Merge pull request #6676 from roidelapluie/postings

Make head Postings only return series in time range
This commit is contained in:
Julien Pivotto 2020-02-21 14:55:44 +01:00 committed by GitHub
commit fdbb18b445
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
12 changed files with 152 additions and 29 deletions

View file

@ -111,8 +111,9 @@ type ChunkReader interface {
// BlockReader provides reading access to a data block.
type BlockReader interface {
// Index returns an IndexReader over the block's data.
Index() (IndexReader, error)
// Index returns an IndexReader over the block's data within the specified
// timeframe.
Index(mint, maxt int64) (IndexReader, error)
// Chunks returns a ChunkReader over the block's data.
Chunks() (ChunkReader, error)
@ -372,7 +373,7 @@ func (pb *Block) startRead() error {
}
// Index returns a new IndexReader against the block data.
func (pb *Block) Index() (IndexReader, error) {
func (pb *Block) Index(mint, maxt int64) (IndexReader, error) {
if err := pb.startRead(); err != nil {
return nil, err
}

View file

@ -470,7 +470,7 @@ func analyzeBlock(b tsdb.BlockReader, limit int) error {
// Presume 1ms resolution that Prometheus uses.
fmt.Printf("Duration: %s\n", (time.Duration(meta.MaxTime-meta.MinTime) * 1e6).String())
fmt.Printf("Series: %d\n", meta.Stats.NumSeries)
ir, err := b.Index()
ir, err := b.Index(math.MinInt64, math.MaxInt64)
if err != nil {
return err
}

View file

@ -683,7 +683,7 @@ func (c *LeveledCompactor) populateBlock(blocks []BlockReader, meta *BlockMeta,
}
}
indexr, err := b.Index()
indexr, err := b.Index(math.MinInt64, globalMaxt)
if err != nil {
return errors.Wrapf(err, "open index reader for block %s", b)
}

View file

@ -456,10 +456,10 @@ func metaRange(name string, mint, maxt int64, stats *BlockStats) dirMeta {
type erringBReader struct{}
func (erringBReader) Index() (IndexReader, error) { return nil, errors.New("index") }
func (erringBReader) Chunks() (ChunkReader, error) { return nil, errors.New("chunks") }
func (erringBReader) Tombstones() (tombstones.Reader, error) { return nil, errors.New("tombstones") }
func (erringBReader) Meta() BlockMeta { return BlockMeta{} }
func (erringBReader) Index(int64, int64) (IndexReader, error) { return nil, errors.New("index") }
func (erringBReader) Chunks() (ChunkReader, error) { return nil, errors.New("chunks") }
func (erringBReader) Tombstones() (tombstones.Reader, error) { return nil, errors.New("tombstones") }
func (erringBReader) Meta() BlockMeta { return BlockMeta{} }
type nopChunkWriter struct{}
@ -652,7 +652,7 @@ func TestCompaction_populateBlock(t *testing.T) {
expErr: errors.New("found chunk with minTime: 10 maxTime: 30 outside of compacted minTime: 0 maxTime: 20"),
},
{
// Introduced by https://github.com/prometheus/prometheus/tsdb/issues/347.
// Introduced by https://github.com/prometheus/tsdb/issues/347.
title: "Populate from single block containing extra chunk",
inputSeriesSamples: [][]seriesSamples{
{
@ -692,7 +692,7 @@ func TestCompaction_populateBlock(t *testing.T) {
},
},
{
// Introduced by https://github.com/prometheus/prometheus/tsdb/pull/539.
// Introduced by https://github.com/prometheus/tsdb/pull/539.
title: "Populate from three blocks that the last two are overlapping.",
inputSeriesSamples: [][]seriesSamples{
{

View file

@ -1384,7 +1384,7 @@ func TestOverlappingBlocksDetectsAllOverlaps(t *testing.T) {
}, OverlappingBlocks(nc1))
}
// Regression test for https://github.com/prometheus/prometheus/tsdb/issues/347
// Regression test for https://github.com/prometheus/tsdb/issues/347
func TestChunkAtBlockBoundary(t *testing.T) {
db, closeFn := openTestDB(t, nil, nil)
defer func() {
@ -1411,7 +1411,7 @@ func TestChunkAtBlockBoundary(t *testing.T) {
testutil.Ok(t, err)
for _, block := range db.Blocks() {
r, err := block.Index()
r, err := block.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(t, err)
defer r.Close()
@ -1761,7 +1761,7 @@ func TestDB_LabelNames(t *testing.T) {
appendSamples(db, 0, 4, tst.sampleLabels1)
// Testing head.
headIndexr, err := db.head.Index()
headIndexr, err := db.head.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(t, err)
labelNames, err := headIndexr.LabelNames()
testutil.Ok(t, err)
@ -1774,7 +1774,7 @@ func TestDB_LabelNames(t *testing.T) {
// All blocks have same label names, hence check them individually.
// No need to aggregate and check.
for _, b := range db.Blocks() {
blockIndexr, err := b.Index()
blockIndexr, err := b.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(t, err)
labelNames, err = blockIndexr.LabelNames()
testutil.Ok(t, err)

View file

@ -754,7 +754,7 @@ type RangeHead struct {
mint, maxt int64
}
// NewRangeHead returns a *rangeHead.
// NewRangeHead returns a *RangeHead.
func NewRangeHead(head *Head, mint, maxt int64) *RangeHead {
return &RangeHead{
head: head,
@ -763,8 +763,15 @@ func NewRangeHead(head *Head, mint, maxt int64) *RangeHead {
}
}
func (h *RangeHead) Index() (IndexReader, error) {
return h.head.indexRange(h.mint, h.maxt), nil
func (h *RangeHead) Index(mint, maxt int64) (IndexReader, error) {
// rangeHead guarantees that the series returned are within its range.
if mint < h.mint {
mint = h.mint
}
if maxt > h.maxt {
maxt = h.maxt
}
return h.head.indexRange(mint, maxt), nil
}
func (h *RangeHead) Chunks() (ChunkReader, error) {
@ -1162,8 +1169,8 @@ func (h *Head) Tombstones() (tombstones.Reader, error) {
}
// Index returns an IndexReader against the block.
func (h *Head) Index() (IndexReader, error) {
return h.indexRange(math.MinInt64, math.MaxInt64), nil
func (h *Head) Index(mint, maxt int64) (IndexReader, error) {
return h.indexRange(mint, maxt), nil
}
func (h *Head) indexRange(mint, maxt int64) *headIndexReader {
@ -1347,9 +1354,35 @@ func (h *headIndexReader) LabelNames() ([]string, error) {
// Postings returns the postings list iterator for the label pairs.
func (h *headIndexReader) Postings(name string, values ...string) (index.Postings, error) {
fullRange := h.mint <= h.head.MinTime() && h.maxt >= h.head.MaxTime()
res := make([]index.Postings, 0, len(values))
for _, value := range values {
res = append(res, h.head.postings.Get(name, value))
p := h.head.postings.Get(name, value)
if fullRange {
// The head timerange covers the full index reader timerange.
// All the series can the be appended without filtering.
res = append(res, p)
continue
}
// Filter out series not in the time range, to avoid
// later on building up all the chunk metadata just to
// discard it.
filtered := []uint64{}
for p.Next() {
s := h.head.series.getByID(p.At())
if s == nil {
level.Debug(h.head.logger).Log("msg", "looked up series not found")
continue
}
if s.minTime() <= h.maxt && s.maxTime() >= h.mint {
filtered = append(filtered, p.At())
}
}
if p.Err() != nil {
return nil, p.Err()
}
res = append(res, index.NewListPostings(filtered))
}
return index.Merge(res...), nil
}

View file

@ -1322,3 +1322,91 @@ func TestAddDuplicateLabelName(t *testing.T) {
add(labels.Labels{{Name: "a", Value: "c"}, {Name: "a", Value: "c"}}, "a")
add(labels.Labels{{Name: "__name__", Value: "up"}, {Name: "job", Value: "prometheus"}, {Name: "le", Value: "500"}, {Name: "le", Value: "400"}, {Name: "unit", Value: "s"}}, "le")
}
func TestHeadSeriesWithTimeBoundaries(t *testing.T) {
h, err := NewHead(nil, nil, nil, 15, DefaultStripeSize)
testutil.Ok(t, err)
defer h.Close()
app := h.Appender()
s1, err := app.Add(labels.FromStrings("foo1", "bar"), 2, 0)
testutil.Ok(t, err)
for ts := int64(3); ts < 13; ts++ {
err = app.AddFast(s1, ts, 0)
testutil.Ok(t, err)
}
s2, err := app.Add(labels.FromStrings("foo2", "bar"), 5, 0)
testutil.Ok(t, err)
for ts := int64(6); ts < 11; ts++ {
err = app.AddFast(s2, ts, 0)
testutil.Ok(t, err)
}
s3, err := app.Add(labels.FromStrings("foo3", "bar"), 5, 0)
testutil.Ok(t, err)
err = app.AddFast(s3, 6, 0)
testutil.Ok(t, err)
_, err = app.Add(labels.FromStrings("foo4", "bar"), 9, 0)
testutil.Ok(t, err)
testutil.Ok(t, app.Commit())
cases := []struct {
mint int64
maxt int64
seriesCount int
samplesCount int
}{
// foo1 ..00000000000..
// foo2 .....000000....
// foo3 .....00........
// foo4 .........0.....
{mint: 0, maxt: 0, seriesCount: 0, samplesCount: 0},
{mint: 0, maxt: 1, seriesCount: 0, samplesCount: 0},
{mint: 0, maxt: 2, seriesCount: 1, samplesCount: 1},
{mint: 2, maxt: 2, seriesCount: 1, samplesCount: 1},
{mint: 0, maxt: 4, seriesCount: 1, samplesCount: 3},
{mint: 0, maxt: 5, seriesCount: 3, samplesCount: 6},
{mint: 0, maxt: 6, seriesCount: 3, samplesCount: 9},
{mint: 0, maxt: 7, seriesCount: 3, samplesCount: 11},
{mint: 0, maxt: 8, seriesCount: 3, samplesCount: 13},
{mint: 0, maxt: 9, seriesCount: 4, samplesCount: 16},
{mint: 0, maxt: 10, seriesCount: 4, samplesCount: 18},
{mint: 0, maxt: 11, seriesCount: 4, samplesCount: 19},
{mint: 0, maxt: 12, seriesCount: 4, samplesCount: 20},
{mint: 0, maxt: 13, seriesCount: 4, samplesCount: 20},
{mint: 0, maxt: 14, seriesCount: 4, samplesCount: 20},
{mint: 2, maxt: 14, seriesCount: 4, samplesCount: 20},
{mint: 3, maxt: 14, seriesCount: 4, samplesCount: 19},
{mint: 4, maxt: 14, seriesCount: 4, samplesCount: 18},
{mint: 8, maxt: 9, seriesCount: 3, samplesCount: 5},
{mint: 9, maxt: 9, seriesCount: 3, samplesCount: 3},
{mint: 6, maxt: 9, seriesCount: 4, samplesCount: 10},
{mint: 11, maxt: 11, seriesCount: 1, samplesCount: 1},
{mint: 11, maxt: 12, seriesCount: 1, samplesCount: 2},
{mint: 11, maxt: 14, seriesCount: 1, samplesCount: 2},
{mint: 12, maxt: 14, seriesCount: 1, samplesCount: 1},
}
for i, c := range cases {
matcher := labels.MustNewMatcher(labels.MatchEqual, "", "")
q, err := NewBlockQuerier(h, c.mint, c.maxt)
testutil.Ok(t, err)
seriesCount := 0
samplesCount := 0
ss, _, err := q.Select(nil, matcher)
testutil.Ok(t, err)
for ss.Next() {
i := ss.At().Iterator()
for i.Next() {
samplesCount++
}
seriesCount++
}
testutil.Ok(t, ss.Err())
testutil.Equals(t, c.seriesCount, seriesCount, "test series %d", i)
testutil.Equals(t, c.samplesCount, samplesCount, "test samples %d", i)
q.Close()
}
}

View file

@ -71,8 +71,8 @@ type mockBReader struct {
maxt int64
}
func (r *mockBReader) Index() (IndexReader, error) { return r.ir, nil }
func (r *mockBReader) Chunks() (ChunkReader, error) { return r.cr, nil }
func (r *mockBReader) Index(mint, maxt int64) (IndexReader, error) { return r.ir, nil }
func (r *mockBReader) Chunks() (ChunkReader, error) { return r.cr, nil }
func (r *mockBReader) Tombstones() (tombstones.Reader, error) {
return tombstones.NewMemTombstones(), nil
}

View file

@ -160,7 +160,7 @@ func (q *verticalQuerier) sel(p *storage.SelectParams, qs []storage.Querier, ms
// NewBlockQuerier returns a querier against the reader.
func NewBlockQuerier(b BlockReader, mint, maxt int64) (storage.Querier, error) {
indexr, err := b.Index()
indexr, err := b.Index(mint, maxt)
if err != nil {
return nil, errors.Wrapf(err, "open index reader")
}

View file

@ -16,6 +16,7 @@ package tsdb
import (
"fmt"
"io/ioutil"
"math"
"os"
"strconv"
"testing"
@ -54,7 +55,7 @@ func BenchmarkPostingsForMatchers(b *testing.B) {
}
testutil.Ok(b, app.Commit())
ir, err := h.Index()
ir, err := h.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(b, err)
b.Run("Head", func(b *testing.B) {
benchmarkPostingsForMatchers(b, ir)
@ -72,7 +73,7 @@ func BenchmarkPostingsForMatchers(b *testing.B) {
defer func() {
testutil.Ok(b, block.Close())
}()
ir, err = block.Index()
ir, err = block.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(b, err)
defer ir.Close()
b.Run("Block", func(b *testing.B) {

View file

@ -1044,7 +1044,7 @@ func TestSeriesIterator(t *testing.T) {
})
}
// Regression for: https://github.com/prometheus/prometheus/tsdb/pull/97
// Regression for: https://github.com/prometheus/tsdb/pull/97
func TestChunkSeriesIterator_DoubleSeek(t *testing.T) {
chkMetas := []chunks.Meta{
tsdbutil.ChunkFromSamples([]tsdbutil.Sample{}),
@ -2137,7 +2137,7 @@ func TestPostingsForMatchers(t *testing.T) {
},
}
ir, err := h.Index()
ir, err := h.Index(math.MinInt64, math.MaxInt64)
testutil.Ok(t, err)
for _, c := range cases {

View file

@ -49,7 +49,7 @@ func TestWALRepair_ReadingError(t *testing.T) {
},
// Ensures that the page buffer is big enough to fit
// an entire page size without panicking.
// https://github.com/prometheus/prometheus/tsdb/pull/414
// https://github.com/prometheus/tsdb/pull/414
"bad_header": {
1,
func(f *os.File) {