Improve ingestion performance (by around 40%)

Summary
    fzf --sync --bind load:accept < 27M-lines ran
      1.16 ± 0.01 times faster than fzf-41b3511 --sync --bind load:accept < 27M-lines
      1.44 ± 0.01 times faster than fzf-0.48.1 --sync --bind load:accept < 27M-lines
pull/3707/head
Junegunn Choi 2 months ago
parent 41b3511ad9
commit 5234c3759a
No known key found for this signature in database
GPG Key ID: 254BC280FEF9C627

@ -3,7 +3,7 @@ CHANGELOG
0.49.0 0.49.0
------ ------
- Ingestion performance improved by around 20% - Ingestion performance improved by around 40%
- Added two environment variables exported to the child processes - Added two environment variables exported to the child processes
- `FZF_PREVIEW_LABEL` - `FZF_PREVIEW_LABEL`
- `FZF_BORDER_LABEL` - `FZF_BORDER_LABEL`

@ -1,13 +1,12 @@
package fzf package fzf
import ( import (
"bufio" "bytes"
"context" "context"
"io" "io"
"os" "os"
"os/exec" "os/exec"
"path/filepath" "path/filepath"
"strconv"
"sync" "sync"
"sync/atomic" "sync/atomic"
"time" "time"
@ -112,86 +111,82 @@ func (r *Reader) ReadSource(root string, opts walkerOpts, ignores []string) {
} }
func (r *Reader) feed(src io.Reader) { func (r *Reader) feed(src io.Reader) {
readerSlabSize, ae := strconv.Atoi(os.Getenv("SLAB_KB")) /*
if ae != nil { readerSlabSize, ae := strconv.Atoi(os.Getenv("SLAB_KB"))
readerSlabSize = 128 * 1024 if ae != nil {
} else { readerSlabSize = 128 * 1024
readerSlabSize *= 1024 } else {
} readerSlabSize *= 1024
readerBufferSize, be := strconv.Atoi(os.Getenv("BUF_KB")) }
if be != nil { readerBufferSize, be := strconv.Atoi(os.Getenv("BUF_KB"))
readerBufferSize = 64 * 1024 if be != nil {
} else { readerBufferSize = 64 * 1024
readerBufferSize *= 1024 } else {
} readerBufferSize *= 1024
}
*/
slab := make([]byte, readerSlabSize)
pointer := 0
delim := byte('\n') delim := byte('\n')
if r.delimNil { if r.delimNil {
delim = '\000' delim = '\000'
} }
reader := bufio.NewReaderSize(src, readerBufferSize)
// We do not put a slice longer than 10% of the slab to reduce fragmentation
maxBytes := readerBufferSize / 10
slab := make([]byte, readerSlabSize)
leftover := []byte{}
var err error
for { for {
var frags [][]byte n := 0
fragsLen := 0 scope := slab[:util.Min(len(slab), readerBufferSize)]
for { for i := 0; i < 100; i++ {
bytea, err := reader.ReadSlice(delim) n, err = src.Read(scope)
if err == bufio.ErrBufferFull { if n > 0 || err != nil {
// Could not find the delimiter in the reader buffer. break
// Need to collect the fragments and merge them later. }
frags = append(frags, bytea) }
fragsLen += len(bytea)
} else {
byteaLen := len(bytea)
if err == nil {
// No errors. Found the delimiter.
if util.IsWindows() && byteaLen >= 2 && bytea[byteaLen-2] == byte('\r') {
bytea = bytea[:byteaLen-2]
byteaLen -= 2
} else {
bytea = bytea[:byteaLen-1]
byteaLen--
}
}
itemLen := fragsLen + byteaLen // We're not making any progress after 100 tries. Stop.
pointer += itemLen if n == 0 && err == nil {
var slice []byte break
if itemLen <= maxBytes { // We can use the slab }
// Allocate a new slab if it doesn't fit
if pointer > readerSlabSize {
slab = make([]byte, readerSlabSize)
pointer = itemLen
}
slice = slab[pointer-itemLen : pointer]
} else { // We can't use the slab because the item is too large
slice = make([]byte, itemLen)
}
if len(frags) > 0 { buf := slab[:n]
// Collect the fragments slab = slab[n:]
n := 0
for _, frag := range frags { for len(buf) > 0 {
n += copy(slice[n:], frag) if i := bytes.IndexByte(buf, delim); i >= 0 {
} // Found the delimiter
copy(slice[n:], bytea) slice := buf[:i+1]
} else if byteaLen > 0 { buf = buf[i+1:]
copy(slice, bytea) if util.IsWindows() && len(slice) >= 2 && slice[len(slice)-2] == byte('\r') {
slice = slice[:len(slice)-2]
} else {
slice = slice[:len(slice)-1]
} }
if (err == nil || itemLen > 0) && r.pusher(slice) { if len(leftover) > 0 {
atomic.StoreInt32(&r.event, int32(EvtReadNew)) slice = append(leftover, slice...)
leftover = []byte{}
} }
if err != nil { if (err == nil || len(slice) > 0) && r.pusher(slice) {
return atomic.StoreInt32(&r.event, int32(EvtReadNew))
} }
} else {
// Could not find the delimiter in the buffer
leftover = append(leftover, buf...)
break break
} }
} }
if err == io.EOF {
leftover = append(leftover, buf...)
break
}
if len(slab) == 0 {
slab = make([]byte, readerSlabSize)
}
}
if len(leftover) > 0 && r.pusher(leftover) {
atomic.StoreInt32(&r.event, int32(EvtReadNew))
} }
} }

Loading…
Cancel
Save