Modify IndexReader API to accomodate compaction

This changes the IndexReader API to expose plain labels
and chunk meta information instead of a Series interface.
Dropping of irrelevant chunks is moved into the querier.

A LabelIndices method is added to query for existing label
value indices.

db.go

@@ -284,6 +284,9 @@ func (s *Shard) Close() error {
 }
 
 func (s *Shard) appendBatch(samples []hashedSample) error {
+	if len(samples) == 0 {
+		return nil
+	}
 	s.mtx.Lock()
 	defer s.mtx.Unlock()
 
@@ -291,7 +294,7 @@ func (s *Shard) appendBatch(samples []hashedSample) error {
 	// different time blocks. Those may occurr during transition to still
 	// allow late samples to arrive for a previous block.
 	err := s.head.appendBatch(samples)
-	if err != nil {
+	if err == nil {
 		s.metrics.samplesAppended.Add(float64(len(samples)))
 	}
 
@@ -451,9 +454,14 @@ func (es MultiError) Error() string {
 }
 
 // Add adds the error to the error list if it is not nil.
-func (es MultiError) Add(err error) {
-	if err != nil {
-		es = append(es, err)
+func (es *MultiError) Add(err error) {
+	if err == nil {
+		return
+	}
+	if merr, ok := err.(MultiError); ok {
+		*es = append(*es, merr...)
+	} else {
+		*es = append(*es, err)
 	}
 }
 

head.go

@@ -1,6 +1,7 @@
 package tsdb
 
 import (
+	"errors"
 	"os"
 	"sort"
 	"sync"
@@ -112,25 +113,22 @@ func (h *HeadBlock) Postings(name, value string) (Postings, error) {
 }
 
 // Series returns the series for the given reference.
-func (h *HeadBlock) Series(ref uint32, mint, maxt int64) (Series, error) {
+func (h *HeadBlock) Series(ref uint32) (labels.Labels, []ChunkMeta, error) {
 	if int(ref) >= len(h.descs) {
-		return nil, errNotFound
+		return nil, nil, errNotFound
 	}
 	cd := h.descs[ref]
 
-	if !intervalOverlap(cd.firsTimestamp, cd.lastTimestamp, mint, maxt) {
-		return nil, nil
+	return cd.lset, []ChunkMeta{{MinTime: h.stats.MinTime, Ref: ref}}, nil
+}
+
+func (h *HeadBlock) LabelIndices() ([][]string, error) {
+	res := [][]string{}
+
+	for s := range h.values {
+		res = append(res, []string{s})
 	}
-	s := &chunkSeries{
-		labels: cd.lset,
-		chunks: []ChunkMeta{
-			{MinTime: h.stats.MinTime, Ref: 0},
-		},
-		chunk: func(ref uint32) (chunks.Chunk, error) {
-			return cd.chunk, nil
-		},
-	}
-	return s, nil
+	return res, nil
 }
 
 // get retrieves the chunk with the hash and label set and creates
@@ -186,6 +184,11 @@ func (h *HeadBlock) create(hash uint64, lset labels.Labels) *chunkDesc {
 	return cd
 }
 
+var (
+	ErrOutOfOrderSample = errors.New("out of order sample")
+	ErrAmendSample      = errors.New("amending sample")
+)
+
 func (h *HeadBlock) appendBatch(samples []hashedSample) error {
 	// Find head chunks for all samples and allocate new IDs/refs for
 	// ones we haven't seen before.
@@ -200,6 +203,13 @@ func (h *HeadBlock) appendBatch(samples []hashedSample) error {
 
 		cd := h.get(s.hash, s.labels)
 		if cd != nil {
+			// Samples must only occur in order.
+			if s.t < cd.lastTimestamp {
+				return ErrOutOfOrderSample
+			}
+			if cd.lastTimestamp == s.t && cd.lastValue != s.v {
+				return ErrAmendSample
+			}
 			// TODO(fabxc): sample refs are only scoped within a block for
 			// now and we ignore any previously set value
 			s.ref = cd.ref
@@ -232,14 +242,17 @@ func (h *HeadBlock) appendBatch(samples []hashedSample) error {
 	}
 
 	for _, s := range samples {
-		h.descs[s.ref].append(s.t, s.v)
-
-		appended.Inc()
-		h.stats.SampleCount++
+		cd := h.descs[s.ref]
+		// Skip duplicate samples.
+		if cd.lastTimestamp == s.t && cd.lastValue != s.v {
+			continue
+		}
+		cd.append(s.t, s.v)
 
 		if s.t > h.stats.MaxTime {
 			h.stats.MaxTime = s.t
 		}
+		h.stats.SampleCount++
 	}
 
 	return nil
@@ -287,5 +300,13 @@ func (h *HeadBlock) persist(p string) (int64, error) {
 		}
 	}
 
-	return iw.Size() + sw.Size(), nil
+	// Everything written successfully, we can remove the WAL.
+	if err := h.wal.Close(); err != nil {
+		return 0, err
+	}
+	if err := os.Remove(h.wal.f.Name()); err != nil {
+		return 0, err
+	}
+
+	return iw.Size() + sw.Size(), err
 }

querier.go

@@ -236,6 +236,7 @@ func (q *blockQuerier) Select(ms ...labels.Matcher) SeriesSet {
 
 	return &blockSeriesSet{
 		index:  q.index,
+		chunks: q.series,
 		it:     Intersect(its...),
 		absent: absent,
 		mint:   q.mint,
@@ -422,6 +423,7 @@ func (s *shardSeriesSet) Next() bool {
 // blockSeriesSet is a set of series from an inverted index query.
 type blockSeriesSet struct {
 	index      IndexReader
+	chunks     SeriesReader
 	it         Postings // postings list referencing series
 	absent     []string // labels that must not be set for result series
 	mint, maxt int64    // considered time range
@@ -434,24 +436,40 @@ func (s *blockSeriesSet) Next() bool {
 	// Step through the postings iterator to find potential series.
 outer:
 	for s.it.Next() {
-		series, err := s.index.Series(s.it.Value(), s.mint, s.maxt)
+		lset, chunks, err := s.index.Series(s.it.Value())
 		if err != nil {
 			s.err = err
 			return false
 		}
-		// Resolving series may return nil if no applicable data for the
-		// time range exists and we can skip to the next series.
-		if series == nil {
-			continue
-		}
+
 		// If a series contains a label that must be absent, it is skipped as well.
 		for _, abs := range s.absent {
-			if series.Labels().Get(abs) != "" {
+			if lset.Get(abs) != "" {
 				continue outer
 			}
 		}
 
-		s.cur = series
+		ser := &chunkSeries{
+			labels: lset,
+			chunks: make([]ChunkMeta, 0, len(chunks)),
+			chunk:  s.chunks.Chunk,
+		}
+		// Only use chunks that fit the time range.
+		for _, c := range chunks {
+			if c.MaxTime < s.mint {
+				continue
+			}
+			if c.MinTime > s.maxt {
+				break
+			}
+			ser.chunks = append(ser.chunks, c)
+		}
+		// If no chunks of the series apply to the time range, skip it.
+		if len(ser.chunks) == 0 {
+			continue
+		}
+
+		s.cur = ser
 		return true
 	}
 	if s.it.Err() != nil {

reader.go

@@ -63,7 +63,10 @@ type IndexReader interface {
 	Postings(name, value string) (Postings, error)
 
 	// Series returns the series for the given reference.
-	Series(ref uint32, mint, maxt int64) (Series, error)
+	Series(ref uint32) (labels.Labels, []ChunkMeta, error)
+
+	// LabelIndices returns the label pairs for which indices exist.
+	LabelIndices() ([][]string, error)
 }
 
 // StringTuples provides access to a sorted list of string tuples.
@@ -178,18 +181,18 @@ func (r *indexReader) section(o uint32) (byte, []byte, error) {
 	return flag, b[:l], nil
 }
 
-func (r *indexReader) lookupSymbol(o uint32) ([]byte, error) {
+func (r *indexReader) lookupSymbol(o uint32) (string, error) {
 	l, n := binary.Uvarint(r.b[o:])
 	if n < 0 {
-		return nil, fmt.Errorf("reading symbol length failed")
+		return "", fmt.Errorf("reading symbol length failed")
 	}
 
 	end := int(o) + n + int(l)
 	if end > len(r.b) {
-		return nil, fmt.Errorf("invalid length")
+		return "", fmt.Errorf("invalid length")
 	}
 
-	return r.b[int(o)+n : end], nil
+	return yoloString(r.b[int(o)+n : end]), nil
 }
 
 func (r *indexReader) Stats() (BlockStats, error) {
@@ -241,56 +244,55 @@ func (r *indexReader) LabelValues(names ...string) (StringTuples, error) {
 	return st, nil
 }
 
-func (r *indexReader) Series(ref uint32, mint, maxt int64) (Series, error) {
+func (r *indexReader) LabelIndices() ([][]string, error) {
+	res := [][]string{}
+
+	for s := range r.labels {
+		res = append(res, strings.Split(s, string(sep)))
+	}
+	return res, nil
+}
+
+func (r *indexReader) Series(ref uint32) (labels.Labels, []ChunkMeta, error) {
 	k, n := binary.Uvarint(r.b[ref:])
 	if n < 1 {
-		return nil, errors.Wrap(errInvalidSize, "number of labels")
+		return nil, nil, errors.Wrap(errInvalidSize, "number of labels")
 	}
 
 	b := r.b[int(ref)+n:]
-	offsets := make([]uint32, 0, 2*k)
-
-	for i := 0; i < 2*int(k); i++ {
-		o, n := binary.Uvarint(b)
-		if n < 1 {
-			return nil, errors.Wrap(errInvalidSize, "symbol offset")
-		}
-		offsets = append(offsets, uint32(o))
-
-		b = b[n:]
-	}
-	// Symbol offests must occur in pairs representing name and value.
-	if len(offsets)&1 != 0 {
-		return nil, errors.New("odd number of symbol references")
-	}
-
-	// TODO(fabxc): Fully materialize series symbols for now. Figure out later if it
-	// makes sense to decode those lazily.
-	// If we use unsafe strings the there'll be no copy overhead.
-	//
-	// The references are expected to be sorted and match the order of
-	// the underlying strings.
 	lbls := make(labels.Labels, 0, k)
 
-	for i := 0; i < len(offsets); i += 2 {
-		n, err := r.lookupSymbol(offsets[i])
-		if err != nil {
-			return nil, errors.Wrap(err, "symbol lookup")
+	for i := 0; i < 2*int(k); i += 2 {
+		o, m := binary.Uvarint(b)
+		if m < 1 {
+			return nil, nil, errors.Wrap(errInvalidSize, "symbol offset")
 		}
-		v, err := r.lookupSymbol(offsets[i+1])
+		n, err := r.lookupSymbol(uint32(o))
 		if err != nil {
-			return nil, errors.Wrap(err, "symbol lookup")
+			return nil, nil, errors.Wrap(err, "symbol lookup")
 		}
+		b = b[m:]
+
+		o, m = binary.Uvarint(b)
+		if m < 1 {
+			return nil, nil, errors.Wrap(errInvalidSize, "symbol offset")
+		}
+		v, err := r.lookupSymbol(uint32(o))
+		if err != nil {
+			return nil, nil, errors.Wrap(err, "symbol lookup")
+		}
+		b = b[m:]
+
 		lbls = append(lbls, labels.Label{
-			Name:  string(n),
-			Value: string(v),
+			Name:  n,
+			Value: v,
 		})
 	}
 
 	// Read the chunks meta data.
 	k, n = binary.Uvarint(b)
 	if n < 1 {
-		return nil, errors.Wrap(errInvalidSize, "number of chunks")
+		return nil, nil, errors.Wrap(errInvalidSize, "number of chunks")
 	}
 
 	b = b[n:]
@@ -299,49 +301,30 @@ func (r *indexReader) Series(ref uint32, mint, maxt int64) (Series, error) {
 	for i := 0; i < int(k); i++ {
 		firstTime, n := binary.Varint(b)
 		if n < 1 {
-			return nil, errors.Wrap(errInvalidSize, "first time")
+			return nil, nil, errors.Wrap(errInvalidSize, "first time")
 		}
 		b = b[n:]
 
-		// Terminate early if we exceeded the queried time range.
-		if firstTime > maxt {
-			break
-		}
-
 		lastTime, n := binary.Varint(b)
 		if n < 1 {
-			return nil, errors.Wrap(errInvalidSize, "last time")
+			return nil, nil, errors.Wrap(errInvalidSize, "last time")
 		}
 		b = b[n:]
 
 		o, n := binary.Uvarint(b)
 		if n < 1 {
-			return nil, errors.Wrap(errInvalidSize, "chunk offset")
+			return nil, nil, errors.Wrap(errInvalidSize, "chunk offset")
 		}
 		b = b[n:]
 
-		// Skip the chunk if it is before the queried time range.
-		if lastTime < mint {
-			continue
-		}
-
 		chunks = append(chunks, ChunkMeta{
 			Ref:     uint32(o),
 			MinTime: firstTime,
 			MaxTime: lastTime,
 		})
 	}
-	// If no chunks applicable to the time range were found, the series
-	// can be skipped.
-	if len(chunks) == 0 {
-		return nil, nil
-	}
 
-	return &chunkSeries{
-		labels: lbls,
-		chunks: chunks,
-		chunk:  r.series.Chunk,
-	}, nil
+	return lbls, chunks, nil
 }
 
 func (r *indexReader) Postings(name, value string) (Postings, error) {
@@ -419,7 +402,7 @@ func (t *stringTuples) Less(i, j int) bool {
 type serializedStringTuples struct {
 	l      int
 	b      []byte
-	lookup func(uint32) ([]byte, error)
+	lookup func(uint32) (string, error)
 }
 
 func (t *serializedStringTuples) Len() int {
@@ -436,11 +419,11 @@ func (t *serializedStringTuples) At(i int) ([]string, error) {
 	for k := 0; k < t.l; k++ {
 		offset := binary.BigEndian.Uint32(t.b[(i+k)*4:])
 
-		b, err := t.lookup(offset)
+		s, err := t.lookup(offset)
 		if err != nil {
 			return nil, errors.Wrap(err, "symbol lookup")
 		}
-		res = append(res, string(b))
+		res = append(res, s)
 	}
 
 	return res, nil