Stream symbols during compaction. (prometheus#6468)

Rather than buffer up symbols in RAM, do it one by one during compaction. Then use the reader's symbol handling for symbol lookups during the rest of the index write. There is some slowdown in compaction, due to having to look through a file rather than a hash lookup. This is noise to the overall cost of compacting series with thousands of samples though. benchmark old ns/op new ns/op delta BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 539917175 675341565 +25.08% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 2441815993 2477453524 +1.46% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 3978543559 3922909687 -1.40% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 8430219716 8586610007 +1.86% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 1786424591 1909552782 +6.89% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 5328998202 6020839950 +12.98% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 10085059958 11085278690 +9.92% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 25497010155 27018079806 +5.97% BenchmarkCompactionFromHead/labelnames=1,labelvalues=100000-4 2427391406 2817217987 +16.06% BenchmarkCompactionFromHead/labelnames=10,labelvalues=10000-4 2592965497 2538805050 -2.09% BenchmarkCompactionFromHead/labelnames=100,labelvalues=1000-4 2437388343 2668012858 +9.46% BenchmarkCompactionFromHead/labelnames=1000,labelvalues=100-4 2317095324 2787423966 +20.30% BenchmarkCompactionFromHead/labelnames=10000,labelvalues=10-4 2600239857 2096973860 -19.35% benchmark old allocs new allocs delta BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 500851 470794 -6.00% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 821527 791451 -3.66% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 1141562 1111508 -2.63% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 2141576 2111504 -1.40% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 871466 841424 -3.45% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 1941428 1911415 -1.55% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 3071573 3041510 -0.98% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 6771648 6741509 -0.45% BenchmarkCompactionFromHead/labelnames=1,labelvalues=100000-4 731493 824888 +12.77% BenchmarkCompactionFromHead/labelnames=10,labelvalues=10000-4 793918 887311 +11.76% BenchmarkCompactionFromHead/labelnames=100,labelvalues=1000-4 811842 905204 +11.50% BenchmarkCompactionFromHead/labelnames=1000,labelvalues=100-4 832244 925081 +11.16% BenchmarkCompactionFromHead/labelnames=10000,labelvalues=10-4 921553 1019162 +10.59% benchmark old bytes new bytes delta BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 40532648 35698276 -11.93% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 60340216 53409568 -11.49% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 81087336 72065552 -11.13% BenchmarkCompaction/type=normal,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 142485576 120878544 -15.16% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=101-4 208661368 203831136 -2.31% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=1001-4 347345904 340484696 -1.98% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=2001-4 585185856 576244648 -1.53% BenchmarkCompaction/type=vertical,blocks=4,series=10000,samplesPerSeriesPerBlock=5001-4 1357641792 1358966528 +0.10% BenchmarkCompactionFromHead/labelnames=1,labelvalues=100000-4 126486664 119666744 -5.39% BenchmarkCompactionFromHead/labelnames=10,labelvalues=10000-4 122323192 115117224 -5.89% BenchmarkCompactionFromHead/labelnames=100,labelvalues=1000-4 126404504 119469864 -5.49% BenchmarkCompactionFromHead/labelnames=1000,labelvalues=100-4 119047832 112230408 -5.73% BenchmarkCompactionFromHead/labelnames=10000,labelvalues=10-4 136576016 116634800 -14.60% Signed-off-by: Brian Brazil <[email protected]>
GeorgeSen · Dec 17, 2019 · d782387 · d782387
1 parent 767fa70
commit d782387
Show file tree

Hide file tree

Showing 8 changed files with 275 additions and 123 deletions.
diff --git a/tsdb/block.go b/tsdb/block.go
@@ -39,8 +39,8 @@ import (
 // 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
+	// and other indices. Symbols must be added in sorted order.
+	AddSymbol(sym string) error
 
 	// AddSeries populates the index writer with a series and its offsets
 	// of chunks that the index can reference.
@@ -61,9 +61,10 @@ type IndexWriter interface {
 
 // 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)
+	// Symbols return an iterator over sorted string symbols that may occur in
+	// series' labels and indices. It is not safe to use the returned strings
+	// beyond the lifetime of the index reader.
+	Symbols() index.StringIter
 
 	// LabelValues returns sorted possible label values.
 	LabelValues(names ...string) (index.StringTuples, error)
@@ -432,9 +433,8 @@ type blockIndexReader struct {
 	b  *Block
 }
 
-func (r blockIndexReader) Symbols() (map[string]struct{}, error) {
-	s, err := r.ir.Symbols()
-	return s, errors.Wrapf(err, "block: %s", r.b.Meta().ULID)
+func (r blockIndexReader) Symbols() index.StringIter {
+	return r.ir.Symbols()
 }
 
 func (r blockIndexReader) LabelValues(names ...string) (index.StringTuples, error) {

diff --git a/tsdb/compact.go b/tsdb/compact.go
@@ -650,7 +650,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 
 	var (
 		set         ChunkSeriesSet
-		allSymbols  = make(map[string]struct{}, 1<<16)
+		symbols     index.StringIter
 		closers     = []io.Closer{}
 		overlapping bool
 	)
@@ -700,14 +700,6 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 		}
 		closers = append(closers, tombsr)
 
-		symbols, err := indexr.Symbols()
-		if err != nil {
-			return errors.Wrap(err, "read symbols")
-		}
-		for s := range symbols {
-			allSymbols[s] = struct{}{}
-		}
-
 		k, v := index.AllPostingsKey()
 		all, err := indexr.Postings(k, v)
 		if err != nil {
@@ -716,24 +708,32 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
 		all = indexr.SortedPostings(all)
 
 		s := newCompactionSeriesSet(indexr, chunkr, tombsr, all)
+		syms := indexr.Symbols()
 
 		if i == 0 {
 			set = s
+			symbols = syms
 			continue
 		}
 		set, err = newCompactionMerger(set, s)
 		if err != nil {
 			return err
 		}
+		symbols = newMergedStringIter(symbols, syms)
 	}
 
 	var (
 		values = map[string]stringset{}
 		ref    = uint64(0)
 	)
 
-	if err := indexw.AddSymbols(allSymbols); err != nil {
-		return errors.Wrap(err, "add symbols")
+	for symbols.Next() {
+		if err := indexw.AddSymbol(symbols.At()); err != nil {
+			return errors.Wrap(err, "add symbol")
+		}
+	}
+	if symbols.Err() != nil {
+		return errors.Wrap(symbols.Err(), "next symbol")
 	}
 
 	delIter := &deletedIterator{}
@@ -1026,3 +1026,47 @@ func (c *compactionMerger) Err() error {
 func (c *compactionMerger) At() (labels.Labels, []chunks.Meta, tombstones.Intervals) {
 	return c.l, c.c, c.intervals
 }
+
+func newMergedStringIter(a index.StringIter, b index.StringIter) index.StringIter {
+	return &mergedStringIter{a: a, b: b, aok: a.Next(), bok: b.Next()}
+}
+
+type mergedStringIter struct {
+	a        index.StringIter
+	b        index.StringIter
+	aok, bok bool
+	cur      string
+}
+
+func (m *mergedStringIter) Next() bool {
+	if (!m.aok && !m.bok) || (m.Err() != nil) {
+		return false
+	}
+
+	if !m.aok {
+		m.cur = m.b.At()
+		m.bok = m.b.Next()
+	} else if !m.bok {
+		m.cur = m.a.At()
+		m.aok = m.a.Next()
+	} else if m.b.At() > m.a.At() {
+		m.cur = m.a.At()
+		m.aok = m.a.Next()
+	} else if m.a.At() > m.b.At() {
+		m.cur = m.b.At()
+		m.bok = m.b.Next()
+	} else { // Equal.
+		m.cur = m.b.At()
+		m.aok = m.a.Next()
+		m.bok = m.b.Next()
+	}
+
+	return true
+}
+func (m mergedStringIter) At() string { return m.cur }
+func (m mergedStringIter) Err() error {
+	if m.a.Err() != nil {
+		return m.a.Err()
+	}
+	return m.b.Err()
+}
diff --git a/tsdb/docs/format/index.md b/tsdb/docs/format/index.md
@@ -126,7 +126,7 @@ After the labels, the number of indexed chunks is encoded, followed by a sequenc
 ### Label Index
 
 A label index section indexes the existing (combined) values for one or more label names.
-The `#names` field determines the number of indexed label names, followed by the total number of entries in the `#entries` field. The body holds #entries / #names tuples of symbol table references, each tuple being of #names length. The value tuples are sorted in lexicographically increasing order.
+The `#names` field determines the number of indexed label names, followed by the total number of entries in the `#entries` field. The body holds #entries / #names tuples of symbol table references, each tuple being of #names length. The value tuples are sorted in lexicographically increasing order. This is no longer used.
 
 ```
 ┌───────────────┬────────────────┬────────────────┐
@@ -181,7 +181,7 @@ The sequence of postings sections is finalized by a [postings offset table](#pos
 
 A label offset table stores a sequence of label offset entries.
 Every label offset entry holds the label name and the offset to its values in the label index section.
-They are used to track label index sections. They are read into memory when an index file is loaded.
+They are used to track label index sections. This is no longer used.
 
 ```
 ┌─────────────────────┬──────────────────────┐

diff --git a/tsdb/head.go b/tsdb/head.go
@@ -1339,16 +1339,17 @@ func (h *headIndexReader) Close() error {
 	return nil
 }
 
-func (h *headIndexReader) Symbols() (map[string]struct{}, error) {
+func (h *headIndexReader) Symbols() index.StringIter {
 	h.head.symMtx.RLock()
-	defer h.head.symMtx.RUnlock()
-
-	res := make(map[string]struct{}, len(h.head.symbols))
+	res := make([]string, 0, len(h.head.symbols))
 
 	for s := range h.head.symbols {
-		res[s] = struct{}{}
+		res = append(res, s)
 	}
-	return res, nil
+	h.head.symMtx.RUnlock()
+
+	sort.Strings(res)
+	return index.NewStringListIter(res)
 }
 
 // LabelValues returns the possible label values