david
06a95597e9
Bitstream primitives (bstream) and XOR encoder/decoder following the Prometheus TSDB variant of the Gorilla scheme. Delta-of-delta timestamps with 14/17/20/64-bit buckets, XOR-compressed float64 values. Decoder is total: arbitrary bytes produce values or ErrShortStream.
384 lines
11 KiB
Go
384 lines
11 KiB
Go
package chunkenc
|
|
|
|
import (
|
|
"encoding/binary"
|
|
"errors"
|
|
"fmt"
|
|
"math"
|
|
"math/rand"
|
|
"testing"
|
|
|
|
"github.com/stretchr/testify/assert"
|
|
)
|
|
|
|
// encodeChunk builds a valid XOR chunk from samples and returns its bytes.
|
|
func encodeChunk(samples [][2]float64) []byte {
|
|
c := NewXORChunk()
|
|
a, _ := c.Appender()
|
|
for _, s := range samples {
|
|
a.Append(int64(s[0]), s[1])
|
|
}
|
|
return append([]byte(nil), c.Bytes()...)
|
|
}
|
|
|
|
func TestXORChunk(t *testing.T) {
|
|
rnd := rand.New(rand.NewSource(42))
|
|
|
|
type readOp struct {
|
|
wantNext bool
|
|
wantT int64
|
|
wantVBits uint64
|
|
}
|
|
|
|
type testCase struct {
|
|
name string
|
|
data []byte
|
|
reads []readOp
|
|
wantIterErr error
|
|
maxBytes int
|
|
wantAppenderErr error
|
|
}
|
|
|
|
nonEmptyErr := errors.New("chunkenc: appender on non-empty chunk")
|
|
|
|
// successReads: one successful read per sample, then a terminal Next()=false.
|
|
successReads := func(samples [][2]float64) []readOp {
|
|
ops := make([]readOp, len(samples)+1)
|
|
for i, s := range samples {
|
|
ops[i] = readOp{true, int64(s[0]), math.Float64bits(s[1])}
|
|
}
|
|
term := readOp{wantNext: false}
|
|
if len(samples) > 0 {
|
|
last := samples[len(samples)-1]
|
|
term.wantT = int64(last[0])
|
|
term.wantVBits = math.Float64bits(last[1])
|
|
}
|
|
ops[len(samples)] = term
|
|
return ops
|
|
}
|
|
|
|
// --- Build sample sets for named round-trip cases ---
|
|
|
|
adversarialVals := []float64{
|
|
math.NaN(), math.Inf(1), math.Inf(-1), 0, math.Copysign(0, -1),
|
|
math.MaxFloat64, math.SmallestNonzeroFloat64, -1e300,
|
|
}
|
|
adversarial := make([][2]float64, len(adversarialVals))
|
|
for i, v := range adversarialVals {
|
|
adversarial[i] = [2]float64{float64(1000 + i*15), v}
|
|
}
|
|
|
|
jitteryTimes := []int64{1000, 1015, 1030, 1031, 1500, 1501, 200000, 200015, 9000000000}
|
|
jittery := make([][2]float64, len(jitteryTimes))
|
|
for i, ts := range jitteryTimes {
|
|
jittery[i] = [2]float64{float64(ts), float64(i) * 1.1}
|
|
}
|
|
|
|
singleSample := [][2]float64{{1000, 71.3}}
|
|
repeatedValue := [][2]float64{{1000, 71.3}, {1015, 71.3}}
|
|
changedValue := [][2]float64{{1000, 71.3}, {1015, 71.4}}
|
|
traceSeq := [][2]float64{{1000, 71.3}, {1015, 71.3}, {1030, 71.4}, {1045, 71.4}}
|
|
|
|
tests := []testCase{
|
|
// --- Round-trip: encode then decode ---
|
|
{
|
|
name: "single_sample",
|
|
data: encodeChunk(singleSample),
|
|
reads: successReads(singleSample),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
name: "repeated_value",
|
|
data: encodeChunk(repeatedValue),
|
|
reads: successReads(repeatedValue),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
name: "changed_value",
|
|
data: encodeChunk(changedValue),
|
|
reads: successReads(changedValue),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
name: "trace_sequence",
|
|
data: encodeChunk(traceSeq),
|
|
reads: successReads(traceSeq),
|
|
wantIterErr: nil,
|
|
maxBytes: 30,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
name: "adversarial_values",
|
|
data: encodeChunk(adversarial),
|
|
reads: successReads(adversarial),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
name: "negative_and_jittery_dods",
|
|
data: encodeChunk(jittery),
|
|
reads: successReads(jittery),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
|
|
// --- Error paths: truncated or empty data ---
|
|
{
|
|
name: "zero_samples",
|
|
data: []byte{0, 0},
|
|
reads: []readOp{{false, 0, 0}},
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nil, // empty chunk, Appender succeeds
|
|
},
|
|
{
|
|
// Header claims 1 sample, no data after header.
|
|
// Case 0: readBits(64) for timestamp fails immediately.
|
|
name: "no_data_after_header",
|
|
data: []byte{0, 1},
|
|
reads: []readOp{{false, 0, 0}},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
// Header claims 1, only 4 bytes of data (partial timestamp).
|
|
// Case 0: readBits(64) for timestamp fails mid-read.
|
|
name: "truncated_first_timestamp",
|
|
data: func() []byte {
|
|
d := make([]byte, 6) // 2 header + 4 data
|
|
binary.BigEndian.PutUint16(d, 1)
|
|
return d
|
|
}(),
|
|
reads: []readOp{{false, 0, 0}},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
// Header claims 1, full timestamp but only 4 bytes of value.
|
|
// Case 0: readBits(64) for value fails; it.t and it.v not set
|
|
// (assigned only after both reads succeed).
|
|
name: "truncated_first_value",
|
|
data: func() []byte {
|
|
d := make([]byte, 14) // 2 header + 8 timestamp + 4 partial value
|
|
binary.BigEndian.PutUint16(d, 1)
|
|
binary.BigEndian.PutUint64(d[2:], uint64(1000))
|
|
return d
|
|
}(),
|
|
reads: []readOp{{false, 0, 0}},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
// Header claims 2, data for exactly 1 sample (16 bytes).
|
|
// Case 1: readBits(14) for delta fails (no data remains).
|
|
// At() returns last successful sample.
|
|
name: "truncated_second_delta",
|
|
data: func() []byte {
|
|
d := make([]byte, 18) // 2 header + 16 data = exactly 1 sample
|
|
binary.BigEndian.PutUint16(d, 2)
|
|
binary.BigEndian.PutUint64(d[2:], uint64(1000))
|
|
binary.BigEndian.PutUint64(d[10:], math.Float64bits(71.3))
|
|
return d
|
|
}(),
|
|
reads: []readOp{
|
|
{true, 1000, math.Float64bits(71.3)},
|
|
{false, 1000, math.Float64bits(71.3)},
|
|
},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
// Header claims 2. Data: 1 valid sample + 14-bit delta + value-changed(1) +
|
|
// new-window(1), then truncated before leading-zeros field.
|
|
// Case 1: delta succeeds (it.t updated), readVDelta fails inside
|
|
// new-window branch at readBits(5).
|
|
name: "truncated_second_vdelta",
|
|
data: func() []byte {
|
|
d := make([]byte, 20) // 2 header + 18 data bytes (144 bits)
|
|
binary.BigEndian.PutUint16(d, 2)
|
|
binary.BigEndian.PutUint64(d[2:], uint64(1000))
|
|
binary.BigEndian.PutUint64(d[10:], math.Float64bits(71.3))
|
|
// Bits 128-141: 14-bit delta = 15 (0b00000000001111)
|
|
// Byte 18 (bits 128-135): 0x00
|
|
// Byte 19 (bits 136-143):
|
|
// 136-141 = bottom 6 bits of delta (001111)
|
|
// 142 = value-changed (1)
|
|
// 143 = new-window (1)
|
|
// = 0b00111111 = 0x3F
|
|
d[19] = 0x3F
|
|
return d
|
|
}(),
|
|
reads: []readOp{
|
|
{true, 1000, math.Float64bits(71.3)},
|
|
{false, 1015, math.Float64bits(71.3)}, // t advanced, v unchanged
|
|
},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
{
|
|
// Header claims 3. Data: 2 valid samples (same value, so sample 1
|
|
// is 14-bit delta + 1-bit value-unchanged = 15 bits). Total data =
|
|
// 143 bits in 18 bytes (1 padding bit). Sample 2 DoD prefix reads
|
|
// the padding zero (dod=0, t advances), then readVDelta fails.
|
|
name: "truncated_third_vdelta",
|
|
data: func() []byte {
|
|
d := make([]byte, 20) // 2 header + 18 data (144 bits)
|
|
binary.BigEndian.PutUint16(d, 3)
|
|
binary.BigEndian.PutUint64(d[2:], uint64(1000))
|
|
binary.BigEndian.PutUint64(d[10:], math.Float64bits(71.3))
|
|
// Bits 128-141: 14-bit delta = 15
|
|
// Bit 142: value-unchanged (0)
|
|
// Bit 143: padding (0)
|
|
// Byte 18 = 0x00, Byte 19 = 0b00111100 = 0x3C
|
|
d[19] = 0x3C
|
|
return d
|
|
}(),
|
|
reads: []readOp{
|
|
{true, 1000, math.Float64bits(71.3)},
|
|
{true, 1015, math.Float64bits(71.3)},
|
|
{false, 1030, math.Float64bits(71.3)}, // dod=0 decoded from padding, vdelta fails
|
|
},
|
|
wantIterErr: ErrShortStream,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
},
|
|
}
|
|
|
|
// Random walk round-trip cases.
|
|
for i := 0; i < 100; i++ {
|
|
var samples [][2]float64
|
|
ts, v := int64(rnd.Intn(1e6)), rnd.Float64()*100
|
|
for j := 0; j < 120; j++ {
|
|
samples = append(samples, [2]float64{float64(ts), v})
|
|
ts += 15000 + int64(rnd.Intn(100)) - 50
|
|
v += rnd.Float64() - 0.5
|
|
}
|
|
tests = append(tests, testCase{
|
|
name: fmt.Sprintf("random_walk_%03d", i),
|
|
data: encodeChunk(samples),
|
|
reads: successReads(samples),
|
|
wantIterErr: nil,
|
|
maxBytes: math.MaxInt,
|
|
wantAppenderErr: nonEmptyErr,
|
|
})
|
|
}
|
|
|
|
for _, tc := range tests {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
c := &XORChunk{b: bstream{stream: tc.data}}
|
|
|
|
it := c.Iterator()
|
|
for i, r := range tc.reads {
|
|
next := it.Next()
|
|
gotT, gotV := it.At()
|
|
assert.Equal(t, r.wantNext, next, "read %d Next()", i)
|
|
assert.Equal(t, r.wantT, gotT, "read %d timestamp", i)
|
|
assert.Equal(t, r.wantVBits, math.Float64bits(gotV), "read %d value", i)
|
|
}
|
|
assert.Equal(t, tc.wantIterErr, it.Err())
|
|
|
|
assert.LessOrEqual(t, len(tc.data), tc.maxBytes)
|
|
|
|
_, appErr := c.Appender()
|
|
assert.Equal(t, tc.wantAppenderErr, appErr)
|
|
})
|
|
}
|
|
}
|
|
|
|
func FuzzXORIterator(f *testing.F) {
|
|
c := NewXORChunk()
|
|
a, _ := c.Appender()
|
|
a.Append(1000, 71.3)
|
|
a.Append(1015, 71.4)
|
|
f.Add(c.Bytes())
|
|
|
|
f.Fuzz(func(t *testing.T, data []byte) {
|
|
if len(data) < 2 {
|
|
return
|
|
}
|
|
chunk := &XORChunk{b: bstream{stream: data}}
|
|
it := chunk.Iterator()
|
|
for it.Next() {
|
|
}
|
|
// Termination without panic is the only assertion.
|
|
})
|
|
}
|
|
|
|
// benchChunk fills a chunk with 120 samples from gen and returns it.
|
|
func benchChunk(gen func(i int) float64) *XORChunk {
|
|
c := NewXORChunk()
|
|
a, _ := c.Appender()
|
|
ts := int64(0)
|
|
for i := 0; i < 120; i++ {
|
|
a.Append(ts, gen(i))
|
|
ts += 15000
|
|
}
|
|
return c
|
|
}
|
|
|
|
func BenchmarkAppend(b *testing.B) {
|
|
rnd := rand.New(rand.NewSource(42))
|
|
vals := make([]float64, 120)
|
|
v := 70.0
|
|
for i := range vals {
|
|
vals[i] = v
|
|
v += rnd.Float64() - 0.5
|
|
}
|
|
|
|
b.ReportAllocs()
|
|
b.ResetTimer()
|
|
for i := 0; i < b.N; i++ {
|
|
c := NewXORChunk()
|
|
a, _ := c.Appender()
|
|
ts := int64(0)
|
|
for j := 0; j < 120; j++ {
|
|
a.Append(ts, vals[j])
|
|
ts += 15000
|
|
}
|
|
}
|
|
b.ReportMetric(float64(b.N*120)/b.Elapsed().Seconds(), "appends/sec")
|
|
}
|
|
|
|
func TestBytesPerSample(t *testing.T) {
|
|
rnd := rand.New(rand.NewSource(42))
|
|
|
|
tests := []struct {
|
|
name string
|
|
max float64
|
|
gen func(i int) float64
|
|
}{
|
|
{"constant", 0.5, func(i int) float64 { return 71.3 }},
|
|
{"stepped_sensor", 3.0, func(i int) float64 {
|
|
return 70 + math.Floor(float64(i)/8)*0.1
|
|
}},
|
|
{"integer_counter", 3.0, func(i int) float64 {
|
|
return float64(1000 + i*3)
|
|
}},
|
|
{"full_precision_walk", 10.0, func(i int) float64 {
|
|
return 70 + rnd.NormFloat64()
|
|
}},
|
|
}
|
|
|
|
for _, tc := range tests {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
c := benchChunk(tc.gen)
|
|
bps := float64(len(c.Bytes())) / 120.0
|
|
t.Logf("%s: %.3f bytes/sample (%d bytes total)", tc.name, bps, len(c.Bytes()))
|
|
assert.LessOrEqual(t, bps, tc.max, "%s bytes/sample exceeds ceiling", tc.name)
|
|
})
|
|
}
|
|
}
|