vendor: update tsdb

Signed-off-by: Goutham Veeramachaneni <cs14btech11014@iith.ac.in>

vendor/github.com/prometheus/tsdb/block.go

@@ -23,10 +23,101 @@ import (
 
 	"github.com/oklog/ulid"
 	"github.com/pkg/errors"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/chunks"
+	"github.com/prometheus/tsdb/index"
 	"github.com/prometheus/tsdb/labels"
 )
 
+// IndexWriter serializes the index for a block of series data.
+// The methods must be called in the order they are specified in.
+type IndexWriter interface {
+	// AddSymbols registers all string symbols that are encountered in series
+	// and other indices.
+	AddSymbols(sym map[string]struct{}) error
+
+	// AddSeries populates the index writer with a series and its offsets
+	// of chunks that the index can reference.
+	// Implementations may require series to be insert in increasing order by
+	// their labels.
+	// The reference numbers are used to resolve entries in postings lists that
+	// are added later.
+	AddSeries(ref uint64, l labels.Labels, chunks ...chunks.Meta) error
+
+	// WriteLabelIndex serializes an index from label names to values.
+	// The passed in values chained tuples of strings of the length of names.
+	WriteLabelIndex(names []string, values []string) error
+
+	// WritePostings writes a postings list for a single label pair.
+	// The Postings here contain refs to the series that were added.
+	WritePostings(name, value string, it index.Postings) error
+
+	// Close writes any finalization and closes the resources associated with
+	// the underlying writer.
+	Close() error
+}
+
+// IndexReader provides reading access of serialized index data.
+type IndexReader interface {
+	// Symbols returns a set of string symbols that may occur in series' labels
+	// and indices.
+	Symbols() (map[string]struct{}, error)
+
+	// LabelValues returns the possible label values.
+	LabelValues(names ...string) (index.StringTuples, error)
+
+	// Postings returns the postings list iterator for the label pair.
+	// The Postings here contain the offsets to the series inside the index.
+	// Found IDs are not strictly required to point to a valid Series, e.g. during
+	// background garbage collections.
+	Postings(name, value string) (index.Postings, error)
+
+	// SortedPostings returns a postings list that is reordered to be sorted
+	// by the label set of the underlying series.
+	SortedPostings(index.Postings) index.Postings
+
+	// Series populates the given labels and chunk metas for the series identified
+	// by the reference.
+	// Returns ErrNotFound if the ref does not resolve to a known series.
+	Series(ref uint64, lset *labels.Labels, chks *[]chunks.Meta) error
+
+	// LabelIndices returns a list of string tuples for which a label value index exists.
+	LabelIndices() ([][]string, error)
+
+	// Close releases the underlying resources of the reader.
+	Close() error
+}
+
+// StringTuples provides access to a sorted list of string tuples.
+type StringTuples interface {
+	// Total number of tuples in the list.
+	Len() int
+	// At returns the tuple at position i.
+	At(i int) ([]string, error)
+}
+
+// ChunkWriter serializes a time block of chunked series data.
+type ChunkWriter interface {
+	// WriteChunks writes several chunks. The Chunk field of the ChunkMetas
+	// must be populated.
+	// After returning successfully, the Ref fields in the ChunkMetas
+	// are set and can be used to retrieve the chunks from the written data.
+	WriteChunks(chunks ...chunks.Meta) error
+
+	// Close writes any required finalization and closes the resources
+	// associated with the underlying writer.
+	Close() error
+}
+
+// ChunkReader provides reading access of serialized time series data.
+type ChunkReader interface {
+	// Chunk returns the series data chunk with the given reference.
+	Chunk(ref uint64) (chunkenc.Chunk, error)
+
+	// Close releases all underlying resources of the reader.
+	Close() error
+}
+
 // BlockReader provides reading access to a data block.
 type BlockReader interface {
 	// Index returns an IndexReader over the block's data.
@@ -91,8 +182,12 @@ type blockMeta struct {
 	*BlockMeta
 }
 
+const indexFilename = "index"
 const metaFilename = "meta.json"
 
+func chunkDir(dir string) string { return filepath.Join(dir, "chunks") }
+func walDir(dir string) string   { return filepath.Join(dir, "wal") }
+
 func readMetaFile(dir string) (*BlockMeta, error) {
 	b, err := ioutil.ReadFile(filepath.Join(dir, metaFilename))
 	if err != nil {
@@ -150,17 +245,17 @@ type Block struct {
 
 // OpenBlock opens the block in the directory. It can be passed a chunk pool, which is used
 // to instantiate chunk structs.
-func OpenBlock(dir string, pool chunks.Pool) (*Block, error) {
+func OpenBlock(dir string, pool chunkenc.Pool) (*Block, error) {
 	meta, err := readMetaFile(dir)
 	if err != nil {
 		return nil, err
 	}
 
-	cr, err := NewDirChunkReader(chunkDir(dir), pool)
+	cr, err := chunks.NewDirReader(chunkDir(dir), pool)
 	if err != nil {
 		return nil, err
 	}
-	ir, err := NewFileIndexReader(filepath.Join(dir, "index"))
+	ir, err := index.NewFileReader(filepath.Join(dir, "index"))
 	if err != nil {
 		return nil, err
 	}
@@ -289,7 +384,7 @@ func (pb *Block) Delete(mint, maxt int64, ms ...labels.Matcher) error {
 		return ErrClosing
 	}
 
-	p, absent, err := PostingsForMatchers(pb.indexr, ms...)
+	p, err := PostingsForMatchers(pb.indexr, ms...)
 	if err != nil {
 		return errors.Wrap(err, "select series")
 	}
@@ -300,7 +395,7 @@ func (pb *Block) Delete(mint, maxt int64, ms ...labels.Matcher) error {
 	stones := memTombstones{}
 
 	var lset labels.Labels
-	var chks []ChunkMeta
+	var chks []chunks.Meta
 
 Outer:
 	for p.Next() {
@@ -309,12 +404,6 @@ Outer:
 			return err
 		}
 
-		for _, abs := range absent {
-			if lset.Get(abs) != "" {
-				continue Outer
-			}
-		}
-
 		for _, chk := range chks {
 			if intervalOverlap(mint, maxt, chk.MinTime, chk.MaxTime) {
 				// Delete only until the current vlaues and not beyond.
@@ -411,9 +500,6 @@ func (pb *Block) Snapshot(dir string) error {
 	return nil
 }
 
-func chunkDir(dir string) string { return filepath.Join(dir, "chunks") }
-func walDir(dir string) string   { return filepath.Join(dir, "wal") }
-
 func clampInterval(a, b, mint, maxt int64) (int64, int64) {
 	if a < mint {
 		a = mint
@@ -423,36 +509,3 @@ func clampInterval(a, b, mint, maxt int64) (int64, int64) {
 	}
 	return a, b
 }
-
-type mmapFile struct {
-	f *os.File
-	b []byte
-}
-
-func openMmapFile(path string) (*mmapFile, error) {
-	f, err := os.Open(path)
-	if err != nil {
-		return nil, errors.Wrap(err, "try lock file")
-	}
-	info, err := f.Stat()
-	if err != nil {
-		return nil, errors.Wrap(err, "stat")
-	}
-
-	b, err := mmap(f, int(info.Size()))
-	if err != nil {
-		return nil, errors.Wrap(err, "mmap")
-	}
-
-	return &mmapFile{f: f, b: b}, nil
-}
-
-func (f *mmapFile) Close() error {
-	err0 := munmap(f.b)
-	err1 := f.f.Close()
-
-	if err0 != nil {
-		return err0
-	}
-	return err1
-}

vendor/github.com/prometheus/tsdb/chunkenc/bstream.go

@@ -39,7 +39,7 @@
 // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-package chunks
+package chunkenc
 
 import "io"
 

vendor/github.com/prometheus/tsdb/chunkenc/chunk.go

@@ -11,7 +11,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package chunks
+package chunkenc
 
 import (
 	"fmt"

vendor/github.com/prometheus/tsdb/chunkenc/xor.go

@@ -41,7 +41,7 @@
 // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-package chunks
+package chunkenc
 
 import (
 	"encoding/binary"

vendor/github.com/prometheus/tsdb/chunks/chunks.go

@@ -11,18 +11,22 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package tsdb
+package chunks
 
 import (
 	"bufio"
 	"encoding/binary"
 	"fmt"
 	"hash"
+	"hash/crc32"
 	"io"
+	"io/ioutil"
 	"os"
+	"path/filepath"
+	"strconv"
 
 	"github.com/pkg/errors"
-	"github.com/prometheus/tsdb/chunks"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/fileutil"
 )
 
@@ -31,19 +35,19 @@ const (
 	MagicChunks = 0x85BD40DD
 )
 
-// ChunkMeta holds information about a chunk of data.
+// Meta holds information about a chunk of data.
-type ChunkMeta struct {
+type Meta struct {
 	// Ref and Chunk hold either a reference that can be used to retrieve
 	// chunk data or the data itself.
 	// Generally, only one of them is set.
 	Ref   uint64
-	Chunk chunks.Chunk
+	Chunk chunkenc.Chunk
 
 	MinTime, MaxTime int64 // time range the data covers
 }
 
 // writeHash writes the chunk encoding and raw data into the provided hash.
-func (cm *ChunkMeta) writeHash(h hash.Hash) error {
+func (cm *Meta) writeHash(h hash.Hash) error {
 	if _, err := h.Write([]byte{byte(cm.Chunk.Encoding())}); err != nil {
 		return err
 	}
@@ -53,62 +57,27 @@ func (cm *ChunkMeta) writeHash(h hash.Hash) error {
 	return nil
 }
 
-// deletedIterator wraps an Iterator and makes sure any deleted metrics are not
+var (
-// returned.
+	errInvalidSize     = fmt.Errorf("invalid size")
-type deletedIterator struct {
+	errInvalidFlag     = fmt.Errorf("invalid flag")
-	it chunks.Iterator
+	errInvalidChecksum = fmt.Errorf("invalid checksum")
+)
 
-	intervals Intervals
+var castagnoliTable *crc32.Table
+
+func init() {
+	castagnoliTable = crc32.MakeTable(crc32.Castagnoli)
 }
 
-func (it *deletedIterator) At() (int64, float64) {
+// newCRC32 initializes a CRC32 hash with a preconfigured polynomial, so the
-	return it.it.At()
+// polynomial may be easily changed in one location at a later time, if necessary.
+func newCRC32() hash.Hash32 {
+	return crc32.New(castagnoliTable)
 }
 
-func (it *deletedIterator) Next() bool {
+// Writer implements the ChunkWriter interface for the standard
-Outer:
-	for it.it.Next() {
-		ts, _ := it.it.At()
-
-		for _, tr := range it.intervals {
-			if tr.inBounds(ts) {
-				continue Outer
-			}
-
-			if ts > tr.Maxt {
-				it.intervals = it.intervals[1:]
-				continue
-			}
-
-			return true
-		}
-
-		return true
-	}
-
-	return false
-}
-
-func (it *deletedIterator) Err() error {
-	return it.it.Err()
-}
-
-// ChunkWriter serializes a time block of chunked series data.
-type ChunkWriter interface {
-	// WriteChunks writes several chunks. The Chunk field of the ChunkMetas
-	// must be populated.
-	// After returning successfully, the Ref fields in the ChunkMetas
-	// are set and can be used to retrieve the chunks from the written data.
-	WriteChunks(chunks ...ChunkMeta) error
-
-	// Close writes any required finalization and closes the resources
-	// associated with the underlying writer.
-	Close() error
-}
-
-// chunkWriter implements the ChunkWriter interface for the standard
 // serialization format.
-type chunkWriter struct {
+type Writer struct {
 	dirFile *os.File
 	files   []*os.File
 	wbuf    *bufio.Writer
@@ -124,7 +93,8 @@ const (
 	chunksFormatV1 = 1
 )
 
-func newChunkWriter(dir string) (*chunkWriter, error) {
+// NewWriter returns a new writer against the given directory.
+func NewWriter(dir string) (*Writer, error) {
 	if err := os.MkdirAll(dir, 0777); err != nil {
 		return nil, err
 	}
@@ -132,7 +102,7 @@ func newChunkWriter(dir string) (*chunkWriter, error) {
 	if err != nil {
 		return nil, err
 	}
-	cw := &chunkWriter{
+	cw := &Writer{
 		dirFile:     dirFile,
 		n:           0,
 		crc32:       newCRC32(),
@@ -141,7 +111,7 @@ func newChunkWriter(dir string) (*chunkWriter, error) {
 	return cw, nil
 }
 
-func (w *chunkWriter) tail() *os.File {
+func (w *Writer) tail() *os.File {
 	if len(w.files) == 0 {
 		return nil
 	}
@@ -150,7 +120,7 @@ func (w *chunkWriter) tail() *os.File {
 
 // finalizeTail writes all pending data to the current tail file,
 // truncates its size, and closes it.
-func (w *chunkWriter) finalizeTail() error {
+func (w *Writer) finalizeTail() error {
 	tf := w.tail()
 	if tf == nil {
 		return nil
@@ -174,7 +144,7 @@ func (w *chunkWriter) finalizeTail() error {
 	return tf.Close()
 }
 
-func (w *chunkWriter) cut() error {
+func (w *Writer) cut() error {
 	// Sync current tail to disk and close.
 	if err := w.finalizeTail(); err != nil {
 		return err
@@ -216,13 +186,13 @@ func (w *chunkWriter) cut() error {
 	return nil
 }
 
-func (w *chunkWriter) write(b []byte) error {
+func (w *Writer) write(b []byte) error {
 	n, err := w.wbuf.Write(b)
 	w.n += int64(n)
 	return err
 }
 
-func (w *chunkWriter) WriteChunks(chks ...ChunkMeta) error {
+func (w *Writer) WriteChunks(chks ...Meta) error {
 	// Calculate maximum space we need and cut a new segment in case
 	// we don't fit into the current one.
 	maxLen := int64(binary.MaxVarintLen32) // The number of chunks.
@@ -272,11 +242,11 @@ func (w *chunkWriter) WriteChunks(chks ...ChunkMeta) error {
 	return nil
 }
 
-func (w *chunkWriter) seq() int {
+func (w *Writer) seq() int {
 	return len(w.files) - 1
 }
 
-func (w *chunkWriter) Close() error {
+func (w *Writer) Close() error {
 	if err := w.finalizeTail(); err != nil {
 		return err
 	}
@@ -285,29 +255,40 @@ func (w *chunkWriter) Close() error {
 	return w.dirFile.Close()
 }
 
-// ChunkReader provides reading access of serialized time series data.
+// ByteSlice abstracts a byte slice.
-type ChunkReader interface {
+type ByteSlice interface {
-	// Chunk returns the series data chunk with the given reference.
+	Len() int
-	Chunk(ref uint64) (chunks.Chunk, error)
+	Range(start, end int) []byte
-
-	// Close releases all underlying resources of the reader.
-	Close() error
 }
 
-// chunkReader implements a SeriesReader for a serialized byte stream
+type realByteSlice []byte
+
+func (b realByteSlice) Len() int {
+	return len(b)
+}
+
+func (b realByteSlice) Range(start, end int) []byte {
+	return b[start:end]
+}
+
+func (b realByteSlice) Sub(start, end int) ByteSlice {
+	return b[start:end]
+}
+
+// Reader implements a SeriesReader for a serialized byte stream
 // of series data.
-type chunkReader struct {
+type Reader struct {
 	// The underlying bytes holding the encoded series data.
 	bs []ByteSlice
 
 	// Closers for resources behind the byte slices.
 	cs []io.Closer
 
-	pool chunks.Pool
+	pool chunkenc.Pool
 }
 
-func newChunkReader(bs []ByteSlice, cs []io.Closer, pool chunks.Pool) (*chunkReader, error) {
+func newReader(bs []ByteSlice, cs []io.Closer, pool chunkenc.Pool) (*Reader, error) {
-	cr := chunkReader{pool: pool, bs: bs, cs: cs}
+	cr := Reader{pool: pool, bs: bs, cs: cs}
 
 	for i, b := range cr.bs {
 		if b.Len() < 4 {
@@ -321,44 +302,44 @@ func newChunkReader(bs []ByteSlice, cs []io.Closer, pool chunks.Pool) (*chunkRea
 	return &cr, nil
 }
 
-// NewChunkReader returns a new chunk reader against the given byte slices.
+// NewReader returns a new chunk reader against the given byte slices.
-func NewChunkReader(bs []ByteSlice, pool chunks.Pool) (ChunkReader, error) {
+func NewReader(bs []ByteSlice, pool chunkenc.Pool) (*Reader, error) {
 	if pool == nil {
-		pool = chunks.NewPool()
+		pool = chunkenc.NewPool()
 	}
-	return newChunkReader(bs, nil, pool)
+	return newReader(bs, nil, pool)
 }
 
-// NewDirChunkReader returns a new ChunkReader against sequentially numbered files in the
+// NewDirReader returns a new Reader against sequentially numbered files in the
 // given directory.
-func NewDirChunkReader(dir string, pool chunks.Pool) (ChunkReader, error) {
+func NewDirReader(dir string, pool chunkenc.Pool) (*Reader, error) {
 	files, err := sequenceFiles(dir)
 	if err != nil {
 		return nil, err
 	}
 	if pool == nil {
-		pool = chunks.NewPool()
+		pool = chunkenc.NewPool()
 	}
 
 	var bs []ByteSlice
 	var cs []io.Closer
 
 	for _, fn := range files {
-		f, err := openMmapFile(fn)
+		f, err := fileutil.OpenMmapFile(fn)
 		if err != nil {
 			return nil, errors.Wrapf(err, "mmap files")
 		}
 		cs = append(cs, f)
-		bs = append(bs, realByteSlice(f.b))
+		bs = append(bs, realByteSlice(f.Bytes()))
 	}
-	return newChunkReader(bs, cs, pool)
+	return newReader(bs, cs, pool)
 }
 
-func (s *chunkReader) Close() error {
+func (s *Reader) Close() error {
 	return closeAll(s.cs...)
 }
 
-func (s *chunkReader) Chunk(ref uint64) (chunks.Chunk, error) {
+func (s *Reader) Chunk(ref uint64) (chunkenc.Chunk, error) {
 	var (
 		seq = int(ref >> 32)
 		off = int((ref << 32) >> 32)
@@ -381,5 +362,47 @@ func (s *chunkReader) Chunk(ref uint64) (chunks.Chunk, error) {
 	}
 	r = b.Range(off+n, off+n+int(l))
 
-	return s.pool.Get(chunks.Encoding(r[0]), r[1:1+l])
+	return s.pool.Get(chunkenc.Encoding(r[0]), r[1:1+l])
+}
+
+func nextSequenceFile(dir string) (string, int, error) {
+	names, err := fileutil.ReadDir(dir)
+	if err != nil {
+		return "", 0, err
+	}
+
+	i := uint64(0)
+	for _, n := range names {
+		j, err := strconv.ParseUint(n, 10, 64)
+		if err != nil {
+			continue
+		}
+		i = j
+	}
+	return filepath.Join(dir, fmt.Sprintf("%0.6d", i+1)), int(i + 1), nil
+}
+
+func sequenceFiles(dir string) ([]string, error) {
+	files, err := ioutil.ReadDir(dir)
+	if err != nil {
+		return nil, err
+	}
+	var res []string
+
+	for _, fi := range files {
+		if _, err := strconv.ParseUint(fi.Name(), 10, 64); err != nil {
+			continue
+		}
+		res = append(res, filepath.Join(dir, fi.Name()))
+	}
+	return res, nil
+}
+
+func closeAll(cs ...io.Closer) (err error) {
+	for _, c := range cs {
+		if e := c.Close(); e != nil {
+			err = e
+		}
+	}
+	return err
 }

vendor/github.com/prometheus/tsdb/compact.go

@@ -26,8 +26,10 @@ import (
 	"github.com/oklog/ulid"
 	"github.com/pkg/errors"
 	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/chunks"
 	"github.com/prometheus/tsdb/fileutil"
+	"github.com/prometheus/tsdb/index"
 	"github.com/prometheus/tsdb/labels"
 )
 
@@ -56,7 +58,7 @@ type Compactor interface {
 
 	// Compact runs compaction against the provided directories. Must
 	// only be called concurrently with results of Plan().
-	Compact(dest string, dirs ...string) error
+	Compact(dest string, dirs ...string) (ulid.ULID, error)
 }
 
 // LeveledCompactor implements the Compactor interface.
@@ -65,7 +67,7 @@ type LeveledCompactor struct {
 	metrics   *compactorMetrics
 	logger    log.Logger
 	ranges    []int64
-	chunkPool chunks.Pool
+	chunkPool chunkenc.Pool
 }
 
 type compactorMetrics struct {
@@ -123,12 +125,12 @@ func newCompactorMetrics(r prometheus.Registerer) *compactorMetrics {
 }
 
 // NewLeveledCompactor returns a LeveledCompactor.
-func NewLeveledCompactor(r prometheus.Registerer, l log.Logger, ranges []int64, pool chunks.Pool) (*LeveledCompactor, error) {
+func NewLeveledCompactor(r prometheus.Registerer, l log.Logger, ranges []int64, pool chunkenc.Pool) (*LeveledCompactor, error) {
 	if len(ranges) == 0 {
 		return nil, errors.Errorf("at least one range must be provided")
 	}
 	if pool == nil {
-		pool = chunks.NewPool()
+		pool = chunkenc.NewPool()
 	}
 	return &LeveledCompactor{
 		ranges:    ranges,
@@ -303,7 +305,7 @@ func compactBlockMetas(uid ulid.ULID, blocks ...*BlockMeta) *BlockMeta {
 
 // Compact creates a new block in the compactor's directory from the blocks in the
 // provided directories.
-func (c *LeveledCompactor) Compact(dest string, dirs ...string) (err error) {
+func (c *LeveledCompactor) Compact(dest string, dirs ...string) (uid ulid.ULID, err error) {
 	var blocks []BlockReader
 	var bs []*Block
 	var metas []*BlockMeta
@@ -311,13 +313,13 @@ func (c *LeveledCompactor) Compact(dest string, dirs ...string) (err error) {
 	for _, d := range dirs {
 		b, err := OpenBlock(d, c.chunkPool)
 		if err != nil {
-			return err
+			return uid, err
 		}
 		defer b.Close()
 
 		meta, err := readMetaFile(d)
 		if err != nil {
-			return err
+			return uid, err
 		}
 
 		metas = append(metas, meta)
@@ -326,11 +328,11 @@ func (c *LeveledCompactor) Compact(dest string, dirs ...string) (err error) {
 	}
 
 	entropy := rand.New(rand.NewSource(time.Now().UnixNano()))
-	uid := ulid.MustNew(ulid.Now(), entropy)
+	uid = ulid.MustNew(ulid.Now(), entropy)
 
 	err = c.write(dest, compactBlockMetas(uid, metas...), blocks...)
 	if err == nil {
-		return nil
+		return uid, nil
 	}
 
 	var merr MultiError
@@ -342,7 +344,7 @@ func (c *LeveledCompactor) Compact(dest string, dirs ...string) (err error) {
 		}
 	}
 
-	return merr
+	return uid, merr
 }
 
 func (c *LeveledCompactor) Write(dest string, b BlockReader, mint, maxt int64) (ulid.ULID, error) {
@@ -370,7 +372,7 @@ type instrumentedChunkWriter struct {
 	trange  prometheus.Histogram
 }
 
-func (w *instrumentedChunkWriter) WriteChunks(chunks ...ChunkMeta) error {
+func (w *instrumentedChunkWriter) WriteChunks(chunks ...chunks.Meta) error {
 	for _, c := range chunks {
 		w.size.Observe(float64(len(c.Chunk.Bytes())))
 		w.samples.Observe(float64(c.Chunk.NumSamples()))
@@ -411,7 +413,7 @@ func (c *LeveledCompactor) write(dest string, meta *BlockMeta, blocks ...BlockRe
 	// data of all blocks.
 	var chunkw ChunkWriter
 
-	chunkw, err = newChunkWriter(chunkDir(tmp))
+	chunkw, err = chunks.NewWriter(chunkDir(tmp))
 	if err != nil {
 		return errors.Wrap(err, "open chunk writer")
 	}
@@ -425,7 +427,7 @@ func (c *LeveledCompactor) write(dest string, meta *BlockMeta, blocks ...BlockRe
 		}
 	}
 
-	indexw, err := newIndexWriter(tmp)
+	indexw, err := index.NewWriter(filepath.Join(tmp, indexFilename))
 	if err != nil {
 		return errors.Wrap(err, "open index writer")
 	}
@@ -514,7 +516,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 			allSymbols[s] = struct{}{}
 		}
 
-		all, err := indexr.Postings(allPostingsKey.Name, allPostingsKey.Value)
+		all, err := indexr.Postings(index.AllPostingsKey())
 		if err != nil {
 			return err
 		}
@@ -534,7 +536,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 
 	// We fully rebuild the postings list index from merged series.
 	var (
-		postings = newMemPostings()
+		postings = index.NewMemPostings()
 		values   = map[string]stringset{}
 		i        = uint64(0)
 	)
@@ -558,7 +560,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 					continue
 				}
 
-				newChunk := chunks.NewXORChunk()
+				newChunk := chunkenc.NewXORChunk()
 				app, err := newChunk.Appender()
 				if err != nil {
 					return err
@@ -599,7 +601,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 			}
 			valset.set(l.Value)
 		}
-		postings.add(i, lset)
+		postings.Add(i, lset)
 
 		i++
 	}
@@ -619,8 +621,8 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 		}
 	}
 
-	for _, l := range postings.sortedKeys() {
+	for _, l := range postings.SortedKeys() {
-		if err := indexw.WritePostings(l.Name, l.Value, postings.get(l.Name, l.Value)); err != nil {
+		if err := indexw.WritePostings(l.Name, l.Value, postings.Get(l.Name, l.Value)); err != nil {
 			return errors.Wrap(err, "write postings")
 		}
 	}
@@ -628,18 +630,18 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 }
 
 type compactionSeriesSet struct {
-	p          Postings
+	p          index.Postings
 	index      IndexReader
 	chunks     ChunkReader
 	tombstones TombstoneReader
 
 	l         labels.Labels
-	c         []ChunkMeta
+	c         []chunks.Meta
 	intervals Intervals
 	err       error
 }
 
-func newCompactionSeriesSet(i IndexReader, c ChunkReader, t TombstoneReader, p Postings) *compactionSeriesSet {
+func newCompactionSeriesSet(i IndexReader, c ChunkReader, t TombstoneReader, p index.Postings) *compactionSeriesSet {
 	return &compactionSeriesSet{
 		index:      i,
 		chunks:     c,
@@ -667,7 +669,7 @@ func (c *compactionSeriesSet) Next() bool {
 
 	// Remove completely deleted chunks.
 	if len(c.intervals) > 0 {
-		chks := make([]ChunkMeta, 0, len(c.c))
+		chks := make([]chunks.Meta, 0, len(c.c))
 		for _, chk := range c.c {
 			if !(Interval{chk.MinTime, chk.MaxTime}.isSubrange(c.intervals)) {
 				chks = append(chks, chk)
@@ -697,7 +699,7 @@ func (c *compactionSeriesSet) Err() error {
 	return c.p.Err()
 }
 
-func (c *compactionSeriesSet) At() (labels.Labels, []ChunkMeta, Intervals) {
+func (c *compactionSeriesSet) At() (labels.Labels, []chunks.Meta, Intervals) {
 	return c.l, c.c, c.intervals
 }
 
@@ -706,13 +708,13 @@ type compactionMerger struct {
 
 	aok, bok  bool
 	l         labels.Labels
-	c         []ChunkMeta
+	c         []chunks.Meta
 	intervals Intervals
 }
 
 type compactionSeries struct {
 	labels labels.Labels
-	chunks []*ChunkMeta
+	chunks []*chunks.Meta
 }
 
 func newCompactionMerger(a, b ChunkSeriesSet) (*compactionMerger, error) {
@@ -747,7 +749,7 @@ func (c *compactionMerger) Next() bool {
 	// While advancing child iterators the memory used for labels and chunks
 	// may be reused. When picking a series we have to store the result.
 	var lset labels.Labels
-	var chks []ChunkMeta
+	var chks []chunks.Meta
 
 	d := c.compare()
 	// Both sets contain the current series. Chain them into a single one.
@@ -788,7 +790,7 @@ func (c *compactionMerger) Err() error {
 	return c.b.Err()
 }
 
-func (c *compactionMerger) At() (labels.Labels, []ChunkMeta, Intervals) {
+func (c *compactionMerger) At() (labels.Labels, []chunks.Meta, Intervals) {
 	return c.l, c.c, c.intervals
 }
 

vendor/github.com/prometheus/tsdb/db.go

@@ -36,7 +36,7 @@ import (
 	"github.com/oklog/ulid"
 	"github.com/pkg/errors"
 	"github.com/prometheus/client_golang/prometheus"
-	"github.com/prometheus/tsdb/chunks"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/fileutil"
 	"github.com/prometheus/tsdb/labels"
 )
@@ -99,7 +99,7 @@ type DB struct {
 	logger    log.Logger
 	metrics   *dbMetrics
 	opts      *Options
-	chunkPool chunks.Pool
+	chunkPool chunkenc.Pool
 	compactor Compactor
 
 	// Mutex for that must be held when modifying the general block layout.
@@ -185,7 +185,7 @@ func Open(dir string, l log.Logger, r prometheus.Registerer, opts *Options) (db
 		donec:              make(chan struct{}),
 		stopc:              make(chan struct{}),
 		compactionsEnabled: true,
-		chunkPool:          chunks.NewPool(),
+		chunkPool:          chunkenc.NewPool(),
 	}
 	db.metrics = newDBMetrics(db, r)
 
@@ -386,7 +386,7 @@ func (db *DB) compact() (changes bool, err error) {
 		default:
 		}
 
-		if err := db.compactor.Compact(db.dir, plan...); err != nil {
+		if _, err := db.compactor.Compact(db.dir, plan...); err != nil {
 			return changes, errors.Wrapf(err, "compact %s", plan)
 		}
 		changes = true

vendor/github.com/prometheus/tsdb/encoding_helpers.go

@@ -5,8 +5,12 @@ import (
 	"hash"
 	"hash/crc32"
 	"unsafe"
+
+	"github.com/pkg/errors"
 )
 
+var errInvalidSize = errors.New("invalid size")
+
 // enbuf is a helper type to populate a byte slice with various types.
 type encbuf struct {
 	b []byte

vendor/github.com/prometheus/tsdb/fileutil/mmap.go

@@ -0,0 +1,48 @@
+package fileutil
+
+import (
+	"os"
+
+	"github.com/pkg/errors"
+)
+
+type MmapFile struct {
+	f *os.File
+	b []byte
+}
+
+func OpenMmapFile(path string) (*MmapFile, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, errors.Wrap(err, "try lock file")
+	}
+	info, err := f.Stat()
+	if err != nil {
+		return nil, errors.Wrap(err, "stat")
+	}
+
+	b, err := mmap(f, int(info.Size()))
+	if err != nil {
+		return nil, errors.Wrap(err, "mmap")
+	}
+
+	return &MmapFile{f: f, b: b}, nil
+}
+
+func (f *MmapFile) Close() error {
+	err0 := munmap(f.b)
+	err1 := f.f.Close()
+
+	if err0 != nil {
+		return err0
+	}
+	return err1
+}
+
+func (f *MmapFile) File() *os.File {
+	return f.f
+}
+
+func (f *MmapFile) Bytes() []byte {
+	return f.b
+}

vendor/github.com/prometheus/tsdb/fileutil/mmap_unix.go

@@ -13,7 +13,7 @@
 
 // +build !windows,!plan9
 
-package tsdb
+package fileutil
 
 import (
 	"os"

vendor/github.com/prometheus/tsdb/fileutil/mmap_windows.go

@@ -11,7 +11,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package tsdb
+package fileutil
 
 import (
 	"os"

vendor/github.com/prometheus/tsdb/head.go

@@ -17,6 +17,7 @@ import (
 	"math"
 	"runtime"
 	"sort"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -25,7 +26,9 @@ import (
 	"github.com/go-kit/kit/log/level"
 	"github.com/pkg/errors"
 	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/chunks"
+	"github.com/prometheus/tsdb/index"
 	"github.com/prometheus/tsdb/labels"
 )
 
@@ -64,7 +67,7 @@ type Head struct {
 	symbols map[string]struct{}
 	values  map[string]stringset // label names to possible values
 
-	postings *memPostings // postings lists for terms
+	postings *index.MemPostings // postings lists for terms
 
 	tombstones memTombstones
 }
@@ -185,7 +188,7 @@ func NewHead(r prometheus.Registerer, l log.Logger, wal WAL, chunkRange int64) (
 		series:     newStripeSeries(),
 		values:     map[string]stringset{},
 		symbols:    map[string]struct{}{},
-		postings:   newUnorderedMemPostings(),
+		postings:   index.NewUnorderedMemPostings(),
 		tombstones: memTombstones{},
 	}
 	h.metrics = newHeadMetrics(h, r)
@@ -226,7 +229,7 @@ func (h *Head) processWALSamples(
 
 // ReadWAL initializes the head by consuming the write ahead log.
 func (h *Head) ReadWAL() error {
-	defer h.postings.ensureOrder()
+	defer h.postings.EnsureOrder()
 
 	r := h.wal.Reader()
 	mint := h.MinTime()
@@ -574,23 +577,16 @@ func (h *Head) Delete(mint, maxt int64, ms ...labels.Matcher) error {
 
 	ir := h.indexRange(mint, maxt)
 
-	p, absent, err := PostingsForMatchers(ir, ms...)
+	p, err := PostingsForMatchers(ir, ms...)
 	if err != nil {
 		return errors.Wrap(err, "select series")
 	}
 
 	var stones []Stone
 
-Outer:
 	for p.Next() {
 		series := h.series.getByID(p.At())
 
-		for _, abs := range absent {
-			if series.lset.Get(abs) != "" {
-				continue Outer
-			}
-		}
-
 		// Delete only until the current values and not beyond.
 		t0, t1 := clampInterval(mint, maxt, series.minTime(), series.maxTime())
 		stones = append(stones, Stone{p.At(), Intervals{{t0, t1}}})
@@ -623,64 +619,14 @@ func (h *Head) gc() {
 	h.metrics.chunksRemoved.Add(float64(chunksRemoved))
 	h.metrics.chunks.Sub(float64(chunksRemoved))
 
-	// Remove deleted series IDs from the postings lists. First do a collection
+	// Remove deleted series IDs from the postings lists.
-	// run where we rebuild all postings that have something to delete
+	h.postings.Delete(deleted)
-	h.postings.mtx.RLock()
-
-	type replEntry struct {
-		idx int
-		l   []uint64
-	}
-	collected := map[labels.Label]replEntry{}
-
-	for t, p := range h.postings.m {
-		repl := replEntry{idx: len(p)}
-
-		for i, id := range p {
-			if _, ok := deleted[id]; ok {
-				// First ID that got deleted, initialize replacement with
-				// all remaining IDs so far.
-				if repl.l == nil {
-					repl.l = make([]uint64, 0, len(p))
-					repl.l = append(repl.l, p[:i]...)
-				}
-				continue
-			}
-			// Only add to the replacement once we know we have to do it.
-			if repl.l != nil {
-				repl.l = append(repl.l, id)
-			}
-		}
-		if repl.l != nil {
-			collected[t] = repl
-		}
-	}
-
-	h.postings.mtx.RUnlock()
-
-	// Replace all postings that have changed. Append all IDs that may have
-	// been added while we switched locks.
-	h.postings.mtx.Lock()
-
-	for t, repl := range collected {
-		l := append(repl.l, h.postings.m[t][repl.idx:]...)
-
-		if len(l) > 0 {
-			h.postings.m[t] = l
-		} else {
-			delete(h.postings.m, t)
-		}
-	}
-
-	h.postings.mtx.Unlock()
 
 	// Rebuild symbols and label value indices from what is left in the postings terms.
-	h.postings.mtx.RLock()
-
 	symbols := make(map[string]struct{})
 	values := make(map[string]stringset, len(h.values))
 
-	for t := range h.postings.m {
+	h.postings.Iter(func(t labels.Label, _ index.Postings) error {
 		symbols[t.Name] = struct{}{}
 		symbols[t.Value] = struct{}{}
 
@@ -690,9 +636,8 @@ func (h *Head) gc() {
 			values[t.Name] = ss
 		}
 		ss.set(t.Value)
-	}
+		return nil
-
+	})
-	h.postings.mtx.RUnlock()
 
 	h.symMtx.Lock()
 
@@ -772,13 +717,24 @@ func unpackChunkID(id uint64) (seriesID, chunkID uint64) {
 }
 
 // Chunk returns the chunk for the reference number.
-func (h *headChunkReader) Chunk(ref uint64) (chunks.Chunk, error) {
+func (h *headChunkReader) Chunk(ref uint64) (chunkenc.Chunk, error) {
 	sid, cid := unpackChunkID(ref)
 
 	s := h.head.series.getByID(sid)
+	// This means that the series has been garbage collected.
+	if s == nil {
+		return nil, ErrNotFound
+	}
 
 	s.Lock()
 	c := s.chunk(int(cid))
+
+	// This means that the chunk has been garbage collected.
+	if c == nil {
+		s.Unlock()
+		return nil, ErrNotFound
+	}
+
 	mint, maxt := c.minTime, c.maxTime
 	s.Unlock()
 
@@ -794,12 +750,12 @@ func (h *headChunkReader) Chunk(ref uint64) (chunks.Chunk, error) {
 }
 
 type safeChunk struct {
-	chunks.Chunk
+	chunkenc.Chunk
 	s   *memSeries
 	cid int
 }
 
-func (c *safeChunk) Iterator() chunks.Iterator {
+func (c *safeChunk) Iterator() chunkenc.Iterator {
 	c.s.Lock()
 	it := c.s.iterator(c.cid)
 	c.s.Unlock()
@@ -832,7 +788,7 @@ func (h *headIndexReader) Symbols() (map[string]struct{}, error) {
 }
 
 // LabelValues returns the possible label values
-func (h *headIndexReader) LabelValues(names ...string) (StringTuples, error) {
+func (h *headIndexReader) LabelValues(names ...string) (index.StringTuples, error) {
 	if len(names) != 1 {
 		return nil, errInvalidSize
 	}
@@ -846,22 +802,22 @@ func (h *headIndexReader) LabelValues(names ...string) (StringTuples, error) {
 	}
 	sort.Strings(sl)
 
-	return &stringTuples{l: len(names), s: sl}, nil
+	return index.NewStringTuples(sl, len(names))
 }
 
 // Postings returns the postings list iterator for the label pair.
-func (h *headIndexReader) Postings(name, value string) (Postings, error) {
+func (h *headIndexReader) Postings(name, value string) (index.Postings, error) {
-	return h.head.postings.get(name, value), nil
+	return h.head.postings.Get(name, value), nil
 }
 
-func (h *headIndexReader) SortedPostings(p Postings) Postings {
+func (h *headIndexReader) SortedPostings(p index.Postings) index.Postings {
 	ep := make([]uint64, 0, 128)
 
 	for p.Next() {
 		ep = append(ep, p.At())
 	}
 	if err := p.Err(); err != nil {
-		return errPostings{err: errors.Wrap(err, "expand postings")}
+		return index.ErrPostings(errors.Wrap(err, "expand postings"))
 	}
 
 	sort.Slice(ep, func(i, j int) bool {
@@ -874,11 +830,11 @@ func (h *headIndexReader) SortedPostings(p Postings) Postings {
 		}
 		return labels.Compare(a.lset, b.lset) < 0
 	})
-	return newListPostings(ep)
+	return index.NewListPostings(ep)
 }
 
 // Series returns the series for the given reference.
-func (h *headIndexReader) Series(ref uint64, lbls *labels.Labels, chks *[]ChunkMeta) error {
+func (h *headIndexReader) Series(ref uint64, lbls *labels.Labels, chks *[]chunks.Meta) error {
 	s := h.head.series.getByID(ref)
 
 	if s == nil {
@@ -897,7 +853,7 @@ func (h *headIndexReader) Series(ref uint64, lbls *labels.Labels, chks *[]ChunkM
 		if !intervalOverlap(c.minTime, c.maxTime, h.mint, h.maxt) {
 			continue
 		}
-		*chks = append(*chks, ChunkMeta{
+		*chks = append(*chks, chunks.Meta{
 			MinTime: c.minTime,
 			MaxTime: c.maxTime,
 			Ref:     packChunkID(s.ref, uint64(s.chunkID(i))),
@@ -945,7 +901,7 @@ func (h *Head) getOrCreateWithID(id, hash uint64, lset labels.Labels) (*memSerie
 	h.metrics.series.Inc()
 	h.metrics.seriesCreated.Inc()
 
-	h.postings.add(id, lset)
+	h.postings.Add(id, lset)
 
 	h.symMtx.Lock()
 	defer h.symMtx.Unlock()
@@ -1150,7 +1106,7 @@ type memSeries struct {
 	lastValue float64
 	sampleBuf [4]sample
 
-	app chunks.Appender // Current appender for the chunk.
+	app chunkenc.Appender // Current appender for the chunk.
 }
 
 func (s *memSeries) minTime() int64 {
@@ -1163,7 +1119,7 @@ func (s *memSeries) maxTime() int64 {
 
 func (s *memSeries) cut(mint int64) *memChunk {
 	c := &memChunk{
-		chunk:   chunks.NewXORChunk(),
+		chunk:   chunkenc.NewXORChunk(),
 		minTime: mint,
 		maxTime: math.MinInt64,
 	}
@@ -1291,13 +1247,13 @@ func computeChunkEndTime(start, cur, max int64) int64 {
 	return start + (max-start)/a
 }
 
-func (s *memSeries) iterator(id int) chunks.Iterator {
+func (s *memSeries) iterator(id int) chunkenc.Iterator {
 	c := s.chunk(id)
 	// TODO(fabxc): Work around! A querier may have retrieved a pointer to a series' chunk,
 	// which got then garbage collected before it got accessed.
 	// We must ensure to not garbage collect as long as any readers still hold a reference.
 	if c == nil {
-		return chunks.NewNopIterator()
+		return chunkenc.NewNopIterator()
 	}
 
 	if id-s.firstChunkID < len(s.chunks)-1 {
@@ -1322,12 +1278,12 @@ func (s *memSeries) head() *memChunk {
 }
 
 type memChunk struct {
-	chunk            chunks.Chunk
+	chunk            chunkenc.Chunk
 	minTime, maxTime int64
 }
 
 type memSafeIterator struct {
-	chunks.Iterator
+	chunkenc.Iterator
 
 	i     int
 	total int
@@ -1352,3 +1308,27 @@ func (it *memSafeIterator) At() (int64, float64) {
 	s := it.buf[4-(it.total-it.i)]
 	return s.t, s.v
 }
+
+type stringset map[string]struct{}
+
+func (ss stringset) set(s string) {
+	ss[s] = struct{}{}
+}
+
+func (ss stringset) has(s string) bool {
+	_, ok := ss[s]
+	return ok
+}
+
+func (ss stringset) String() string {
+	return strings.Join(ss.slice(), ",")
+}
+
+func (ss stringset) slice() []string {
+	slice := make([]string, 0, len(ss))
+	for k := range ss {
+		slice = append(slice, k)
+	}
+	sort.Strings(slice)
+	return slice
+}

vendor/github.com/prometheus/tsdb/index/encoding_helpers.go

@@ -0,0 +1,179 @@
+package index
+
+import (
+	"encoding/binary"
+	"hash"
+	"hash/crc32"
+	"unsafe"
+)
+
+// enbuf is a helper type to populate a byte slice with various types.
+type encbuf struct {
+	b []byte
+	c [binary.MaxVarintLen64]byte
+}
+
+func (e *encbuf) reset()      { e.b = e.b[:0] }
+func (e *encbuf) get() []byte { return e.b }
+func (e *encbuf) len() int    { return len(e.b) }
+
+func (e *encbuf) putString(s string) { e.b = append(e.b, s...) }
+func (e *encbuf) putBytes(b []byte)  { e.b = append(e.b, b...) }
+func (e *encbuf) putByte(c byte)     { e.b = append(e.b, c) }
+
+func (e *encbuf) putBE32int(x int)      { e.putBE32(uint32(x)) }
+func (e *encbuf) putBE64int(x int)      { e.putBE64(uint64(x)) }
+func (e *encbuf) putBE64int64(x int64)  { e.putBE64(uint64(x)) }
+func (e *encbuf) putUvarint32(x uint32) { e.putUvarint64(uint64(x)) }
+func (e *encbuf) putUvarint(x int)      { e.putUvarint64(uint64(x)) }
+
+func (e *encbuf) putBE32(x uint32) {
+	binary.BigEndian.PutUint32(e.c[:], x)
+	e.b = append(e.b, e.c[:4]...)
+}
+
+func (e *encbuf) putBE64(x uint64) {
+	binary.BigEndian.PutUint64(e.c[:], x)
+	e.b = append(e.b, e.c[:8]...)
+}
+
+func (e *encbuf) putUvarint64(x uint64) {
+	n := binary.PutUvarint(e.c[:], x)
+	e.b = append(e.b, e.c[:n]...)
+}
+
+func (e *encbuf) putVarint64(x int64) {
+	n := binary.PutVarint(e.c[:], x)
+	e.b = append(e.b, e.c[:n]...)
+}
+
+// putVarintStr writes a string to the buffer prefixed by its varint length (in bytes!).
+func (e *encbuf) putUvarintStr(s string) {
+	b := *(*[]byte)(unsafe.Pointer(&s))
+	e.putUvarint(len(b))
+	e.putString(s)
+}
+
+// putHash appends a hash over the buffers current contents to the buffer.
+func (e *encbuf) putHash(h hash.Hash) {
+	h.Reset()
+	_, err := h.Write(e.b)
+	if err != nil {
+		panic(err) // The CRC32 implementation does not error
+	}
+	e.b = h.Sum(e.b)
+}
+
+// decbuf provides safe methods to extract data from a byte slice. It does all
+// necessary bounds checking and advancing of the byte slice.
+// Several datums can be extracted without checking for errors. However, before using
+// any datum, the err() method must be checked.
+type decbuf struct {
+	b []byte
+	e error
+}
+
+func (d *decbuf) uvarint() int      { return int(d.uvarint64()) }
+func (d *decbuf) uvarint32() uint32 { return uint32(d.uvarint64()) }
+func (d *decbuf) be32int() int      { return int(d.be32()) }
+func (d *decbuf) be64int64() int64  { return int64(d.be64()) }
+
+// crc32 returns a CRC32 checksum over the remaining bytes.
+func (d *decbuf) crc32() uint32 {
+	return crc32.Checksum(d.b, castagnoliTable)
+}
+
+func (d *decbuf) uvarintStr() string {
+	l := d.uvarint64()
+	if d.e != nil {
+		return ""
+	}
+	if len(d.b) < int(l) {
+		d.e = errInvalidSize
+		return ""
+	}
+	s := string(d.b[:l])
+	d.b = d.b[l:]
+	return s
+}
+
+func (d *decbuf) varint64() int64 {
+	if d.e != nil {
+		return 0
+	}
+	x, n := binary.Varint(d.b)
+	if n < 1 {
+		d.e = errInvalidSize
+		return 0
+	}
+	d.b = d.b[n:]
+	return x
+}
+
+func (d *decbuf) uvarint64() uint64 {
+	if d.e != nil {
+		return 0
+	}
+	x, n := binary.Uvarint(d.b)
+	if n < 1 {
+		d.e = errInvalidSize
+		return 0
+	}
+	d.b = d.b[n:]
+	return x
+}
+
+func (d *decbuf) be64() uint64 {
+	if d.e != nil {
+		return 0
+	}
+	if len(d.b) < 4 {
+		d.e = errInvalidSize
+		return 0
+	}
+	x := binary.BigEndian.Uint64(d.b)
+	d.b = d.b[8:]
+	return x
+}
+
+func (d *decbuf) be32() uint32 {
+	if d.e != nil {
+		return 0
+	}
+	if len(d.b) < 4 {
+		d.e = errInvalidSize
+		return 0
+	}
+	x := binary.BigEndian.Uint32(d.b)
+	d.b = d.b[4:]
+	return x
+}
+
+func (d *decbuf) byte() byte {
+	if d.e != nil {
+		return 0
+	}
+	if len(d.b) < 1 {
+		d.e = errInvalidSize
+		return 0
+	}
+	x := d.b[0]
+	d.b = d.b[1:]
+	return x
+}
+
+func (d *decbuf) decbuf(l int) decbuf {
+	if d.e != nil {
+		return decbuf{e: d.e}
+	}
+	if l > len(d.b) {
+		return decbuf{e: errInvalidSize}
+	}
+	r := decbuf{b: d.b[:l]}
+	d.b = d.b[l:]
+	return r
+}
+
+func (d *decbuf) err() error  { return d.e }
+func (d *decbuf) len() int    { return len(d.b) }
+func (d *decbuf) get() []byte { return d.b }

vendor/github.com/prometheus/tsdb/index/index.go

@@ -11,13 +11,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package tsdb
+package index
 
 import (
 	"bufio"
 	"encoding/binary"
 	"fmt"
 	"hash"
+	"hash/crc32"
 	"io"
 	"math"
 	"os"
@@ -26,6 +27,7 @@ import (
 	"strings"
 
 	"github.com/pkg/errors"
+	"github.com/prometheus/tsdb/chunks"
 	"github.com/prometheus/tsdb/fileutil"
 	"github.com/prometheus/tsdb/labels"
 )
@@ -35,18 +37,12 @@ const (
 	MagicIndex = 0xBAAAD700
 
 	indexFormatV1 = 1
-
-	size_unit = 4
 )
 
-const indexFilename = "index"
-
-const compactionPageBytes = minSectorSize * 64
-
 type indexWriterSeries struct {
 	labels labels.Labels
-	chunks []ChunkMeta // series file offset of chunks
+	chunks []chunks.Meta // series file offset of chunks
-	offset uint32      // index file offset of series reference
+	offset uint32        // index file offset of series reference
 }
 
 type indexWriterSeriesSlice []*indexWriterSeries
@@ -87,37 +83,24 @@ func (s indexWriterStage) String() string {
 	return "<unknown>"
 }
 
-// IndexWriter serializes the index for a block of series data.
+// The table gets initialized with sync.Once but may still cause a race
-// The methods must be called in the order they are specified in.
+// with any other use of the crc32 package anywhere. Thus we initialize it
-type IndexWriter interface {
+// before.
-	// AddSymbols registers all string symbols that are encountered in series
+var castagnoliTable *crc32.Table
-	// and other indices.
-	AddSymbols(sym map[string]struct{}) error
 
-	// AddSeries populates the index writer with a series and its offsets
+func init() {
-	// of chunks that the index can reference.
+	castagnoliTable = crc32.MakeTable(crc32.Castagnoli)
-	// Implementations may require series to be insert in increasing order by
+}
-	// their labels.
-	// The reference numbers are used to resolve entries in postings lists that
-	// are added later.
-	AddSeries(ref uint64, l labels.Labels, chunks ...ChunkMeta) error
 
-	// WriteLabelIndex serializes an index from label names to values.
+// newCRC32 initializes a CRC32 hash with a preconfigured polynomial, so the
-	// The passed in values chained tuples of strings of the length of names.
+// polynomial may be easily changed in one location at a later time, if necessary.
-	WriteLabelIndex(names []string, values []string) error
+func newCRC32() hash.Hash32 {
-
+	return crc32.New(castagnoliTable)
-	// WritePostings writes a postings list for a single label pair.
-	// The Postings here contain refs to the series that were added.
-	WritePostings(name, value string, it Postings) error
-
-	// Close writes any finalization and closes the resources associated with
-	// the underlying writer.
-	Close() error
 }
 
 // indexWriter implements the IndexWriter interface for the standard
 // serialization format.
-type indexWriter struct {
+type Writer struct {
 	f    *os.File
 	fbuf *bufio.Writer
 	pos  uint64
@@ -150,14 +133,17 @@ type indexTOC struct {
 	postingsTable     uint64
 }
 
-func newIndexWriter(dir string) (*indexWriter, error) {
+// NewWriter returns a new Writer to the given filename.
+func NewWriter(fn string) (*Writer, error) {
+	dir := filepath.Dir(fn)
+
 	df, err := fileutil.OpenDir(dir)
 	if err != nil {
 		return nil, err
 	}
 	defer df.Close() // close for flatform windows
 
-	f, err := os.OpenFile(filepath.Join(dir, indexFilename), os.O_CREATE|os.O_WRONLY, 0666)
+	f, err := os.OpenFile(fn, os.O_CREATE|os.O_WRONLY, 0666)
 	if err != nil {
 		return nil, err
 	}
@@ -165,7 +151,7 @@ func newIndexWriter(dir string) (*indexWriter, error) {
 		return nil, errors.Wrap(err, "sync dir")
 	}
 
-	iw := &indexWriter{
+	iw := &Writer{
 		f:     f,
 		fbuf:  bufio.NewWriterSize(f, 1<<22),
 		pos:   0,
@@ -187,7 +173,7 @@ func newIndexWriter(dir string) (*indexWriter, error) {
 	return iw, nil
 }
 
-func (w *indexWriter) write(bufs ...[]byte) error {
+func (w *Writer) write(bufs ...[]byte) error {
 	for _, b := range bufs {
 		n, err := w.fbuf.Write(b)
 		w.pos += uint64(n)
@@ -206,18 +192,18 @@ func (w *indexWriter) write(bufs ...[]byte) error {
 }
 
 // addPadding adds zero byte padding until the file size is a multiple size_unit.
-func (w *indexWriter) addPadding() error {
+func (w *Writer) addPadding(size int) error {
-	p := w.pos % size_unit
+	p := w.pos % uint64(size)
 	if p == 0 {
 		return nil
 	}
-	p = size_unit - p
+	p = uint64(size) - p
 	return errors.Wrap(w.write(make([]byte, p)), "add padding")
 }
 
 // ensureStage handles transitions between write stages and ensures that IndexWriter
 // methods are called in an order valid for the implementation.
-func (w *indexWriter) ensureStage(s indexWriterStage) error {
+func (w *Writer) ensureStage(s indexWriterStage) error {
 	if w.stage == s {
 		return nil
 	}
@@ -256,7 +242,7 @@ func (w *indexWriter) ensureStage(s indexWriterStage) error {
 	return nil
 }
 
-func (w *indexWriter) writeMeta() error {
+func (w *Writer) writeMeta() error {
 	w.buf1.reset()
 	w.buf1.putBE32(MagicIndex)
 	w.buf1.putByte(indexFormatV1)
@@ -264,7 +250,7 @@ func (w *indexWriter) writeMeta() error {
 	return w.write(w.buf1.get())
 }
 
-func (w *indexWriter) AddSeries(ref uint64, lset labels.Labels, chunks ...ChunkMeta) error {
+func (w *Writer) AddSeries(ref uint64, lset labels.Labels, chunks ...chunks.Meta) error {
 	if err := w.ensureStage(idxStageSeries); err != nil {
 		return err
 	}
@@ -328,7 +314,7 @@ func (w *indexWriter) AddSeries(ref uint64, lset labels.Labels, chunks ...ChunkM
 	return nil
 }
 
-func (w *indexWriter) AddSymbols(sym map[string]struct{}) error {
+func (w *Writer) AddSymbols(sym map[string]struct{}) error {
 	if err := w.ensureStage(idxStageSymbols); err != nil {
 		return err
 	}
@@ -361,7 +347,7 @@ func (w *indexWriter) AddSymbols(sym map[string]struct{}) error {
 	return errors.Wrap(err, "write symbols")
 }
 
-func (w *indexWriter) WriteLabelIndex(names []string, values []string) error {
+func (w *Writer) WriteLabelIndex(names []string, values []string) error {
 	if len(values)%len(names) != 0 {
 		return errors.Errorf("invalid value list length %d for %d names", len(values), len(names))
 	}
@@ -369,14 +355,14 @@ func (w *indexWriter) WriteLabelIndex(names []string, values []string) error {
 		return errors.Wrap(err, "ensure stage")
 	}
 
-	valt, err := newStringTuples(values, len(names))
+	valt, err := NewStringTuples(values, len(names))
 	if err != nil {
 		return err
 	}
 	sort.Sort(valt)
 
 	// Align beginning to 4 bytes for more efficient index list scans.
-	if err := w.addPadding(); err != nil {
+	if err := w.addPadding(4); err != nil {
 		return err
 	}
 
@@ -407,7 +393,7 @@ func (w *indexWriter) WriteLabelIndex(names []string, values []string) error {
 }
 
 // writeOffsetTable writes a sequence of readable hash entries.
-func (w *indexWriter) writeOffsetTable(entries []hashEntry) error {
+func (w *Writer) writeOffsetTable(entries []hashEntry) error {
 	w.buf2.reset()
 	w.buf2.putBE32int(len(entries))
 
@@ -428,7 +414,7 @@ func (w *indexWriter) writeOffsetTable(entries []hashEntry) error {
 
 const indexTOCLen = 6*8 + 4
 
-func (w *indexWriter) writeTOC() error {
+func (w *Writer) writeTOC() error {
 	w.buf1.reset()
 
 	w.buf1.putBE64(w.toc.symbols)
@@ -443,13 +429,13 @@ func (w *indexWriter) writeTOC() error {
 	return w.write(w.buf1.get())
 }
 
-func (w *indexWriter) WritePostings(name, value string, it Postings) error {
+func (w *Writer) WritePostings(name, value string, it Postings) error {
 	if err := w.ensureStage(idxStagePostings); err != nil {
 		return errors.Wrap(err, "ensure stage")
 	}
 
 	// Align beginning to 4 bytes for more efficient postings list scans.
-	if err := w.addPadding(); err != nil {
+	if err := w.addPadding(4); err != nil {
 		return err
 	}
 
@@ -506,7 +492,7 @@ type hashEntry struct {
 	offset uint64
 }
 
-func (w *indexWriter) Close() error {
+func (w *Writer) Close() error {
 	if err := w.ensureStage(idxStageDone); err != nil {
 		return err
 	}
@@ -519,37 +505,6 @@ func (w *indexWriter) Close() error {
 	return w.f.Close()
 }
 
-// IndexReader provides reading access of serialized index data.
-type IndexReader interface {
-	// Symbols returns a set of string symbols that may occur in series' labels
-	// and indices.
-	Symbols() (map[string]struct{}, error)
-
-	// LabelValues returns the possible label values
-	LabelValues(names ...string) (StringTuples, error)
-
-	// Postings returns the postings list iterator for the label pair.
-	// The Postings here contain the offsets to the series inside the index.
-	// Found IDs are not strictly required to point to a valid Series, e.g. during
-	// background garbage collections.
-	Postings(name, value string) (Postings, error)
-
-	// SortedPostings returns a postings list that is reordered to be sorted
-	// by the label set of the underlying series.
-	SortedPostings(Postings) Postings
-
-	// Series populates the given labels and chunk metas for the series identified
-	// by the reference.
-	// Returns ErrNotFound if the ref does not resolve to a known series.
-	Series(ref uint64, lset *labels.Labels, chks *[]ChunkMeta) error
-
-	// LabelIndices returns the label pairs for which indices exist.
-	LabelIndices() ([][]string, error)
-
-	// Close released the underlying resources of the reader.
-	Close() error
-}
-
 // StringTuples provides access to a sorted list of string tuples.
 type StringTuples interface {
 	// Total number of tuples in the list.
@@ -558,7 +513,7 @@ type StringTuples interface {
 	At(i int) ([]string, error)
 }
 
-type indexReader struct {
+type Reader struct {
 	// The underlying byte slice holding the encoded series data.
 	b   ByteSlice
 	toc indexTOC
@@ -568,7 +523,7 @@ type indexReader struct {
 
 	// Cached hashmaps of section offsets.
 	labels   map[string]uint32
-	postings map[string]uint32
+	postings map[labels.Label]uint32
 	// Cache of read symbols. Strings that are returned when reading from the
 	// block are always backed by true strings held in here rather than
 	// strings that are backed by byte slices from the mmap'd index file. This
@@ -576,6 +531,8 @@ type indexReader struct {
 	// the block has been unmapped.
 	symbols map[uint32]string
 
+	dec *DecoderV1
+
 	crc32 hash.Hash32
 }
 
@@ -605,26 +562,28 @@ func (b realByteSlice) Sub(start, end int) ByteSlice {
 	return b[start:end]
 }
 
-// NewIndexReader returns a new IndexReader on the given byte slice.
+// NewReader returns a new IndexReader on the given byte slice.
-func NewIndexReader(b ByteSlice) (IndexReader, error) {
+func NewReader(b ByteSlice) (*Reader, error) {
-	return newIndexReader(b, nil)
+	return newReader(b, nil)
 }
 
-// NewFileIndexReader returns a new index reader against the given index file.
+// NewFileReader returns a new index reader against the given index file.
-func NewFileIndexReader(path string) (IndexReader, error) {
+func NewFileReader(path string) (*Reader, error) {
-	f, err := openMmapFile(path)
+	f, err := fileutil.OpenMmapFile(path)
 	if err != nil {
 		return nil, err
 	}
-	return newIndexReader(realByteSlice(f.b), f)
+	return newReader(realByteSlice(f.Bytes()), f)
 }
 
-func newIndexReader(b ByteSlice, c io.Closer) (*indexReader, error) {
+func newReader(b ByteSlice, c io.Closer) (*Reader, error) {
-	r := &indexReader{
+	r := &Reader{
-		b:       b,
+		b:        b,
-		c:       c,
+		c:        c,
-		symbols: map[uint32]string{},
+		symbols:  map[uint32]string{},
-		crc32:   newCRC32(),
+		labels:   map[string]uint32{},
+		postings: map[labels.Label]uint32{},
+		crc32:    newCRC32(),
 	}
 	// Verify magic number.
 	if b.Len() < 4 {
@@ -642,15 +601,56 @@ func newIndexReader(b ByteSlice, c io.Closer) (*indexReader, error) {
 	}
 	var err error
 
-	r.labels, err = r.readOffsetTable(r.toc.labelIndicesTable)
+	err = r.readOffsetTable(r.toc.labelIndicesTable, func(key []string, off uint32) error {
+		if len(key) != 1 {
+			return errors.Errorf("unexpected key length %d", len(key))
+		}
+		r.labels[key[0]] = off
+		return nil
+	})
 	if err != nil {
 		return nil, errors.Wrap(err, "read label index table")
 	}
-	r.postings, err = r.readOffsetTable(r.toc.postingsTable)
+	err = r.readOffsetTable(r.toc.postingsTable, func(key []string, off uint32) error {
-	return r, errors.Wrap(err, "read postings table")
+		if len(key) != 2 {
+			return errors.Errorf("unexpected key length %d", len(key))
+		}
+		r.postings[labels.Label{Name: key[0], Value: key[1]}] = off
+		return nil
+	})
+	if err != nil {
+		return nil, errors.Wrap(err, "read postings table")
+	}
+
+	r.dec = &DecoderV1{symbols: r.symbols}
+
+	return r, nil
 }
 
-func (r *indexReader) readTOC() error {
+// Range marks a byte range.
+type Range struct {
+	Start, End int64
+}
+
+// PostingsRanges returns a new map of byte range in the underlying index file
+// for all postings lists.
+func (r *Reader) PostingsRanges() (map[labels.Label]Range, error) {
+	m := map[labels.Label]Range{}
+
+	for l, start := range r.postings {
+		d := r.decbufAt(int(start))
+		if d.err() != nil {
+			return nil, d.err()
+		}
+		m[l] = Range{
+			Start: int64(start) + 4,
+			End:   int64(start) + 4 + int64(d.len()),
+		}
+	}
+	return m, nil
+}
+
+func (r *Reader) readTOC() error {
 	if r.b.Len() < indexTOCLen {
 		return errInvalidSize
 	}
@@ -676,7 +676,7 @@ func (r *indexReader) readTOC() error {
 // decbufAt returns a new decoding buffer. It expects the first 4 bytes
 // after offset to hold the big endian encoded content length, followed by the contents and the expected
 // checksum.
-func (r *indexReader) decbufAt(off int) decbuf {
+func (r *Reader) decbufAt(off int) decbuf {
 	if r.b.Len() < off+4 {
 		return decbuf{e: errInvalidSize}
 	}
@@ -700,7 +700,7 @@ func (r *indexReader) decbufAt(off int) decbuf {
 // decbufUvarintAt returns a new decoding buffer. It expects the first bytes
 // after offset to hold the uvarint-encoded buffers length, followed by the contents and the expected
 // checksum.
-func (r *indexReader) decbufUvarintAt(off int) decbuf {
+func (r *Reader) decbufUvarintAt(off int) decbuf {
 	// We never have to access this method at the far end of the byte slice. Thus just checking
 	// against the MaxVarintLen32 is sufficient.
 	if r.b.Len() < off+binary.MaxVarintLen32 {
@@ -730,7 +730,7 @@ func (r *indexReader) decbufUvarintAt(off int) decbuf {
 // readSymbols reads the symbol table fully into memory and allocates proper strings for them.
 // Strings backed by the mmap'd memory would cause memory faults if applications keep using them
 // after the reader is closed.
-func (r *indexReader) readSymbols(off int) error {
+func (r *Reader) readSymbols(off int) error {
 	if off == 0 {
 		return nil
 	}
@@ -752,16 +752,13 @@ func (r *indexReader) readSymbols(off int) error {
 	return d.err()
 }
 
-// readOffsetTable reads an offset table at the given position and returns a map
+// readOffsetTable reads an offset table at the given position calls f for each
-// with the key strings concatenated by the 0xff unicode non-character.
+// found entry.f
-func (r *indexReader) readOffsetTable(off uint64) (map[string]uint32, error) {
+// If f returns an error it stops decoding and returns the received error,
-	const sep = "\xff"
+func (r *Reader) readOffsetTable(off uint64, f func([]string, uint32) error) error {
-
 	d := r.decbufAt(int(off))
 	cnt := d.be32()
 
-	res := make(map[string]uint32, cnt)
-
 	for d.err() == nil && d.len() > 0 && cnt > 0 {
 		keyCount := int(d.uvarint())
 		keys := make([]string, 0, keyCount)
@@ -769,18 +766,24 @@ func (r *indexReader) readOffsetTable(off uint64) (map[string]uint32, error) {
 		for i := 0; i < keyCount; i++ {
 			keys = append(keys, d.uvarintStr())
 		}
-		res[strings.Join(keys, sep)] = uint32(d.uvarint())
+		o := uint32(d.uvarint())
-
+		if d.err() != nil {
+			break
+		}
+		if err := f(keys, o); err != nil {
+			return err
+		}
 		cnt--
 	}
-	return res, d.err()
+	return d.err()
 }
 
-func (r *indexReader) Close() error {
+// Close the reader and its underlying resources.
+func (r *Reader) Close() error {
 	return r.c.Close()
 }
 
-func (r *indexReader) lookupSymbol(o uint32) (string, error) {
+func (r *Reader) lookupSymbol(o uint32) (string, error) {
 	s, ok := r.symbols[o]
 	if !ok {
 		return "", errors.Errorf("unknown symbol offset %d", o)
@@ -788,7 +791,8 @@ func (r *indexReader) lookupSymbol(o uint32) (string, error) {
 	return s, nil
 }
 
-func (r *indexReader) Symbols() (map[string]struct{}, error) {
+// Symbols returns a set of symbols that exist within the index.
+func (r *Reader) Symbols() (map[string]struct{}, error) {
 	res := make(map[string]struct{}, len(r.symbols))
 
 	for _, s := range r.symbols {
@@ -797,7 +801,13 @@ func (r *indexReader) Symbols() (map[string]struct{}, error) {
 	return res, nil
 }
 
-func (r *indexReader) LabelValues(names ...string) (StringTuples, error) {
+// SymbolTable returns the symbol table that is used to resolve symbol references.
+func (r *Reader) SymbolTable() map[uint32]string {
+	return r.symbols
+}
+
+// LabelValues returns value tuples that exist for the given label name tuples.
+func (r *Reader) LabelValues(names ...string) (StringTuples, error) {
 	const sep = "\xff"
 
 	key := strings.Join(names, sep)
@@ -830,7 +840,8 @@ type emptyStringTuples struct{}
 func (emptyStringTuples) At(i int) ([]string, error) { return nil, nil }
 func (emptyStringTuples) Len() int                   { return 0 }
 
-func (r *indexReader) LabelIndices() ([][]string, error) {
+// LabelIndices returns a for which labels or label tuples value indices exist.
+func (r *Reader) LabelIndices() ([][]string, error) {
 	const sep = "\xff"
 
 	res := [][]string{}
@@ -841,87 +852,38 @@ func (r *indexReader) LabelIndices() ([][]string, error) {
 	return res, nil
 }
 
-func (r *indexReader) Series(ref uint64, lbls *labels.Labels, chks *[]ChunkMeta) error {
+// Series the series with the given ID and writes its labels and chunks into lbls and chks.
-	d := r.decbufUvarintAt(int(ref))
+func (r *Reader) Series(id uint64, lbls *labels.Labels, chks *[]chunks.Meta) error {
-
+	d := r.decbufUvarintAt(int(id))
-	*lbls = (*lbls)[:0]
+	if d.err() != nil {
-	*chks = (*chks)[:0]
+		return d.err()
-
-	k := int(d.uvarint())
-
-	for i := 0; i < k; i++ {
-		lno := uint32(d.uvarint())
-		lvo := uint32(d.uvarint())
-
-		if d.err() != nil {
-			return errors.Wrap(d.err(), "read series label offsets")
-		}
-
-		ln, err := r.lookupSymbol(lno)
-		if err != nil {
-			return errors.Wrap(err, "lookup label name")
-		}
-		lv, err := r.lookupSymbol(lvo)
-		if err != nil {
-			return errors.Wrap(err, "lookup label value")
-		}
-
-		*lbls = append(*lbls, labels.Label{Name: ln, Value: lv})
 	}
-
+	return r.dec.Series(d.get(), lbls, chks)
-	// Read the chunks meta data.
-	k = int(d.uvarint())
-
-	if k == 0 {
-		return nil
-	}
-
-	t0 := d.varint64()
-	maxt := int64(d.uvarint64()) + t0
-	ref0 := int64(d.uvarint64())
-
-	*chks = append(*chks, ChunkMeta{
-		Ref:     uint64(ref0),
-		MinTime: t0,
-		MaxTime: maxt,
-	})
-	t0 = maxt
-
-	for i := 1; i < k; i++ {
-		mint := int64(d.uvarint64()) + t0
-		maxt := int64(d.uvarint64()) + mint
-
-		ref0 += d.varint64()
-		t0 = maxt
-
-		if d.err() != nil {
-			return errors.Wrapf(d.err(), "read meta for chunk %d", i)
-		}
-
-		*chks = append(*chks, ChunkMeta{
-			Ref:     uint64(ref0),
-			MinTime: mint,
-			MaxTime: maxt,
-		})
-	}
-	return d.err()
 }
 
-func (r *indexReader) Postings(name, value string) (Postings, error) {
+// Postings returns a postings list for the given label pair.
-	const sep = "\xff"
+func (r *Reader) Postings(name, value string) (Postings, error) {
-	key := strings.Join([]string{name, value}, sep)
+	off, ok := r.postings[labels.Label{
-
+		Name:  name,
-	off, ok := r.postings[key]
+		Value: value,
+	}]
 	if !ok {
-		return emptyPostings, nil
+		return EmptyPostings(), nil
 	}
 	d := r.decbufAt(int(off))
-	d.be32() // consume unused postings list length.
+	if d.err() != nil {
-
+		return nil, errors.Wrap(d.err(), "get postings entry")
-	return newBigEndianPostings(d.get()), errors.Wrap(d.err(), "get postings bytes")
+	}
+	_, p, err := r.dec.Postings(d.get())
+	if err != nil {
+		return nil, errors.Wrap(err, "decode postings")
+	}
+	return p, nil
 }
 
-func (r *indexReader) SortedPostings(p Postings) Postings {
+// SortedPostings returns the given postings list reordered so that the backing series
+// are sorted.
+func (r *Reader) SortedPostings(p Postings) Postings {
 	return p
 }
 
@@ -930,7 +892,7 @@ type stringTuples struct {
 	s []string // flattened tuple entries
 }
 
-func newStringTuples(s []string, l int) (*stringTuples, error) {
+func NewStringTuples(s []string, l int) (*stringTuples, error) {
 	if len(s)%l != 0 {
 		return nil, errors.Wrap(errInvalidSize, "string tuple list")
 	}
@@ -992,3 +954,100 @@ func (t *serializedStringTuples) At(i int) ([]string, error) {
 
 	return res, nil
 }
+
+// DecoderV1 provides decoding methods for the v1 index file format.
+//
+// It currently does not contain decoding methods for all entry types but can be extended
+// by them if there's demand.
+type DecoderV1 struct {
+	symbols map[uint32]string
+}
+
+func (dec *DecoderV1) lookupSymbol(o uint32) (string, error) {
+	s, ok := dec.symbols[o]
+	if !ok {
+		return "", errors.Errorf("unknown symbol offset %d", o)
+	}
+	return s, nil
+}
+
+// SetSymbolTable set the symbol table to be used for lookups when decoding series
+// and label indices
+func (dec *DecoderV1) SetSymbolTable(t map[uint32]string) {
+	dec.symbols = t
+}
+
+// Postings returns a postings list for b and its number of elements.
+func (dec *DecoderV1) Postings(b []byte) (int, Postings, error) {
+	d := decbuf{b: b}
+	n := d.be32int()
+	l := d.get()
+	return n, newBigEndianPostings(l), d.err()
+}
+
+// Series decodes a series entry from the given byte slice into lset and chks.
+func (dec *DecoderV1) Series(b []byte, lbls *labels.Labels, chks *[]chunks.Meta) error {
+	*lbls = (*lbls)[:0]
+	*chks = (*chks)[:0]
+
+	d := decbuf{b: b}
+
+	k := int(d.uvarint())
+
+	for i := 0; i < k; i++ {
+		lno := uint32(d.uvarint())
+		lvo := uint32(d.uvarint())
+
+		if d.err() != nil {
+			return errors.Wrap(d.err(), "read series label offsets")
+		}
+
+		ln, err := dec.lookupSymbol(lno)
+		if err != nil {
+			return errors.Wrap(err, "lookup label name")
+		}
+		lv, err := dec.lookupSymbol(lvo)
+		if err != nil {
+			return errors.Wrap(err, "lookup label value")
+		}
+
+		*lbls = append(*lbls, labels.Label{Name: ln, Value: lv})
+	}
+
+	// Read the chunks meta data.
+	k = int(d.uvarint())
+
+	if k == 0 {
+		return nil
+	}
+
+	t0 := d.varint64()
+	maxt := int64(d.uvarint64()) + t0
+	ref0 := int64(d.uvarint64())
+
+	*chks = append(*chks, chunks.Meta{
+		Ref:     uint64(ref0),
+		MinTime: t0,
+		MaxTime: maxt,
+	})
+	t0 = maxt
+
+	for i := 1; i < k; i++ {
+		mint := int64(d.uvarint64()) + t0
+		maxt := int64(d.uvarint64()) + mint
+
+		ref0 += d.varint64()
+		t0 = maxt
+
+		if d.err() != nil {
+			return errors.Wrapf(d.err(), "read meta for chunk %d", i)
+		}
+
+		*chks = append(*chks, chunks.Meta{
+			Ref:     uint64(ref0),
+			MinTime: mint,
+			MaxTime: maxt,
+		})
+	}
+	return d.err()
+}

vendor/github.com/prometheus/tsdb/index/postings.go

@@ -11,7 +11,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package tsdb
+package index
 
 import (
 	"encoding/binary"
@@ -23,35 +23,42 @@ import (
 	"github.com/prometheus/tsdb/labels"
 )
 
-// memPostings holds postings list for series ID per label pair. They may be written
+var allPostingsKey = labels.Label{}
+
+// AllPostingsKey returns the label key that is used to store the postings list of all existing IDs.
+func AllPostingsKey() (name, value string) {
+	return allPostingsKey.Name, allPostingsKey.Value
+}
+
+// MemPostings holds postings list for series ID per label pair. They may be written
 // to out of order.
 // ensureOrder() must be called once before any reads are done. This allows for quick
 // unordered batch fills on startup.
-type memPostings struct {
+type MemPostings struct {
 	mtx     sync.RWMutex
 	m       map[labels.Label][]uint64
 	ordered bool
 }
 
-// newMemPoistings returns a memPostings that's ready for reads and writes.
+// NewMemPostings returns a memPostings that's ready for reads and writes.
-func newMemPostings() *memPostings {
+func NewMemPostings() *MemPostings {
-	return &memPostings{
+	return &MemPostings{
 		m:       make(map[labels.Label][]uint64, 512),
 		ordered: true,
 	}
 }
 
-// newUnorderedMemPostings returns a memPostings that is not safe to be read from
+// NewUnorderedMemPostings returns a memPostings that is not safe to be read from
 // until ensureOrder was called once.
-func newUnorderedMemPostings() *memPostings {
+func NewUnorderedMemPostings() *MemPostings {
-	return &memPostings{
+	return &MemPostings{
 		m:       make(map[labels.Label][]uint64, 512),
 		ordered: false,
 	}
 }
 
-// sortedKeys returns a list of sorted label keys of the postings.
+// SortedKeys returns a list of sorted label keys of the postings.
-func (p *memPostings) sortedKeys() []labels.Label {
+func (p *MemPostings) SortedKeys() []labels.Label {
 	p.mtx.RLock()
 	keys := make([]labels.Label, 0, len(p.m))
 
@@ -69,23 +76,26 @@ func (p *memPostings) sortedKeys() []labels.Label {
 	return keys
 }
 
-// Postings returns an iterator over the postings list for s.
+// Get returns a postings list for the given label pair.
-func (p *memPostings) get(name, value string) Postings {
+func (p *MemPostings) Get(name, value string) Postings {
 	p.mtx.RLock()
 	l := p.m[labels.Label{Name: name, Value: value}]
 	p.mtx.RUnlock()
 
 	if l == nil {
-		return emptyPostings
+		return EmptyPostings()
 	}
 	return newListPostings(l)
 }
 
-var allPostingsKey = labels.Label{}
+// All returns a postings list over all documents ever added.
+func (p *MemPostings) All() Postings {
+	return p.Get(AllPostingsKey())
+}
 
-// ensurePostings ensures that all postings lists are sorted. After it returns all further
+// EnsureOrder ensures that all postings lists are sorted. After it returns all further
 // calls to add and addFor will insert new IDs in a sorted manner.
-func (p *memPostings) ensureOrder() {
+func (p *MemPostings) EnsureOrder() {
 	p.mtx.Lock()
 	defer p.mtx.Unlock()
 
@@ -117,9 +127,66 @@ func (p *memPostings) ensureOrder() {
 	p.ordered = true
 }
 
-// add adds a document to the index. The caller has to ensure that no
+// Delete removes all ids in the given map from the postings lists.
-// term argument appears twice.
+func (p *MemPostings) Delete(deleted map[uint64]struct{}) {
-func (p *memPostings) add(id uint64, lset labels.Labels) {
+	var keys []labels.Label
+
+	// Collect all keys relevant for deletion once. New keys added afterwards
+	// can by definition not be affected by any of the given deletes.
+	p.mtx.RLock()
+	for l := range p.m {
+		keys = append(keys, l)
+	}
+	p.mtx.RUnlock()
+
+	// For each key we first analyse whether the postings list is affected by the deletes.
+	// If yes, we actually reallocate a new postings list.
+	for _, l := range keys {
+		// Only lock for processing one postings list so we don't block reads for too long.
+		p.mtx.Lock()
+
+		found := false
+		for _, id := range p.m[l] {
+			if _, ok := deleted[id]; ok {
+				found = true
+				break
+			}
+		}
+		if !found {
+			p.mtx.Unlock()
+			continue
+		}
+		repl := make([]uint64, 0, len(p.m[l]))
+
+		for _, id := range p.m[l] {
+			if _, ok := deleted[id]; !ok {
+				repl = append(repl, id)
+			}
+		}
+		if len(repl) > 0 {
+			p.m[l] = repl
+		} else {
+			delete(p.m, l)
+		}
+		p.mtx.Unlock()
+	}
+}
+
+// Iter calls f for each postings list. It aborts if f returns an error and returns it.
+func (p *MemPostings) Iter(f func(labels.Label, Postings) error) error {
+	p.mtx.RLock()
+	defer p.mtx.RUnlock()
+
+	for l, p := range p.m {
+		if err := f(l, newListPostings(p)); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// Add a label set to the postings index.
+func (p *MemPostings) Add(id uint64, lset labels.Labels) {
 	p.mtx.Lock()
 
 	for _, l := range lset {
@@ -130,7 +197,7 @@ func (p *memPostings) add(id uint64, lset labels.Labels) {
 	p.mtx.Unlock()
 }
 
-func (p *memPostings) addFor(id uint64, l labels.Label) {
+func (p *MemPostings) addFor(id uint64, l labels.Label) {
 	list := append(p.m[l], id)
 	p.m[l] = list
 
@@ -149,7 +216,8 @@ func (p *memPostings) addFor(id uint64, l labels.Label) {
 	}
 }
 
-func expandPostings(p Postings) (res []uint64, err error) {
+// ExpandPostings returns the postings expanded as a slice.
+func ExpandPostings(p Postings) (res []uint64, err error) {
 	for p.Next() {
 		res = append(res, p.At())
 	}
@@ -189,6 +257,11 @@ func EmptyPostings() Postings {
 	return emptyPostings
 }
 
+// ErrPostings returns new postings that immediately error.
+func ErrPostings(err error) Postings {
+	return errPostings{err}
+}
+
 // Intersect returns a new postings list over the intersection of the
 // input postings.
 func Intersect(its ...Postings) Postings {
@@ -259,7 +332,7 @@ func (it *intersectPostings) Err() error {
 // Merge returns a new iterator over the union of the input iterators.
 func Merge(its ...Postings) Postings {
 	if len(its) == 0 {
-		return nil
+		return EmptyPostings()
 	}
 	if len(its) == 1 {
 		return its[0]
@@ -340,12 +413,96 @@ func (it *mergedPostings) Err() error {
 	return it.b.Err()
 }
 
+// Without returns a new postings list that contains all elements from the full list that
+// are not in the drop list
+func Without(full, drop Postings) Postings {
+	return newRemovedPostings(full, drop)
+}
+
+type removedPostings struct {
+	full, remove Postings
+
+	cur uint64
+
+	initialized bool
+	fok, rok    bool
+}
+
+func newRemovedPostings(full, remove Postings) *removedPostings {
+	return &removedPostings{
+		full:   full,
+		remove: remove,
+	}
+}
+
+func (rp *removedPostings) At() uint64 {
+	return rp.cur
+}
+
+func (rp *removedPostings) Next() bool {
+	if !rp.initialized {
+		rp.fok = rp.full.Next()
+		rp.rok = rp.remove.Next()
+		rp.initialized = true
+	}
+
+	if !rp.fok {
+		return false
+	}
+
+	if !rp.rok {
+		rp.cur = rp.full.At()
+		rp.fok = rp.full.Next()
+		return true
+	}
+
+	fcur, rcur := rp.full.At(), rp.remove.At()
+	if fcur < rcur {
+		rp.cur = fcur
+		rp.fok = rp.full.Next()
+
+		return true
+	} else if rcur < fcur {
+		// Forward the remove postings to the right position.
+		rp.rok = rp.remove.Seek(fcur)
+	} else {
+		// Skip the current posting.
+		rp.fok = rp.full.Next()
+	}
+
+	return rp.Next()
+}
+
+func (rp *removedPostings) Seek(id uint64) bool {
+	if rp.cur >= id {
+		return true
+	}
+
+	rp.fok = rp.full.Seek(id)
+	rp.rok = rp.remove.Seek(id)
+	rp.initialized = true
+
+	return rp.Next()
+}
+
+func (rp *removedPostings) Err() error {
+	if rp.full.Err() != nil {
+		return rp.full.Err()
+	}
+
+	return rp.remove.Err()
+}
+
 // listPostings implements the Postings interface over a plain list.
 type listPostings struct {
 	list []uint64
 	cur  uint64
 }
 
+func NewListPostings(list []uint64) Postings {
+	return newListPostings(list)
+}
+
 func newListPostings(list []uint64) *listPostings {
 	return &listPostings{list: list}
 }
@@ -434,27 +591,3 @@ func (it *bigEndianPostings) Seek(x uint64) bool {
 func (it *bigEndianPostings) Err() error {
 	return nil
 }
-
-type stringset map[string]struct{}
-
-func (ss stringset) set(s string) {
-	ss[s] = struct{}{}
-}
-
-func (ss stringset) has(s string) bool {
-	_, ok := ss[s]
-	return ok
-}
-
-func (ss stringset) String() string {
-	return strings.Join(ss.slice(), ",")
-}
-
-func (ss stringset) slice() []string {
-	slice := make([]string, 0, len(ss))
-	for k := range ss {
-		slice = append(slice, k)
-	}
-	sort.Strings(slice)
-	return slice
-}

vendor/github.com/prometheus/tsdb/labels/labels.go

@@ -14,12 +14,15 @@
 package labels
 
 import (
+	"bufio"
 	"bytes"
+	"os"
 	"sort"
 	"strconv"
 	"strings"
 
 	"github.com/cespare/xxhash"
+	"github.com/pkg/errors"
 )
 
 const sep = '\xff'
@@ -161,3 +164,49 @@ type Slice []Labels
 func (s Slice) Len() int           { return len(s) }
 func (s Slice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
 func (s Slice) Less(i, j int) bool { return Compare(s[i], s[j]) < 0 }
+
+// ReadLabels reads up to n label sets in a JSON formatted file fn. It is mostly useful
+// to load testing data.
+func ReadLabels(fn string, n int) ([]Labels, error) {
+	f, err := os.Open(fn)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	scanner := bufio.NewScanner(f)
+
+	var mets []Labels
+	hashes := map[uint64]struct{}{}
+	i := 0
+
+	for scanner.Scan() && i < n {
+		m := make(Labels, 0, 10)
+
+		r := strings.NewReplacer("\"", "", "{", "", "}", "")
+		s := r.Replace(scanner.Text())
+
+		labelChunks := strings.Split(s, ",")
+		for _, labelChunk := range labelChunks {
+			split := strings.Split(labelChunk, ":")
+			m = append(m, Label{Name: split[0], Value: split[1]})
+		}
+		// Order of the k/v labels matters, don't assume we'll always receive them already sorted.
+		sort.Sort(m)
+
+		h := m.Hash()
+		if _, ok := hashes[h]; ok {
+			continue
+		}
+		mets = append(mets, m)
+		hashes[h] = struct{}{}
+		i++
+	}
+	if err != nil {
+		return nil, err
+	}
+	if i != n {
+		return mets, errors.Errorf("requested %d metrics but found %d", n, i)
+	}
+	return mets, nil
+}

vendor/github.com/prometheus/tsdb/labels/selector.go

@@ -76,6 +76,18 @@ func NewRegexpMatcher(name, pattern string) (Matcher, error) {
 	return &regexpMatcher{name: name, re: re}, nil
 }
 
+// NewRegexpMatcher returns a new matcher verifying that a value matches
+// the regular expression pattern. Will panic if the pattern is not a valid
+// regular expression.
+func NewMustRegexpMatcher(name, pattern string) Matcher {
+	re, err := regexp.Compile(pattern)
+	if err != nil {
+		panic(err)
+	}
+	return &regexpMatcher{name: name, re: re}
+
+}
+
 // notMatcher inverts the matching result for a matcher.
 type notMatcher struct {
 	Matcher

vendor/github.com/prometheus/tsdb/querier.go

@@ -19,7 +19,9 @@ import (
 	"strings"
 
 	"github.com/pkg/errors"
+	"github.com/prometheus/tsdb/chunkenc"
 	"github.com/prometheus/tsdb/chunks"
+	"github.com/prometheus/tsdb/index"
 	"github.com/prometheus/tsdb/labels"
 )
 
@@ -202,25 +204,17 @@ func (q *blockQuerier) Close() error {
 // PostingsForMatchers assembles a single postings iterator against the index reader
 // based on the given matchers. It returns a list of label names that must be manually
 // checked to not exist in series the postings list points to.
-func PostingsForMatchers(index IndexReader, ms ...labels.Matcher) (Postings, []string, error) {
+func PostingsForMatchers(ix IndexReader, ms ...labels.Matcher) (index.Postings, error) {
-	var (
+	var its []index.Postings
-		its    []Postings
+
-		absent []string
-	)
 	for _, m := range ms {
-		// If the matcher checks absence of a label, don't select them
+		it, err := postingsForMatcher(ix, m)
-		// but propagate the check into the series set.
-		if _, ok := m.(*labels.EqualMatcher); ok && m.Matches("") {
-			absent = append(absent, m.Name())
-			continue
-		}
-		it, err := postingsForMatcher(index, m)
 		if err != nil {
-			return nil, nil, err
+			return nil, err
 		}
 		its = append(its, it)
 	}
-	return index.SortedPostings(Intersect(its...)), absent, nil
+	return ix.SortedPostings(index.Intersect(its...)), nil
 }
 
 // tuplesByPrefix uses binary search to find prefix matches within ts.
@@ -254,17 +248,24 @@ func tuplesByPrefix(m *labels.PrefixMatcher, ts StringTuples) ([]string, error)
 	return matches, nil
 }
 
-func postingsForMatcher(index IndexReader, m labels.Matcher) (Postings, error) {
+func postingsForMatcher(ix IndexReader, m labels.Matcher) (index.Postings, error) {
+	// If the matcher selects an empty value, it selects all the series which dont
+	// have the label name set too. See: https://github.com/prometheus/prometheus/issues/3575
+	// and https://github.com/prometheus/prometheus/pull/3578#issuecomment-351653555
+	if m.Matches("") {
+		return postingsForUnsetLabelMatcher(ix, m)
+	}
+
 	// Fast-path for equal matching.
 	if em, ok := m.(*labels.EqualMatcher); ok {
-		it, err := index.Postings(em.Name(), em.Value())
+		it, err := ix.Postings(em.Name(), em.Value())
 		if err != nil {
 			return nil, err
 		}
 		return it, nil
 	}
 
-	tpls, err := index.LabelValues(m.Name())
+	tpls, err := ix.LabelValues(m.Name())
 	if err != nil {
 		return nil, err
 	}
@@ -289,20 +290,55 @@ func postingsForMatcher(index IndexReader, m labels.Matcher) (Postings, error) {
 	}
 
 	if len(res) == 0 {
-		return EmptyPostings(), nil
+		return index.EmptyPostings(), nil
 	}
 
-	var rit []Postings
+	var rit []index.Postings
 
 	for _, v := range res {
-		it, err := index.Postings(m.Name(), v)
+		it, err := ix.Postings(m.Name(), v)
 		if err != nil {
 			return nil, err
 		}
 		rit = append(rit, it)
 	}
 
-	return Merge(rit...), nil
+	return index.Merge(rit...), nil
+}
+
+func postingsForUnsetLabelMatcher(ix IndexReader, m labels.Matcher) (index.Postings, error) {
+	tpls, err := ix.LabelValues(m.Name())
+	if err != nil {
+		return nil, err
+	}
+
+	var res []string
+	for i := 0; i < tpls.Len(); i++ {
+		vals, err := tpls.At(i)
+		if err != nil {
+			return nil, err
+		}
+
+		if !m.Matches(vals[0]) {
+			res = append(res, vals[0])
+		}
+	}
+
+	var rit []index.Postings
+	for _, v := range res {
+		it, err := ix.Postings(m.Name(), v)
+		if err != nil {
+			return nil, err
+		}
+
+		rit = append(rit, it)
+	}
+
+	allPostings, err := ix.Postings(index.AllPostingsKey())
+	if err != nil {
+		return nil, err
+	}
+	return index.Without(allPostings, index.Merge(rit...)), nil
 }
 
 func mergeStrings(a, b []string) []string {
@@ -417,22 +453,23 @@ func (s *mergedSeriesSet) Next() bool {
 	return true
 }
 
+// ChunkSeriesSet exposes the chunks and intervals of a series instead of the
+// actual series itself.
 type ChunkSeriesSet interface {
 	Next() bool
-	At() (labels.Labels, []ChunkMeta, Intervals)
+	At() (labels.Labels, []chunks.Meta, Intervals)
 	Err() error
 }
 
 // baseChunkSeries loads the label set and chunk references for a postings
 // list from an index. It filters out series that have labels set that should be unset.
 type baseChunkSeries struct {
-	p          Postings
+	p          index.Postings
 	index      IndexReader
 	tombstones TombstoneReader
-	absent     []string // labels that must be unset in results.
 
 	lset      labels.Labels
-	chks      []ChunkMeta
+	chks      []chunks.Meta
 	intervals Intervals
 	err       error
 }
@@ -443,7 +480,7 @@ func LookupChunkSeries(ir IndexReader, tr TombstoneReader, ms ...labels.Matcher)
 	if tr == nil {
 		tr = EmptyTombstoneReader()
 	}
-	p, absent, err := PostingsForMatchers(ir, ms...)
+	p, err := PostingsForMatchers(ir, ms...)
 	if err != nil {
 		return nil, err
 	}
@@ -451,11 +488,10 @@ func LookupChunkSeries(ir IndexReader, tr TombstoneReader, ms ...labels.Matcher)
 		p:          p,
 		index:      ir,
 		tombstones: tr,
-		absent:     absent,
 	}, nil
 }
 
-func (s *baseChunkSeries) At() (labels.Labels, []ChunkMeta, Intervals) {
+func (s *baseChunkSeries) At() (labels.Labels, []chunks.Meta, Intervals) {
 	return s.lset, s.chks, s.intervals
 }
 
@@ -463,14 +499,14 @@ func (s *baseChunkSeries) Err() error { return s.err }
 
 func (s *baseChunkSeries) Next() bool {
 	var (
-		lset   labels.Labels
+		lset     labels.Labels
-		chunks []ChunkMeta
+		chkMetas []chunks.Meta
-		err    error
+		err      error
 	)
-Outer:
+
 	for s.p.Next() {
 		ref := s.p.At()
-		if err := s.index.Series(ref, &lset, &chunks); err != nil {
+		if err := s.index.Series(ref, &lset, &chkMetas); err != nil {
 			// Postings may be stale. Skip if no underlying series exists.
 			if errors.Cause(err) == ErrNotFound {
 				continue
@@ -479,15 +515,8 @@ Outer:
 			return false
 		}
 
-		// If a series contains a label that must be absent, it is skipped as well.
-		for _, abs := range s.absent {
-			if lset.Get(abs) != "" {
-				continue Outer
-			}
-		}
-
 		s.lset = lset
-		s.chks = chunks
+		s.chks = chkMetas
 		s.intervals, err = s.tombstones.Get(s.p.At())
 		if err != nil {
 			s.err = errors.Wrap(err, "get tombstones")
@@ -496,7 +525,7 @@ Outer:
 
 		if len(s.intervals) > 0 {
 			// Only those chunks that are not entirely deleted.
-			chks := make([]ChunkMeta, 0, len(s.chks))
+			chks := make([]chunks.Meta, 0, len(s.chks))
 			for _, chk := range s.chks {
 				if !(Interval{chk.MinTime, chk.MaxTime}.isSubrange(s.intervals)) {
 					chks = append(chks, chk)
@@ -523,14 +552,15 @@ type populatedChunkSeries struct {
 	mint, maxt int64
 
 	err       error
-	chks      []ChunkMeta
+	chks      []chunks.Meta
 	lset      labels.Labels
 	intervals Intervals
 }
 
-func (s *populatedChunkSeries) At() (labels.Labels, []ChunkMeta, Intervals) {
+func (s *populatedChunkSeries) At() (labels.Labels, []chunks.Meta, Intervals) {
 	return s.lset, s.chks, s.intervals
 }
+
 func (s *populatedChunkSeries) Err() error { return s.err }
 
 func (s *populatedChunkSeries) Next() bool {
@@ -544,19 +574,31 @@ func (s *populatedChunkSeries) Next() bool {
 			chks = chks[1:]
 		}
 
-		for i := range chks {
+		// This is to delete in place while iterating.
-			c := &chks[i]
+		for i, rlen := 0, len(chks); i < rlen; i++ {
+			j := i - (rlen - len(chks))
+			c := &chks[j]
 
 			// Break out at the first chunk that has no overlap with mint, maxt.
 			if c.MinTime > s.maxt {
-				chks = chks[:i]
+				chks = chks[:j]
 				break
 			}
+
 			c.Chunk, s.err = s.chunks.Chunk(c.Ref)
 			if s.err != nil {
+				// This means that the chunk has be garbage collected. Remove it from the list.
+				if s.err == ErrNotFound {
+					s.err = nil
+
+					// Delete in-place.
+					chks = append(chks[:j], chks[j+1:]...)
+				}
+
 				return false
 			}
 		}
+
 		if len(chks) == 0 {
 			continue
 		}
@@ -608,7 +650,7 @@ func (s *blockSeriesSet) Err() error { return s.err }
 // time series data.
 type chunkSeries struct {
 	labels labels.Labels
-	chunks []ChunkMeta // in-order chunk refs
+	chunks []chunks.Meta // in-order chunk refs
 
 	mint, maxt int64
 
@@ -711,17 +753,17 @@ func (it *chainedSeriesIterator) Err() error {
 // chunkSeriesIterator implements a series iterator on top
 // of a list of time-sorted, non-overlapping chunks.
 type chunkSeriesIterator struct {
-	chunks []ChunkMeta
+	chunks []chunks.Meta
 
 	i   int
-	cur chunks.Iterator
+	cur chunkenc.Iterator
 
 	maxt, mint int64
 
 	intervals Intervals
 }
 
-func newChunkSeriesIterator(cs []ChunkMeta, dranges Intervals, mint, maxt int64) *chunkSeriesIterator {
+func newChunkSeriesIterator(cs []chunks.Meta, dranges Intervals, mint, maxt int64) *chunkSeriesIterator {
 	it := cs[0].Chunk.Iterator()
 
 	if len(dranges) > 0 {
@@ -810,6 +852,46 @@ func (it *chunkSeriesIterator) Err() error {
 	return it.cur.Err()
 }
 
+// deletedIterator wraps an Iterator and makes sure any deleted metrics are not
+// returned.
+type deletedIterator struct {
+	it chunkenc.Iterator
+
+	intervals Intervals
+}
+
+func (it *deletedIterator) At() (int64, float64) {
+	return it.it.At()
+}
+
+func (it *deletedIterator) Next() bool {
+Outer:
+	for it.it.Next() {
+		ts, _ := it.it.At()
+
+		for _, tr := range it.intervals {
+			if tr.inBounds(ts) {
+				continue Outer
+			}
+
+			if ts > tr.Maxt {
+				it.intervals = it.intervals[1:]
+				continue
+			}
+
+			return true
+		}
+
+		return true
+	}
+
+	return false
+}
+
+func (it *deletedIterator) Err() error {
+	return it.it.Err()
+}
+
 type mockSeriesSet struct {
 	next   func() bool
 	series func() Series

vendor/github.com/prometheus/tsdb/tombstones.go

@@ -109,7 +109,9 @@ type Stone struct {
 
 func readTombstones(dir string) (memTombstones, error) {
 	b, err := ioutil.ReadFile(filepath.Join(dir, tombstoneFilename))
-	if err != nil {
+	if os.IsNotExist(err) {
+		return memTombstones{}, nil
+	} else if err != nil {
 		return nil, err
 	}
 

vendor/vendor.json

@@ -794,28 +794,40 @@
 			"revisionTime": "2016-04-11T19:08:41Z"
 		},
 		{
-			"checksumSHA1": "XgGOJ06okm8qd+x/gdRj48RgXsg=",
+			"checksumSHA1": "C72p7MMwA94LrpqYD3dMcQEKHzY=",
 			"path": "github.com/prometheus/tsdb",
-			"revision": "30bbbe34f8787df072cf04563bc98fb8094ba070",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
-			"revisionTime": "2017-11-30T09:58:01Z"
+			"revisionTime": "2017-12-22T14:44:41Z"
 		},
 		{
-			"checksumSHA1": "C5V8KPHm/gZF0qrNwmIEDdG6rhA=",
+			"checksumSHA1": "XTirmk6Pq5TBGIZEaN5VL4k3i1s=",
+			"path": "github.com/prometheus/tsdb/chunkenc",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
+			"revisionTime": "2017-12-22T14:44:41Z"
+		},
+		{
+			"checksumSHA1": "+zsn1i8cqwgZXL8Bg6jDy32xjAo=",
 			"path": "github.com/prometheus/tsdb/chunks",
-			"revision": "30bbbe34f8787df072cf04563bc98fb8094ba070",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
-			"revisionTime": "2017-11-30T09:58:01Z"
+			"revisionTime": "2017-12-22T14:44:41Z"
 		},
 		{
-			"checksumSHA1": "7RhNAVcmDmLFqn9nWiudT0B76f8=",
+			"checksumSHA1": "h49AAcJ5+iRBwCgbfQf+2T1E1ZE=",
 			"path": "github.com/prometheus/tsdb/fileutil",
-			"revision": "30bbbe34f8787df072cf04563bc98fb8094ba070",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
-			"revisionTime": "2017-11-30T09:58:01Z"
+			"revisionTime": "2017-12-22T14:44:41Z"
 		},
 		{
-			"checksumSHA1": "zhmlvc322RH1L3l9DaA9d/HVVWs=",
+			"checksumSHA1": "y3lUn43gcc3HtYd20UDujyybGq4=",
+			"path": "github.com/prometheus/tsdb/index",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
+			"revisionTime": "2017-12-22T14:44:41Z"
+		},
+		{
+			"checksumSHA1": "Va8HWvOFTwFeewZFadMAOzNGDps=",
 			"path": "github.com/prometheus/tsdb/labels",
-			"revision": "30bbbe34f8787df072cf04563bc98fb8094ba070",
+			"revision": "07ef80820ef1250db82f9544f3fcf7f0f63ccee0",
-			"revisionTime": "2017-11-30T09:58:01Z"
+			"revisionTime": "2017-12-22T14:44:41Z"
 		},
 		{
 			"checksumSHA1": "5SYLEhADhdBVZAGPVHWggQl7H8k=",