mirror of
https://github.com/VictoriaMetrics/VictoriaMetrics.git
synced 2024-11-21 14:44:00 +00:00
5fa9525498
This should reduce chances of unnoticed on-disk data corruption. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/2998 Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3011 This change modifies the format for data exported via /api/v1/export/native - now this data contains MaxTimestamp and PrecisionBits fields from blockHeader. This is OK, since the native export format is undocumented.
388 lines
12 KiB
Go
388 lines
12 KiB
Go
package storage
|
|
|
|
import (
|
|
"fmt"
|
|
"sync"
|
|
"sync/atomic"
|
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
|
|
)
|
|
|
|
const (
|
|
// The maximum number of rows per block.
|
|
maxRowsPerBlock = 8 * 1024
|
|
|
|
// The maximum size of values in the block.
|
|
maxBlockSize = 8 * maxRowsPerBlock
|
|
)
|
|
|
|
// Block represents a block of time series values for a single TSID.
|
|
type Block struct {
|
|
bh blockHeader
|
|
|
|
// nextIdx is the next index for reading timestamps and values.
|
|
nextIdx int
|
|
|
|
timestamps []int64
|
|
values []int64
|
|
|
|
// Marshaled representation of block header.
|
|
headerData []byte
|
|
|
|
// Marshaled representation of timestamps.
|
|
timestampsData []byte
|
|
|
|
// Marshaled representation of values.
|
|
valuesData []byte
|
|
}
|
|
|
|
// Reset resets b.
|
|
func (b *Block) Reset() {
|
|
b.bh = blockHeader{}
|
|
b.nextIdx = 0
|
|
b.timestamps = b.timestamps[:0]
|
|
b.values = b.values[:0]
|
|
|
|
b.headerData = b.headerData[:0]
|
|
b.timestampsData = b.timestampsData[:0]
|
|
b.valuesData = b.valuesData[:0]
|
|
}
|
|
|
|
// CopyFrom copies src to b.
|
|
func (b *Block) CopyFrom(src *Block) {
|
|
b.bh = src.bh
|
|
b.nextIdx = 0
|
|
b.timestamps = append(b.timestamps[:0], src.timestamps[src.nextIdx:]...)
|
|
b.values = append(b.values[:0], src.values[src.nextIdx:]...)
|
|
|
|
b.headerData = append(b.headerData[:0], src.headerData...)
|
|
b.timestampsData = append(b.timestampsData[:0], src.timestampsData...)
|
|
b.valuesData = append(b.valuesData[:0], src.valuesData...)
|
|
}
|
|
|
|
func getBlock() *Block {
|
|
v := blockPool.Get()
|
|
if v == nil {
|
|
return &Block{}
|
|
}
|
|
return v.(*Block)
|
|
}
|
|
|
|
func putBlock(b *Block) {
|
|
b.Reset()
|
|
blockPool.Put(b)
|
|
}
|
|
|
|
var blockPool sync.Pool
|
|
|
|
func (b *Block) fixupTimestamps() {
|
|
b.bh.MinTimestamp = b.timestamps[b.nextIdx]
|
|
b.bh.MaxTimestamp = b.timestamps[len(b.timestamps)-1]
|
|
}
|
|
|
|
// RowsCount returns the number of rows in the block.
|
|
func (b *Block) RowsCount() int {
|
|
return int(b.bh.RowsCount)
|
|
}
|
|
|
|
// Init initializes b with the given tsid, timestamps, values and scale.
|
|
func (b *Block) Init(tsid *TSID, timestamps, values []int64, scale int16, precisionBits uint8) {
|
|
b.Reset()
|
|
b.bh.TSID = *tsid
|
|
b.bh.Scale = scale
|
|
b.bh.PrecisionBits = precisionBits
|
|
b.timestamps = append(b.timestamps[:0], timestamps...)
|
|
b.values = append(b.values[:0], values...)
|
|
}
|
|
|
|
// nextRow advances to the next row.
|
|
//
|
|
// Returns false if there are no more rows in the block.
|
|
func (b *Block) nextRow() bool {
|
|
if b.nextIdx == len(b.values) {
|
|
return false
|
|
}
|
|
b.nextIdx++
|
|
return true
|
|
}
|
|
|
|
// assertUnmarshaled makes sure the block is unmarshaled.
|
|
func (b *Block) assertUnmarshaled() {
|
|
if len(b.valuesData) > 0 {
|
|
logger.Panicf("BUG: valuesData must be empty; got %d bytes", len(b.valuesData))
|
|
}
|
|
if len(b.timestampsData) > 0 {
|
|
logger.Panicf("BUG: timestampsData must be empty; got %d bytes", len(b.timestampsData))
|
|
}
|
|
if len(b.values) != len(b.timestamps) {
|
|
logger.Panicf("BUG: the number of values must match the number of timestamps; got %d vs %d", len(b.values), len(b.timestamps))
|
|
}
|
|
if b.nextIdx > len(b.values) {
|
|
logger.Panicf("BUG: nextIdx cannot exceed the number of values; got %d vs %d", b.nextIdx, len(b.values))
|
|
}
|
|
}
|
|
|
|
// assertMergeable makes sure b and ib are mergeable, i.e. they have the same
|
|
// tsid and scale.
|
|
func (b *Block) assertMergeable(ib *Block) {
|
|
if b.bh.TSID.MetricID != ib.bh.TSID.MetricID {
|
|
logger.Panicf("BUG: unequal TSID: %q vs %q", &b.bh.TSID, &ib.bh.TSID)
|
|
}
|
|
if b.bh.Scale != ib.bh.Scale {
|
|
logger.Panicf("BUG: unequal Scale: %d vs %d", b.bh.Scale, ib.bh.Scale)
|
|
}
|
|
}
|
|
|
|
// tooBig returns true if the block is too big to be extended.
|
|
func (b *Block) tooBig() bool {
|
|
if b.bh.RowsCount >= maxRowsPerBlock || len(b.values[b.nextIdx:]) >= maxRowsPerBlock {
|
|
return true
|
|
}
|
|
if len(b.valuesData) >= maxBlockSize {
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
func (b *Block) deduplicateSamplesDuringMerge() {
|
|
if !isDedupEnabled() {
|
|
// Deduplication is disabled
|
|
return
|
|
}
|
|
// Unmarshal block if it isn't unmarshaled yet in order to apply the de-duplication to unmarshaled samples.
|
|
if err := b.UnmarshalData(); err != nil {
|
|
logger.Panicf("FATAL: cannot unmarshal block: %s", err)
|
|
}
|
|
srcTimestamps := b.timestamps[b.nextIdx:]
|
|
if len(srcTimestamps) < 2 {
|
|
// Nothing to dedup.
|
|
return
|
|
}
|
|
dedupInterval := GetDedupInterval()
|
|
if dedupInterval <= 0 {
|
|
// Deduplication is disabled.
|
|
return
|
|
}
|
|
srcValues := b.values[b.nextIdx:]
|
|
timestamps, values := deduplicateSamplesDuringMerge(srcTimestamps, srcValues, dedupInterval)
|
|
dedups := len(srcTimestamps) - len(timestamps)
|
|
atomic.AddUint64(&dedupsDuringMerge, uint64(dedups))
|
|
b.timestamps = b.timestamps[:b.nextIdx+len(timestamps)]
|
|
b.values = b.values[:b.nextIdx+len(values)]
|
|
}
|
|
|
|
var dedupsDuringMerge uint64
|
|
|
|
func (b *Block) rowsCount() int {
|
|
if len(b.values) == 0 {
|
|
return int(b.bh.RowsCount)
|
|
}
|
|
return len(b.values[b.nextIdx:])
|
|
}
|
|
|
|
// MarshalData marshals the block into binary representation.
|
|
func (b *Block) MarshalData(timestampsBlockOffset, valuesBlockOffset uint64) ([]byte, []byte, []byte) {
|
|
if len(b.values) == 0 {
|
|
// The data has been already marshaled.
|
|
|
|
// b.valuesData and b.timestampsData may be empty for certain
|
|
// b.bh.*MarshalType values, so don't check them.
|
|
|
|
if b.nextIdx != 0 {
|
|
logger.Panicf("BUG: nextIdx must be zero; got %d", b.nextIdx)
|
|
}
|
|
if int(b.bh.TimestampsBlockSize) != len(b.timestampsData) {
|
|
logger.Panicf("BUG: invalid TimestampsBlockSize; got %d; expecting %d", b.bh.TimestampsBlockSize, len(b.timestampsData))
|
|
}
|
|
if int(b.bh.ValuesBlockSize) != len(b.valuesData) {
|
|
logger.Panicf("BUG: invalid ValuesBlockSize; got %d; expecting %d", b.bh.ValuesBlockSize, len(b.valuesData))
|
|
}
|
|
if b.bh.RowsCount <= 0 {
|
|
logger.Panicf("BUG: RowsCount must be greater than 0; got %d", b.bh.RowsCount)
|
|
}
|
|
|
|
// headerData must be always recreated, since it contains timestampsBlockOffset and valuesBlockOffset.
|
|
b.bh.TimestampsBlockOffset = timestampsBlockOffset
|
|
b.bh.ValuesBlockOffset = valuesBlockOffset
|
|
b.headerData = b.bh.Marshal(b.headerData[:0])
|
|
|
|
return b.headerData, b.timestampsData, b.valuesData
|
|
}
|
|
|
|
if b.nextIdx > len(b.values) {
|
|
logger.Panicf("BUG: nextIdx cannot exceed values size; got %d vs %d", b.nextIdx, len(b.values))
|
|
}
|
|
|
|
timestamps := b.timestamps[b.nextIdx:]
|
|
values := b.values[b.nextIdx:]
|
|
if len(values) == 0 {
|
|
logger.Panicf("BUG: values cannot be empty; nextIdx=%d, timestampsBlockOffset=%d, valuesBlockOffset=%d", b.nextIdx, timestampsBlockOffset, valuesBlockOffset)
|
|
}
|
|
if len(values) != len(timestamps) {
|
|
logger.Panicf("BUG: the number of values must match the number of timestamps; got %d vs %d", len(values), len(timestamps))
|
|
}
|
|
|
|
b.valuesData, b.bh.ValuesMarshalType, b.bh.FirstValue = encoding.MarshalValues(b.valuesData[:0], values, b.bh.PrecisionBits)
|
|
b.bh.ValuesBlockOffset = valuesBlockOffset
|
|
b.bh.ValuesBlockSize = uint32(len(b.valuesData))
|
|
b.values = b.values[:0]
|
|
|
|
b.timestampsData, b.bh.TimestampsMarshalType, b.bh.MinTimestamp = encoding.MarshalTimestamps(b.timestampsData[:0], timestamps, b.bh.PrecisionBits)
|
|
b.bh.TimestampsBlockOffset = timestampsBlockOffset
|
|
b.bh.TimestampsBlockSize = uint32(len(b.timestampsData))
|
|
b.bh.MaxTimestamp = timestamps[len(timestamps)-1]
|
|
b.timestamps = b.timestamps[:0]
|
|
|
|
b.bh.RowsCount = uint32(len(values))
|
|
b.headerData = b.bh.Marshal(b.headerData[:0])
|
|
|
|
b.nextIdx = 0
|
|
|
|
return b.headerData, b.timestampsData, b.valuesData
|
|
}
|
|
|
|
// UnmarshalData unmarshals block data.
|
|
func (b *Block) UnmarshalData() error {
|
|
// blockHeader (b.bh) must be already unmarshaled.
|
|
|
|
if len(b.values) > 0 {
|
|
// The data has been already unmarshaled.
|
|
if len(b.valuesData) > 0 {
|
|
logger.Panicf("BUG: valuesData must be empty; contains %d bytes", len(b.valuesData))
|
|
}
|
|
if len(b.timestampsData) > 0 {
|
|
logger.Panicf("BUG: timestampsData must be empty; contains %d bytes", len(b.timestampsData))
|
|
}
|
|
return nil
|
|
}
|
|
|
|
if b.bh.RowsCount <= 0 {
|
|
return fmt.Errorf("RowsCount must be greater than 0; got %d", b.bh.RowsCount)
|
|
}
|
|
|
|
var err error
|
|
|
|
b.timestamps, err = encoding.UnmarshalTimestamps(b.timestamps[:0], b.timestampsData, b.bh.TimestampsMarshalType, b.bh.MinTimestamp, int(b.bh.RowsCount))
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if b.bh.PrecisionBits < 64 {
|
|
// Recover timestamps order after lossy compression.
|
|
encoding.EnsureNonDecreasingSequence(b.timestamps, b.bh.MinTimestamp, b.bh.MaxTimestamp)
|
|
} else {
|
|
// Ensure timestamps are in the range [MinTimestamp ... MaxTimestamps] and are ordered.
|
|
if err := checkTimestampsBounds(b.timestamps, b.bh.MinTimestamp, b.bh.MaxTimestamp); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
b.timestampsData = b.timestampsData[:0]
|
|
|
|
b.values, err = encoding.UnmarshalValues(b.values[:0], b.valuesData, b.bh.ValuesMarshalType, b.bh.FirstValue, int(b.bh.RowsCount))
|
|
if err != nil {
|
|
return err
|
|
}
|
|
b.valuesData = b.valuesData[:0]
|
|
|
|
if len(b.timestamps) != len(b.values) {
|
|
return fmt.Errorf("timestamps and values count mismatch; got %d vs %d", len(b.timestamps), len(b.values))
|
|
}
|
|
|
|
b.nextIdx = 0
|
|
|
|
return nil
|
|
}
|
|
|
|
func checkTimestampsBounds(timestamps []int64, minTimestamp, maxTimestamp int64) error {
|
|
if len(timestamps) == 0 {
|
|
return nil
|
|
}
|
|
tsPrev := timestamps[0]
|
|
if tsPrev < minTimestamp {
|
|
return fmt.Errorf("timestamp for the row 0 out of %d rows cannot be smaller than %d; got %d", len(timestamps), minTimestamp, tsPrev)
|
|
}
|
|
for i, ts := range timestamps[1:] {
|
|
if ts < tsPrev {
|
|
return fmt.Errorf("timestamp for the row %d cannot be smaller than the timestamp for the row %d (total %d rows); got %d vs %d",
|
|
i+1, i, len(timestamps), ts, tsPrev)
|
|
}
|
|
tsPrev = ts
|
|
}
|
|
if tsPrev > maxTimestamp {
|
|
return fmt.Errorf("timestamp for the row %d (the last one) cannot be bigger than %d; got %d", len(timestamps)-1, maxTimestamp, tsPrev)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// AppendRowsWithTimeRangeFilter filters samples from b according to tr and appends them to dst*.
|
|
//
|
|
// It is expected that UnmarshalData has been already called on b.
|
|
func (b *Block) AppendRowsWithTimeRangeFilter(dstTimestamps []int64, dstValues []float64, tr TimeRange) ([]int64, []float64) {
|
|
timestamps, values := b.filterTimestamps(tr)
|
|
dstTimestamps = append(dstTimestamps, timestamps...)
|
|
dstValues = decimal.AppendDecimalToFloat(dstValues, values, b.bh.Scale)
|
|
return dstTimestamps, dstValues
|
|
}
|
|
|
|
func (b *Block) filterTimestamps(tr TimeRange) ([]int64, []int64) {
|
|
timestamps := b.timestamps
|
|
|
|
// Skip timestamps smaller than tr.MinTimestamp.
|
|
i := 0
|
|
for i < len(timestamps) && timestamps[i] < tr.MinTimestamp {
|
|
i++
|
|
}
|
|
|
|
// Skip timestamps bigger than tr.MaxTimestamp.
|
|
j := len(timestamps)
|
|
for j > i && timestamps[j-1] > tr.MaxTimestamp {
|
|
j--
|
|
}
|
|
|
|
if i == j {
|
|
return nil, nil
|
|
}
|
|
return timestamps[i:j], b.values[i:j]
|
|
}
|
|
|
|
// MarshalPortable marshals b to dst, so it could be portably migrated to other VictoriaMetrics instance.
|
|
//
|
|
// The marshaled value must be unmarshaled with UnmarshalPortable function.
|
|
func (b *Block) MarshalPortable(dst []byte) []byte {
|
|
b.MarshalData(0, 0)
|
|
dst = b.bh.marshalPortable(dst)
|
|
dst = encoding.MarshalBytes(dst, b.timestampsData)
|
|
dst = encoding.MarshalBytes(dst, b.valuesData)
|
|
return dst
|
|
}
|
|
|
|
// UnmarshalPortable unmarshals block from src to b and returns the remaining tail.
|
|
//
|
|
// It is assumed that the block has been marshaled with MarshalPortable.
|
|
func (b *Block) UnmarshalPortable(src []byte) ([]byte, error) {
|
|
b.Reset()
|
|
src, err := b.bh.unmarshalPortable(src)
|
|
if err != nil {
|
|
return src, err
|
|
}
|
|
src, timestampsData, err := encoding.UnmarshalBytes(src)
|
|
if err != nil {
|
|
return src, fmt.Errorf("cannot read timestampsData: %w", err)
|
|
}
|
|
b.timestampsData = append(b.timestampsData[:0], timestampsData...)
|
|
src, valuesData, err := encoding.UnmarshalBytes(src)
|
|
if err != nil {
|
|
return src, fmt.Errorf("cannot read valuesData: %w", err)
|
|
}
|
|
b.valuesData = append(b.valuesData[:0], valuesData...)
|
|
|
|
if err := b.bh.validate(); err != nil {
|
|
return src, fmt.Errorf("invalid blockHeader: %w", err)
|
|
}
|
|
if err := b.UnmarshalData(); err != nil {
|
|
return src, fmt.Errorf("invalid data: %w", err)
|
|
}
|
|
|
|
return src, nil
|
|
}
|