VictoriaMetrics/lib/storage/block_stream_merger.go
Aliaksandr Valialkin e7959094f6 lib/storage: remove prioritizing of merging small parts over merging big parts, since it doesn't work as expected
The prioritizing could lead to big merge starvation, which could end up in too big number of parts that must be merged into big parts.

Multiple big merges may be initiated after the migration from v1.39.0 or v1.39.1. It is OK - these merges should be finished soon,
which should return CPU and disk IO usage to normal levels.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/648
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/618
2020-07-30 19:57:27 +03:00

143 lines
2.7 KiB
Go

package storage
import (
"container/heap"
"fmt"
"io"
)
// blockStreamMerger is used for merging block streams.
type blockStreamMerger struct {
// The current block to work with.
Block *Block
bsrHeap blockStreamReaderHeap
// Whether the call to NextBlock must be no-op.
nextBlockNoop bool
// The last error
err error
}
func (bsm *blockStreamMerger) reset() {
bsm.Block = nil
for i := range bsm.bsrHeap {
bsm.bsrHeap[i] = nil
}
bsm.bsrHeap = bsm.bsrHeap[:0]
bsm.nextBlockNoop = false
bsm.err = nil
}
// Init initializes bsm with the given bsrs.
func (bsm *blockStreamMerger) Init(bsrs []*blockStreamReader) {
bsm.reset()
for _, bsr := range bsrs {
if bsr.NextBlock() {
bsm.bsrHeap = append(bsm.bsrHeap, bsr)
continue
}
if err := bsr.Error(); err != nil {
bsm.err = fmt.Errorf("cannot obtain the next block to merge: %w", err)
return
}
}
if len(bsm.bsrHeap) == 0 {
bsm.err = io.EOF
return
}
heap.Init(&bsm.bsrHeap)
bsm.Block = &bsm.bsrHeap[0].Block
bsm.nextBlockNoop = true
}
// NextBlock stores the next block in bsm.Block.
//
// The blocks are sorted by (TDIS, MinTimestamp). Two subsequent blocks
// for the same TSID may contain overlapped time ranges.
func (bsm *blockStreamMerger) NextBlock() bool {
if bsm.err != nil {
return false
}
if bsm.nextBlockNoop {
bsm.nextBlockNoop = false
return true
}
bsm.err = bsm.nextBlock()
switch bsm.err {
case nil:
return true
case io.EOF:
return false
default:
bsm.err = fmt.Errorf("cannot obtain the next block to merge: %w", bsm.err)
return false
}
}
func (bsm *blockStreamMerger) nextBlock() error {
bsrMin := bsm.bsrHeap[0]
if bsrMin.NextBlock() {
heap.Fix(&bsm.bsrHeap, 0)
bsm.Block = &bsm.bsrHeap[0].Block
return nil
}
if err := bsrMin.Error(); err != nil {
return err
}
heap.Pop(&bsm.bsrHeap)
if len(bsm.bsrHeap) == 0 {
return io.EOF
}
bsm.Block = &bsm.bsrHeap[0].Block
return nil
}
func (bsm *blockStreamMerger) Error() error {
if bsm.err == io.EOF {
return nil
}
return bsm.err
}
type blockStreamReaderHeap []*blockStreamReader
func (bsrh *blockStreamReaderHeap) Len() int {
return len(*bsrh)
}
func (bsrh *blockStreamReaderHeap) Less(i, j int) bool {
x := *bsrh
a, b := &x[i].Block.bh, &x[j].Block.bh
if a.TSID.MetricID == b.TSID.MetricID {
// Fast path for identical TSID values.
return a.MinTimestamp < b.MinTimestamp
}
// Slow path for distinct TSID values.
return a.TSID.Less(&b.TSID)
}
func (bsrh *blockStreamReaderHeap) Swap(i, j int) {
x := *bsrh
x[i], x[j] = x[j], x[i]
}
func (bsrh *blockStreamReaderHeap) Push(x interface{}) {
*bsrh = append(*bsrh, x.(*blockStreamReader))
}
func (bsrh *blockStreamReaderHeap) Pop() interface{} {
a := *bsrh
v := a[len(a)-1]
*bsrh = a[:len(a)-1]
return v
}