func (self *Coordinator) getShardsAndProcessor(querySpec *parser.QuerySpec, writer engine.Processor) ([]*cluster.ShardData, engine.Processor, error) {
shards, err := self.clusterConfiguration.GetShardsForQuery(querySpec)
if err != nil {
return nil, nil, err
}
shouldAggregateLocally := shards.ShouldAggregateLocally(querySpec)
q := querySpec.SelectQuery()
if q == nil {
return shards, writer, nil
}
if !shouldAggregateLocally {
// if we should aggregate in the coordinator (i.e. aggregation
// isn't happening locally at the shard level), create an engine
shardIds := make([]uint32, len(shards))
for i, s := range shards {
shardIds[i] = s.Id()
}
writer, err = engine.NewQueryEngine(writer, q, shardIds)
if err != nil {
log.Error(err)
log.Debug("Coordinator processor chain: %s", engine.ProcessorChain(writer))
}
return shards, writer, err
}
// if we have a query with limit, then create an engine, or we can
// make the passthrough limit aware
writer = engine.NewPassthroughEngineWithLimit(writer, 100, q.Limit)
return shards, writer, nil
}
func (self *Shard) Query(querySpec *parser.QuerySpec, processor engine.Processor) error {
self.closeLock.RLock()
defer self.closeLock.RUnlock()
if self.closed {
return fmt.Errorf("Shard is closed")
}
if querySpec.IsListSeriesQuery() {
return fmt.Errorf("List series queries should never come to the shard")
} else if querySpec.IsDeleteFromSeriesQuery() {
return self.executeDeleteQuery(querySpec, processor)
}
if !self.hasReadAccess(querySpec) {
return errors.New("User does not have access to one or more of the series requested.")
}
switch t := querySpec.SelectQuery().FromClause.Type; t {
case parser.FromClauseArray:
log.Debug("Shard %s: running a regular query", self.db.Path())
return self.executeArrayQuery(querySpec, processor)
case parser.FromClauseMerge, parser.FromClauseInnerJoin:
log.Debug("Shard %s: running a merge query", self.db.Path())
return self.executeMergeQuery(querySpec, processor, t)
default:
panic(fmt.Errorf("Unknown from clause type %s", t))
}
}
func (self *Shard) executeArrayQuery(querySpec *parser.QuerySpec, processor engine.Processor) error {
seriesAndColumns := querySpec.SelectQuery().GetReferencedColumns()
for series, columns := range seriesAndColumns {
if regex, ok := series.GetCompiledRegex(); ok {
seriesNames := self.metaStore.GetSeriesForDatabaseAndRegex(querySpec.Database(), regex)
for _, name := range seriesNames {
if !querySpec.HasReadAccess(name) {
continue
}
err := self.executeQueryForSeries(querySpec, name, columns, processor)
if err != nil {
return err
}
}
} else {
err := self.executeQueryForSeries(querySpec, series.Name, columns, processor)
if err != nil {
return err
}
}
}
return nil
}
func (self *Shard) executeMergeQuery(querySpec *parser.QuerySpec, processor engine.Processor, t parser.FromClauseType) error {
seriesAndColumns := querySpec.SelectQuery().GetReferencedColumns()
iterators := make([]*PointIterator, len(seriesAndColumns))
streams := make([]engine.StreamQuery, len(iterators))
i := 0
var err error
for s, c := range seriesAndColumns {
c, iterators[i], err = self.getPointIteratorForSeries(querySpec, s.Name, c)
if err != nil {
log.Error(err)
return err
}
defer iterators[i].Close()
aliases := querySpec.SelectQuery().GetTableAliases(s.Name)
if len(aliases) > 1 {
return fmt.Errorf("Cannot have the same table joined more than once")
}
streams[i] = PointIteratorStream{
pi: iterators[i],
name: aliases[0],
fields: c,
}
i++
}
h := engine.NewSeriesHeap(querySpec.IsAscending())
merger := engine.NewCME("Shard", streams, h, processor, t == parser.FromClauseMerge)
if _, err := merger.Update(); err != nil {
return err
}
return nil
}
func (self *ShardData) getProcessor(querySpec *parser.QuerySpec, processor engine.Processor) (engine.Processor, error) {
switch qt := querySpec.Query().Type(); qt {
case parser.Delete, parser.DropSeries:
return NilProcessor{}, nil
case parser.Select:
// continue
default:
panic(fmt.Errorf("Unexpected query type: %s", qt))
}
if querySpec.IsSinglePointQuery() {
return engine.NewPassthroughEngine(processor, 1), nil
}
query := querySpec.SelectQuery()
var err error
// We should aggregate at the shard level
if self.ShouldAggregateLocally(querySpec) {
log.Debug("creating a query engine")
processor, err = engine.NewQueryEngine(processor, query, nil)
if err != nil {
return nil, err
}
goto addFilter
}
// we shouldn't limit the queries if they have aggregates and aren't
// aggregated locally, otherwise the aggregation result which happen
// in the coordinator will get partial data and will be incorrect
if query.HasAggregates() {
log.Debug("creating a passthrough engine")
processor = engine.NewPassthroughEngine(processor, 1000)
goto addFilter
}
// This is an optimization so we don't send more data that we should
// over the wire. The coordinator has its own Passthrough which does
// the final limit.
if l := query.Limit; l > 0 {
log.Debug("creating a passthrough engine with limit")
processor = engine.NewPassthroughEngineWithLimit(processor, 1000, query.Limit)
}
addFilter:
if query := querySpec.SelectQuery(); query != nil && query.GetFromClause().Type != parser.FromClauseInnerJoin {
// Joins do their own filtering since we need to get all
// points before filtering. This is due to the fact that some
// where expressions will be difficult to compute before the
// points are joined together, think where clause with
// left.column = 'something' or right.column =
// 'something_else'. We can't filter the individual series
// separately. The filtering happens in merge.go:55
processor = engine.NewFilteringEngine(query, processor)
}
return processor, nil
}
func (self *Coordinator) expandRegex(spec *parser.QuerySpec) {
q := spec.SelectQuery()
if q == nil {
return
}
f := func(r *regexp.Regexp) []string {
return self.clusterConfiguration.MetaStore.GetSeriesForDatabaseAndRegex(spec.Database(), r)
}
parser.RewriteMergeQuery(q, f)
}
func (self *Shard) executeQueryForSeries(querySpec *parser.QuerySpec, name string, columns []string, processor engine.Processor) error {
if querySpec.IsSinglePointQuery() {
log.Debug("Running single query for series %s", name)
return self.executeSinglePointQuery(querySpec, name, columns, processor)
}
var pi *PointIterator
var err error
columns, pi, err = self.getPointIteratorForSeries(querySpec, name, columns)
if err != nil {
return err
}
defer pi.Close()
query := querySpec.SelectQuery()
aliases := query.GetTableAliases(name)
seriesOutgoing := &protocol.Series{Name: protocol.String(name), Fields: columns, Points: make([]*protocol.Point, 0, self.pointBatchSize)}
for pi.Valid() {
p := pi.Point()
seriesOutgoing.Points = append(seriesOutgoing.Points, p)
if len(seriesOutgoing.Points) >= self.pointBatchSize {
ok, err := yieldToProcessor(seriesOutgoing, processor, aliases)
if !ok || err != nil {
log.Debug("Stopping processing.")
if err != nil {
log.Error("Error while processing data: %v", err)
return err
}
return nil
}
seriesOutgoing = &protocol.Series{Name: protocol.String(name), Fields: columns, Points: make([]*protocol.Point, 0, self.pointBatchSize)}
}
pi.Next()
}
if err := pi.Error(); err != nil {
return err
}
//Yield remaining data
if ok, err := yieldToProcessor(seriesOutgoing, processor, aliases); !ok || err != nil {
log.Debug("Stopping processing remaining points...")
if err != nil {
log.Error("Error while processing data: %v", err)
return err
}
}
log.Debug("Finished running query %s", query.GetQueryString())
return nil
}
func (self *Permissions) AuthorizeSelectQuery(user common.User, db string, querySpec *parser.QuerySpec) (ok bool, err common.AuthorizationError) {
// if this isn't a regex query do the permission check here
fromClause := querySpec.SelectQuery().GetFromClause()
for _, n := range fromClause.Names {
if _, ok := n.Name.GetCompiledRegex(); ok {
break
} else if name := n.Name.Name; !user.HasReadAccess(name) {
return false, common.NewAuthorizationError("User doesn't have read access to %s", name)
}
}
return true, ""
}
func (self *Shard) executeSinglePointQuery(querySpec *parser.QuerySpec, name string, columns []string, p engine.Processor) error {
fields, err := self.getFieldsForSeries(querySpec.Database(), name, columns)
if err != nil {
log.Error("Error looking up fields for %s: %s", name, err)
return err
}
query := querySpec.SelectQuery()
fieldCount := len(fields)
fieldNames := make([]string, 0, fieldCount)
point := &protocol.Point{Values: make([]*protocol.FieldValue, 0, fieldCount)}
timestamp := common.TimeToMicroseconds(query.GetStartTime())
sequenceNumber, err := query.GetSinglePointQuerySequenceNumber()
if err != nil {
return err
}
// set the timestamp and sequence number
point.SequenceNumber = &sequenceNumber
point.SetTimestampInMicroseconds(timestamp)
for _, field := range fields {
sk := newStorageKey(field.Id, timestamp, sequenceNumber)
data, err := self.db.Get(sk.bytes())
if err != nil {
return err
}
if data == nil {
continue
}
fieldValue := &protocol.FieldValue{}
err = proto.Unmarshal(data, fieldValue)
if err != nil {
return err
}
fieldNames = append(fieldNames, field.Name)
point.Values = append(point.Values, fieldValue)
}
result := &protocol.Series{Name: &name, Fields: fieldNames, Points: []*protocol.Point{point}}
if len(result.Points) > 0 {
_, err := p.Yield(result)
return err
}
return nil
}
func (self *Coordinator) shouldQuerySequentially(shards cluster.Shards, querySpec *parser.QuerySpec) bool {
// if the query isn't a select, then it doesn't matter
if querySpec.SelectQuery() == nil {
return false
}
// if the query is a regex, we can't predic the number of responses
// we get back
if querySpec.IsRegex() {
return true
}
groupByClause := querySpec.SelectQuery().GetGroupByClause()
// if there's no group by clause, then we're returning raw points
// with some math done on them, thus we can't predict the number of
// points
if groupByClause == nil {
return true
}
// if there's a group by clause but no group by interval, we can't
// predict the cardinality of the columns used in the group by
// interval, thus we can't predict the number of responses returned
// from the shard
if querySpec.GetGroupByInterval() == nil {
return true
}
// if there's a group by time and other columns, then the previous
// logic holds
if len(groupByClause.Elems) > 1 {
return true
}
if !shards.ShouldAggregateLocally(querySpec) {
return true
}
for _, shard := range shards {
bufferSize := shard.QueryResponseBufferSize(querySpec, self.config.StoragePointBatchSize)
// if the number of repsonses is too big, do a sequential querying
if bufferSize > self.config.ClusterMaxResponseBufferSize {
return true
}
}
// parallel querying only if we're querying a single series, with
// group by time only
return false
}
func (self *Shard) getPointIteratorForSeries(querySpec *parser.QuerySpec, name string, columns []string) ([]string, *PointIterator, error) {
fields, err := self.getFieldsForSeries(querySpec.Database(), name, columns)
if err != nil {
log.Error("Error looking up fields for %s: %s", name, err)
return nil, nil, err
}
startTime := querySpec.GetStartTime()
endTime := querySpec.GetEndTime()
query := querySpec.SelectQuery()
iterators := self.getIterators(fields, startTime, endTime, query.Ascending)
pi := NewPointIterator(iterators, fields, querySpec.GetStartTime(), querySpec.GetEndTime(), query.Ascending)
columns = make([]string, len(fields))
for i := range fields {
columns[i] = fields[i].Name
}
return columns, pi, nil
}
