, 64)
if err != nil {
return nil, err
}
if limit < 1 {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("function %s() second parameter must be > 0", name))
}
return &TopOrBottomAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
name: name,
isTop: isTop,
defaultValue: wrappedDefaultValue,
alias: v.Alias,
limit: limit}, nil
}
func NewCountAggregator(q *parser.SelectQuery, v *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function count() requires exactly one argument")
}
if v.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.InvalidArgument, "function count() doesn't work with wildcards")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
if v.Elems[0].Type != parser.ValueSimpleName {
innerName := strings.ToLower(v.Elems[0].Name)
init := registeredAggregators[innerName]
if init == nil {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("Unknown function %s", innerName))
}
inner, err := init(q, v.Elems[0], defaultValue)
if err != nil {
return nil, err
}
return NewCompositeAggregator(&CountAggregator{AbstractAggregator{}, wrappedDefaultValue, v.Alias}, inner)
}
return &CountAggregator{AbstractAggregator{}, wrappedDefaultValue, v.Alias}, nil
}
func NewPercentileAggregator(_ *parser.SelectQuery, value *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(value.Elems) != 2 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function percentile() requires exactly two arguments")
}
if value.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "wildcard cannot be used with percentile")
}
percentile, err := strconv.ParseFloat(value.Elems[1].Name, 64)
if err != nil || percentile <= 0 || percentile >= 100 {
return nil, common.NewQueryError(common.InvalidArgument, "function percentile() requires a numeric second argument between 0 and 100")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
functionName := "percentile"
if value.Alias != "" {
functionName = value.Alias
}
return &PercentileAggregator{
AbstractAggregator: AbstractAggregator{
value: value.Elems[0],
},
functionName: functionName,
percentile: percentile,
defaultValue: wrappedDefaultValue,
}, nil
}
func NewDerivativeAggregator(q *parser.SelectQuery, v *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function derivative() requires exactly one argument")
}
if v.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.InvalidArgument, "function derivative() doesn't work with wildcards")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
da := &DerivativeAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
defaultValue: wrappedDefaultValue,
alias: v.Alias,
}
da.duration, _, err = q.GetGroupByClause().GetGroupByTime()
if err != nil {
return nil, err
}
return da, nil
}
func (self *QueryEngine) executeCountQueryWithGroupBy(query *parser.SelectQuery, yield func(*protocol.Series) error) error {
self.aggregateYield = yield
duration, err := query.GetGroupByClause().GetGroupByTime()
if err != nil {
return err
}
self.isAggregateQuery = true
self.duration = duration
self.aggregators = []Aggregator{}
for _, value := range query.GetColumnNames() {
if !value.IsFunctionCall() {
continue
}
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, query.GetGroupByClause().FillValue)
if err != nil {
return common.NewQueryError(common.InvalidArgument, fmt.Sprintf("%s", err))
}
self.aggregators = append(self.aggregators, aggregator)
}
for _, elem := range query.GetGroupByClause().Elems {
if elem.IsFunctionCall() {
continue
}
self.elems = append(self.elems, elem)
}
self.fillWithZero = query.GetGroupByClause().FillWithZero
// This is a special case for issue #426. If the start time is
// specified and there's a group by clause and fill with zero, then
// we need to fill the entire range from start time to end time
if query.IsStartTimeSpecified() && self.duration != nil && self.fillWithZero {
self.startTimeSpecified = true
self.startTime = query.GetStartTime().Truncate(*self.duration).UnixNano() / 1000
self.endTime = query.GetEndTime().Truncate(*self.duration).UnixNano() / 1000
}
self.initializeFields()
err = self.distributeQuery(query, func(series *protocol.Series) error {
if len(series.Points) == 0 {
return nil
}
return self.aggregateValuesForSeries(series)
})
return err
}
func NewAggregatorEngine(query *parser.SelectQuery, next Processor) (*AggregatorEngine, error) {
ae := &AggregatorEngine{
next: next,
seriesStates: make(map[string]*SeriesState),
ascending: query.Ascending,
}
var err error
ae.duration, ae.irregularInterval, err = query.GetGroupByClause().GetGroupByTime()
if err != nil {
return nil, err
}
ae.aggregators = []Aggregator{}
for _, value := range query.GetColumnNames() {
if !value.IsFunctionCall() {
continue
}
lowerCaseName := strings.ToLower(value.Name)
initializer := registeredAggregators[lowerCaseName]
if initializer == nil {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("Unknown function %s", value.Name))
}
aggregator, err := initializer(query, value, query.GetGroupByClause().FillValue)
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("%s", err))
}
ae.aggregators = append(ae.aggregators, aggregator)
}
for _, elem := range query.GetGroupByClause().Elems {
if elem.IsFunctionCall() {
continue
}
ae.elems = append(ae.elems, elem)
}
ae.isFillQuery = query.GetGroupByClause().FillWithZero
// This is a special case for issue #426. If the start time is
// specified and there's a group by clause and fill with zero, then
// we need to fill the entire range from start time to end time
if query.IsStartTimeSpecified() && ae.duration != nil && ae.isFillQuery {
ae.startTimeSpecified = true
ae.startTime = query.GetStartTime().Truncate(*ae.duration).UnixNano() / 1000
ae.endTime = query.GetEndTime().Truncate(*ae.duration).UnixNano() / 1000
}
ae.initializeFields()
return ae, nil
}
func (self *QueryEngine) executeCountQueryWithGroupBy(query *parser.SelectQuery, yield func(*protocol.Series) error) error {
self.aggregateYield = yield
duration, err := query.GetGroupByClause().GetGroupByTime()
if err != nil {
return err
}
self.isAggregateQuery = true
self.duration = duration
self.aggregators = []Aggregator{}
for _, value := range query.GetColumnNames() {
if !value.IsFunctionCall() {
continue
}
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, query.GetGroupByClause().FillValue)
if err != nil {
return common.NewQueryError(common.InvalidArgument, fmt.Sprintf("%s", err))
}
self.aggregators = append(self.aggregators, aggregator)
}
for _, elem := range query.GetGroupByClause().Elems {
if elem.IsFunctionCall() {
continue
}
self.elems = append(self.elems, elem)
}
self.fillWithZero = query.GetGroupByClause().FillWithZero
self.initializeFields()
err = self.distributeQuery(query, func(series *protocol.Series) error {
if len(series.Points) == 0 {
return nil
}
return self.aggregateValuesForSeries(series)
})
return err
}
// Process responses from the given channel. Returns true if
// processing should stop for other channels. False otherwise.
func (p *MergeChannelProcessor) processChannel(channel <-chan *protocol.Response) bool {
for response := range channel {
log4go.Debug("%s received %s", p, response)
switch rt := response.GetType(); rt {
// all these types end the stream
case protocol.Response_HEARTBEAT,
protocol.Response_END_STREAM:
p.e <- nil
return false
case protocol.Response_ERROR:
err := common.NewQueryError(common.InvalidArgument, response.GetErrorMessage())
p.e <- err
return false
case protocol.Response_QUERY:
for _, s := range response.MultiSeries {
log4go.Debug("Yielding to %s: %s", p.next.Name(), s)
_, err := p.next.Yield(s)
if err != nil {
p.e <- err
return true
}
}
default:
panic(fmt.Errorf("Unknown response type: %s", rt))
}
}
panic(errors.New("Reached end of method"))
}
func NewMedianAggregator(_ *parser.SelectQuery, value *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(value.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function median() requires exactly one argument")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
functionName := "median"
if value.Alias != "" {
functionName = value.Alias
}
aggregator := &PercentileAggregator{
AbstractAggregator: AbstractAggregator{
value: value.Elems[0],
},
functionName: functionName,
percentile: 50.0,
defaultValue: wrappedDefaultValue,
alias: value.Alias,
}
return aggregator, nil
}
func NewModeAggregator(_ *parser.SelectQuery, value *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(value.Elems) < 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function mode() requires at least one argument")
}
// TODO: Mode can in fact take two argument, the second specifies
// the "size", but it's not clear if size is set to 2 whether to
// return at least 2 elements, or return the most common values and
// the second most common values. The difference will be apparent if
// the data set is multimodel and there are two most common
// values. In the first case, the two most common values will be
// returned, but in the second case the two most common values and
// the second most common values will be returned
if len(value.Elems) > 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function mode() takes at most one arguments")
}
size := 1
if len(value.Elems) == 2 {
switch value.Elems[1].Type {
case parser.ValueInt:
var err error
size, err = strconv.Atoi(value.Elems[1].Name)
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into an int", value.Elems[1].Name)
}
default:
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a int", value.Elems[1].Name)
}
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
return &ModeAggregator{
AbstractAggregator: AbstractAggregator{
value: value.Elems[0],
},
defaultValue: wrappedDefaultValue,
alias: value.Alias,
size: size,
}, nil
}
func HashPassword(password string) ([]byte, error) {
if length := len(password); length < 4 || length > 56 {
return nil, common.NewQueryError(common.InvalidArgument, "Password must be more than 4 and less than 56 characters")
}
// The second arg is the cost of the hashing, higher is slower but makes it harder
// to brute force, since it will be really slow and impractical
return bcrypt.GenerateFromPassword([]byte(password), 10)
}
func (self *CoordinatorImpl) runQuerySpec(querySpec *parser.QuerySpec, seriesWriter SeriesWriter) error {
shards, processor, seriesClosed, err := self.getShardsAndProcessor(querySpec, seriesWriter)
if err != nil {
return err
}
if len(shards) == 0 {
return fmt.Errorf("Couldn't look up columns")
}
defer func() {
if processor != nil {
processor.Close()
<-seriesClosed
} else {
seriesWriter.Close()
}
}()
shardConcurrentLimit := self.config.ConcurrentShardQueryLimit
if self.shouldQuerySequentially(shards, querySpec) {
log.Debug("Querying shards sequentially")
shardConcurrentLimit = 1
}
log.Debug("Shard concurrent limit: %d", shardConcurrentLimit)
errors := make(chan error, shardConcurrentLimit)
for i := 0; i < shardConcurrentLimit; i++ {
errors <- nil
}
responseChannels := make(chan (<-chan *protocol.Response), shardConcurrentLimit)
go self.readFromResponseChannels(processor, seriesWriter, querySpec.IsExplainQuery(), errors, responseChannels)
err = self.queryShards(querySpec, shards, errors, responseChannels)
// make sure we read the rest of the errors and responses
for _err := range errors {
if err == nil {
err = _err
}
}
for responsechan := range responseChannels {
for response := range responsechan {
if response.GetType() != endStreamResponse {
continue
}
if response.ErrorMessage != nil && err == nil {
err = common.NewQueryError(common.InvalidArgument, *response.ErrorMessage)
}
break
}
}
return err
}
func NewHistogramAggregator(q *parser.SelectQuery, v *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) < 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function histogram() requires at least one arguments")
}
if len(v.Elems) > 2 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function histogram() takes at most two arguments")
}
if v.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.InvalidArgument, "function histogram() doesn't work with wildcards")
}
bucketSize := 1.0
if len(v.Elems) == 2 {
switch v.Elems[1].Type {
case parser.ValueInt, parser.ValueFloat:
var err error
bucketSize, err = strconv.ParseFloat(v.Elems[1].Name, 64)
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[1].Name)
}
default:
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[1].Name)
}
}
columnNames := []string{"bucket_start", "count"}
if v.Alias != "" {
columnNames[0] = fmt.Sprintf("%s_bucket_start", v.Alias)
columnNames[1] = fmt.Sprintf("%s_count", v.Alias)
}
return &HistogramAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
bucketSize: bucketSize,
columnNames: columnNames,
}, nil
}
func (self *CoordinatorImpl) readFromResponseChannels(processor cluster.QueryProcessor,
writer SeriesWriter,
isExplainQuery bool,
errors chan<- error,
responseChannels <-chan (<-chan *protocol.Response)) {
defer close(errors)
for responseChan := range responseChannels {
for response := range responseChan {
//log.Debug("GOT RESPONSE: ", response.Type, response.Series)
log.Debug("GOT RESPONSE: %v", response.Type)
if *response.Type == endStreamResponse || *response.Type == accessDeniedResponse {
if response.ErrorMessage == nil {
break
}
err := common.NewQueryError(common.InvalidArgument, *response.ErrorMessage)
log.Error("Error while executing query: %s", err)
errors <- err
return
}
if response.Series == nil || len(response.Series.Points) == 0 {
log.Debug("Series has no points, continue")
continue
}
// if we don't have a processor, yield the point to the writer
// this happens if shard took care of the query
// otherwise client will get points from passthrough engine
if processor != nil {
// if the data wasn't aggregated at the shard level, aggregate
// the data here
log.Debug("YIELDING: %d points with %d columns for %s", len(response.Series.Points), len(response.Series.Fields), response.Series.GetName())
processor.YieldSeries(response.Series)
continue
}
// If we have EXPLAIN query, we don't write actual points (of
// response.Type Query) to the client
if !(*response.Type == queryResponse && isExplainQuery) {
writer.Write(response.Series)
}
}
// once we're done with a response channel signal queryShards to
// start querying a new shard
errors <- nil
}
return
}
func NewStandardDeviationAggregator(q *parser.SelectQuery, v *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function stddev() requires exactly one argument")
}
if v.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.InvalidArgument, "function stddev() doesn't work with wildcards")
}
value, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
return &StandardDeviationAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
defaultValue: value,
alias: v.Alias,
}, nil
}
func (self GroupByClause) GetGroupByTime() (*time.Duration, error) {
for _, groupBy := range self.Elems {
if groupBy.IsFunctionCall() && strings.ToLower(groupBy.Name) == "time" {
// TODO: check the number of arguments and return an error
if len(groupBy.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "time function only accepts one argument")
}
if groupBy.Elems[0].Type != ValueDuration {
log.Debug("Get a time function without a duration argument %v", groupBy.Elems[0].Type)
}
arg := groupBy.Elems[0].Name
durationInt, err := common.ParseTimeDuration(arg)
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, fmt.Sprintf("invalid argument %s to the time function", arg))
}
duration := time.Duration(durationInt)
return &duration, nil
}
}
return nil, nil
}
func NewMeanAggregator(_ *parser.SelectQuery, value *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(value.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function mean() requires exactly one argument")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
return &MeanAggregator{
AbstractAggregator: AbstractAggregator{
value: value.Elems[0],
},
defaultValue: wrappedDefaultValue,
alias: value.Alias,
}, nil
}
func NewFirstOrLastAggregator(name string, v *parser.Value, isFirst bool, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, fmt.Sprintf("function %s() requires exactly one argument", name))
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
if v.Alias != "" {
name = v.Alias
}
return &FirstOrLastAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
name: name,
isFirst: isFirst,
defaultValue: wrappedDefaultValue,
}, nil
}
func NewCumulativeArithmeticAggregator(name string, value *parser.Value, initialValue float64, defaultValue *parser.Value, operation Operation) (Aggregator, error) {
if len(value.Elems) != 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function max() requires only one argument")
}
wrappedDefaultValue, err := wrapDefaultValue(defaultValue)
if err != nil {
return nil, err
}
if value.Alias != "" {
name = value.Alias
}
return &CumulativeArithmeticAggregator{
AbstractAggregator: AbstractAggregator{
value: value.Elems[0],
},
name: name,
operation: operation,
initialValue: initialValue,
defaultValue: wrappedDefaultValue,
}, nil
}
func NewHistogramAggregator(q *parser.SelectQuery, v *parser.Value, defaultValue *parser.Value) (Aggregator, error) {
if len(v.Elems) < 1 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function histogram() requires at least one arguments")
}
if len(v.Elems) > 4 {
return nil, common.NewQueryError(common.WrongNumberOfArguments, "function histogram() takes at most four arguments")
}
if v.Elems[0].Type == parser.ValueWildcard {
return nil, common.NewQueryError(common.InvalidArgument, "function histogram() doesn't work with wildcards")
}
bucketSize := 1.0
bucketStart := 0.0
explicitBucketStart := false
bucketStop := 0.0
bucketStopIdx := -1
if len(v.Elems) > 1 {
switch v.Elems[1].Type {
case parser.ValueInt, parser.ValueFloat:
var err error
bucketSize, err = strconv.ParseFloat(v.Elems[1].Name, 64)
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[1].Name)
}
default:
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[1].Name)
}
if len(v.Elems) > 2 {
switch v.Elems[2].Type {
case parser.ValueInt, parser.ValueFloat:
var err error
bucketStart, err = strconv.ParseFloat(v.Elems[2].Name, 64)
explicitBucketStart = true
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[2].Name)
}
default:
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[2].Name)
}
if len(v.Elems) == 4 {
switch v.Elems[3].Type {
case parser.ValueInt, parser.ValueFloat:
var err error
bucketStop, err = strconv.ParseFloat(v.Elems[3].Name, 64)
bucketStopIdx = int(math.Floor((bucketStop - bucketStart) / bucketSize))
if err != nil {
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[3].Name)
}
default:
return nil, common.NewQueryError(common.InvalidArgument, "Cannot parse %s into a float", v.Elems[3].Name)
}
}
}
}
columnNames := []string{"bucket_start", "count"}
if v.Alias != "" {
columnNames[0] = fmt.Sprintf("%s_bucket_start", v.Alias)
columnNames[1] = fmt.Sprintf("%s_count", v.Alias)
}
return &HistogramAggregator{
AbstractAggregator: AbstractAggregator{
value: v.Elems[0],
},
bucketSize: bucketSize,
bucketStart: bucketStart,
explicitBucketStart: explicitBucketStart,
bucketStopIdx: bucketStopIdx,
columnNames: columnNames,
}, nil
}