diff --git a/db.go b/db.go
index d9382364c..7ca513ea3 100644
--- a/db.go
+++ b/db.go
@@ -290,7 +290,10 @@ func (s *Shard) appendBatch(samples []hashedSample) error {
 	// TODO(fabxc): distinguish samples between concurrent heads for
 	// different time blocks. Those may occurr during transition to still
 	// allow late samples to arrive for a previous block.
-	err := s.head.appendBatch(samples, s.metrics.samplesAppended)
+	err := s.head.appendBatch(samples)
+	if err != nil {
+		s.metrics.samplesAppended.Add(float64(len(samples)))
+	}
 
 	// TODO(fabxc): randomize over time and use better scoring function.
 	if s.head.stats.SampleCount/(uint64(s.head.stats.ChunkCount)+1) > 400 {
@@ -401,6 +404,7 @@ func (s *Shard) persist() error {
 
 // chunkDesc wraps a plain data chunk and provides cached meta data about it.
 type chunkDesc struct {
+	ref   uint32
 	lset  labels.Labels
 	chunk chunks.Chunk
 
diff --git a/head.go b/head.go
index a58dd9af4..b5e4f063a 100644
--- a/head.go
+++ b/head.go
@@ -7,7 +7,6 @@ import (
 
 	"github.com/fabxc/tsdb/chunks"
 	"github.com/fabxc/tsdb/labels"
-	"github.com/prometheus/client_golang/prometheus"
 )
 
 // HeadBlock handles reads and writes of time series data within a time window.
@@ -18,8 +17,8 @@ type HeadBlock struct {
 	// is assigned the index as its ID.
 	descs []*chunkDesc
 	// hashes contains a collision map of label set hashes of chunks
-	// to their position in the chunk desc slice.
-	hashes map[uint64][]int
+	// to their chunk descs.
+	hashes map[uint64][]*chunkDesc
 
 	values   map[string]stringset // label names to possible values
 	postings *memPostings         // postings lists for terms
@@ -38,7 +37,7 @@ func OpenHeadBlock(dir string, baseTime int64) (*HeadBlock, error) {
 
 	b := &HeadBlock{
 		descs:    []*chunkDesc{},
-		hashes:   map[uint64][]int{},
+		hashes:   map[uint64][]*chunkDesc{},
 		values:   map[string]stringset{},
 		postings: &memPostings{m: make(map[term][]uint32)},
 		wal:      wal,
@@ -136,15 +135,15 @@ func (h *HeadBlock) Series(ref uint32, mint, maxt int64) (Series, error) {
 
 // get retrieves the chunk with the hash and label set and creates
 // a new one if it doesn't exist yet.
-func (h *HeadBlock) get(hash uint64, lset labels.Labels) (*chunkDesc, uint32) {
-	refs := h.hashes[hash]
+func (h *HeadBlock) get(hash uint64, lset labels.Labels) *chunkDesc {
+	cds := h.hashes[hash]
 
-	for _, ref := range refs {
-		if cd := h.descs[ref]; cd.lset.Equals(lset) {
-			return cd, uint32(ref)
+	for _, cd := range cds {
+		if cd.lset.Equals(lset) {
+			return cd
 		}
 	}
-	return nil, 0
+	return nil
 }
 
 func (h *HeadBlock) create(hash uint64, lset labels.Labels) *chunkDesc {
@@ -160,10 +159,10 @@ func (h *HeadBlock) create(hash uint64, lset labels.Labels) *chunkDesc {
 		panic(err)
 	}
 	// Index the new chunk.
-	ref := len(h.descs)
+	cd.ref = uint32(len(h.descs))
 
 	h.descs = append(h.descs, cd)
-	h.hashes[hash] = append(h.hashes[hash], ref)
+	h.hashes[hash] = append(h.hashes[hash], cd)
 
 	// Add each label pair as a term to the inverted index.
 	terms := make([]term, 0, len(lset))
@@ -178,7 +177,7 @@ func (h *HeadBlock) create(hash uint64, lset labels.Labels) *chunkDesc {
 		}
 		valset.set(l.Value)
 	}
-	h.postings.add(uint32(ref), terms...)
+	h.postings.add(cd.ref, terms...)
 
 	// For the head block there's exactly one chunk per series.
 	h.stats.ChunkCount++
@@ -187,7 +186,7 @@ func (h *HeadBlock) create(hash uint64, lset labels.Labels) *chunkDesc {
 	return cd
 }
 
-func (h *HeadBlock) appendBatch(samples []hashedSample, appended prometheus.Counter) error {
+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.
 	var (
@@ -199,11 +198,11 @@ func (h *HeadBlock) appendBatch(samples []hashedSample, appended prometheus.Coun
 	for i := range samples {
 		s := &samples[i]
 
-		cd, ref := h.get(s.hash, s.labels)
+		cd := h.get(s.hash, s.labels)
 		if cd != nil {
 			// TODO(fabxc): sample refs are only scoped within a block for
 			// now and we ignore any previously set value
-			s.ref = ref
+			s.ref = cd.ref
 			continue
 		}
 
@@ -226,6 +225,8 @@ func (h *HeadBlock) appendBatch(samples []hashedSample, appended prometheus.Coun
 		return err
 	}
 
+	// After the samples were successfully written to the WAL, there may
+	// be no further failures.
 	for i, s := range newSeries {
 		h.create(newHashes[i], s)
 	}
diff --git a/postings.go b/postings.go
index 2f4d10975..9dde3e678 100644
--- a/postings.go
+++ b/postings.go
@@ -15,7 +15,11 @@ type term struct {
 
 // Postings returns an iterator over the postings list for s.
 func (p *memPostings) get(t term) Postings {
-	return &listPostings{list: p.m[t], idx: -1}
+	l := p.m[t]
+	if l == nil {
+		return emptyPostings
+	}
+	return &listPostings{list: l, idx: -1}
 }
 
 // add adds a document to the index. The caller has to ensure that no