Remove atomic int usage and some mutexes, replace with channels

This commit is contained in:
Seednode 2023-09-15 13:51:04 -05:00
parent ba827430bb
commit 0d252d0456
4 changed files with 124 additions and 80 deletions

View File

@ -6,6 +6,7 @@ package cmd
import (
"encoding/gob"
"fmt"
"net/http"
"os"
"sync"
@ -64,14 +65,6 @@ func (cache *fileCache) set(val []string) {
cache.list = make([]string, length)
copy(cache.list, val)
cache.mutex.Unlock()
}
func (cache *fileCache) generate(args []string, formats *types.Types) {
cache.mutex.Lock()
cache.list = []string{}
cache.mutex.Unlock()
fileList(args, &filters{}, "", cache, formats)
if Cache && CacheFile != "" {
cache.Export(CacheFile)
@ -138,7 +131,14 @@ func (cache *fileCache) Import(path string) error {
func serveCacheClear(args []string, cache *fileCache, formats *types.Types) httprouter.Handle {
return func(w http.ResponseWriter, r *http.Request, p httprouter.Params) {
cache.generate(args, formats)
list, err := fileList(args, &filters{}, "", &fileCache{}, formats)
if err != nil {
fmt.Println(err)
return
}
cache.set(list)
w.Header().Set("Content-Type", "text/plain")
@ -146,7 +146,7 @@ func serveCacheClear(args []string, cache *fileCache, formats *types.Types) http
}
}
func registerCacheHandlers(mux *httprouter.Router, args []string, cache *fileCache, formats *types.Types) {
func registerCacheHandlers(mux *httprouter.Router, args []string, cache *fileCache, formats *types.Types) error {
skipIndex := false
if CacheFile != "" {
@ -157,8 +157,15 @@ func registerCacheHandlers(mux *httprouter.Router, args []string, cache *fileCac
}
if !skipIndex {
cache.generate(args, formats)
list, err := fileList(args, &filters{}, "", &fileCache{}, formats)
if err != nil {
return err
}
cache.set(list)
}
register(mux, Prefix+"/clear_cache", serveCacheClear(args, cache, formats))
return nil
}

View File

@ -17,7 +17,6 @@ import (
"strconv"
"strings"
"sync"
"sync/atomic"
"time"
"seedno.de/seednode/roulette/types"
@ -41,22 +40,11 @@ type concurrency struct {
fileScans chan int
}
type files struct {
mutex sync.RWMutex
list []string
}
func (f *files) append(path string) {
f.mutex.Lock()
f.list = append(f.list, path)
f.mutex.Unlock()
}
type scanStats struct {
filesMatched atomic.Uint32
filesSkipped atomic.Uint32
directoriesMatched atomic.Uint32
directoriesSkipped atomic.Uint32
filesMatched int
filesSkipped int
directoriesMatched int
directoriesSkipped int
}
type splitPath struct {
@ -102,7 +90,7 @@ func preparePath(path string) string {
func newFile(paths []string, filters *filters, sortOrder string, regexes *regexes, cache *fileCache, formats *types.Types) (string, error) {
path, err := pickFile(paths, filters, sortOrder, cache, formats)
if err != nil {
return "", nil
return "", err
}
splitPath, err := split(path, regexes)
@ -274,9 +262,9 @@ func pathHasSupportedFiles(path string, formats *types.Types) (bool, error) {
}
}
func pathCount(path string) (uint32, uint32, error) {
var directories uint32 = 0
var files uint32 = 0
func pathCount(path string) (int, int, error) {
var directories = 0
var files = 0
nodes, err := os.ReadDir(path)
if err != nil {
@ -294,9 +282,16 @@ func pathCount(path string) (uint32, uint32, error) {
return files, directories, nil
}
func scanPath(path string, files *files, stats *scanStats, concurrency *concurrency, formats *types.Types) error {
func scanPath(path string, fileChannel chan<- string, statChannel chan<- *scanStats, errorChannel chan<- error, concurrency *concurrency, formats *types.Types) {
var wg sync.WaitGroup
stats := &scanStats{
filesMatched: 0,
filesSkipped: 0,
directoriesMatched: 0,
directoriesSkipped: 0,
}
err := filepath.WalkDir(path, func(p string, info os.DirEntry, err error) error {
if err != nil {
return err
@ -318,34 +313,34 @@ func scanPath(path string, files *files, stats *scanStats, concurrency *concurre
path, err := normalizePath(p)
if err != nil {
fmt.Println(err)
errorChannel <- err
}
if !formats.Validate(path) {
stats.filesSkipped.Add(1)
stats.filesSkipped = stats.filesSkipped + 1
return
}
files.append(path)
fileChannel <- path
stats.filesMatched.Add(1)
stats.filesMatched = stats.filesMatched + 1
}()
case info.IsDir():
files, directories, err := pathCount(p)
if err != nil {
fmt.Println(err)
errorChannel <- err
}
if files > 0 && (files < MinimumFileCount) || (files > MaximumFileCount) {
// This count will not otherwise include the parent directory itself, so increment by one
stats.directoriesSkipped.Add(directories + 1)
stats.filesSkipped.Add(files)
stats.directoriesSkipped = stats.directoriesSkipped + directories + 1
stats.filesSkipped = stats.filesSkipped + files
return filepath.SkipDir
}
stats.directoriesMatched.Add(1)
stats.directoriesMatched = stats.directoriesMatched + 1
}
return err
@ -353,24 +348,26 @@ func scanPath(path string, files *files, stats *scanStats, concurrency *concurre
wg.Wait()
if err != nil {
return err
}
statChannel <- stats
return nil
if err != nil {
errorChannel <- err
}
}
func scanPaths(paths []string, sort string, cache *fileCache, formats *types.Types) []string {
files := &files{
mutex: sync.RWMutex{},
list: []string{},
}
func scanPaths(paths []string, sort string, cache *fileCache, formats *types.Types) ([]string, error) {
var list []string
fileChannel := make(chan string)
statChannel := make(chan *scanStats)
errorChannel := make(chan error)
done := make(chan bool, 1)
stats := &scanStats{
filesMatched: atomic.Uint32{},
filesSkipped: atomic.Uint32{},
directoriesMatched: atomic.Uint32{},
directoriesSkipped: atomic.Uint32{},
filesMatched: 0,
filesSkipped: 0,
directoriesMatched: 0,
directoriesSkipped: 0,
}
concurrency := &concurrency{
@ -393,55 +390,95 @@ func scanPaths(paths []string, sort string, cache *fileCache, formats *types.Typ
wg.Done()
}()
err := scanPath(paths[i], files, stats, concurrency, formats)
if err != nil {
fmt.Println(err)
}
scanPath(paths[i], fileChannel, statChannel, errorChannel, concurrency, formats)
}(i)
}
go func() {
wg.Wait()
done <- true
}()
if stats.filesMatched.Load() < 1 {
Poll:
for {
select {
case p := <-fileChannel:
list = append(list, p)
case s := <-statChannel:
stats.filesMatched = stats.filesMatched + s.filesMatched
stats.filesSkipped = stats.filesSkipped + s.filesSkipped
stats.directoriesMatched = stats.directoriesMatched + s.directoriesMatched
stats.directoriesSkipped = stats.directoriesSkipped + s.directoriesSkipped
case e := <-errorChannel:
return []string{}, e
case <-done:
break Poll
}
}
if stats.filesMatched < 1 {
fmt.Println("No files matched")
return []string{}
return []string{}, nil
}
if Verbose {
fmt.Printf("%s | Indexed %d/%d files across %d/%d directories in %s\n",
time.Now().Format(logDate),
stats.filesMatched.Load(),
stats.filesMatched.Load()+stats.filesSkipped.Load(),
stats.directoriesMatched.Load(),
stats.directoriesMatched.Load()+stats.directoriesSkipped.Load(),
stats.filesMatched,
stats.filesMatched+stats.filesSkipped,
stats.directoriesMatched,
stats.directoriesMatched+stats.directoriesSkipped,
time.Since(startTime),
)
}
return files.list
return list, nil
}
func fileList(paths []string, filters *filters, sort string, cache *fileCache, formats *types.Types) []string {
func fileList(paths []string, filters *filters, sort string, cache *fileCache, formats *types.Types) ([]string, error) {
switch {
case Cache && !cache.isEmpty() && filters.isEmpty():
return cache.List()
return cache.List(), nil
case Cache && !cache.isEmpty() && !filters.isEmpty():
return filters.apply(cache.List())
return filters.apply(cache.List()), nil
case Cache && cache.isEmpty() && !filters.isEmpty():
cache.set(scanPaths(paths, sort, cache, formats))
return filters.apply(cache.List())
list, err := scanPaths(paths, sort, cache, formats)
if err != nil {
return []string{}, err
}
cache.set(list)
return filters.apply(cache.List()), nil
case Cache && cache.isEmpty() && filters.isEmpty():
cache.set(scanPaths(paths, sort, cache, formats))
return cache.List()
list, err := scanPaths(paths, sort, cache, formats)
if err != nil {
return []string{}, err
}
cache.set(list)
return cache.List(), nil
case !Cache && !filters.isEmpty():
return filters.apply(scanPaths(paths, sort, cache, formats))
list, err := scanPaths(paths, sort, cache, formats)
if err != nil {
return []string{}, err
}
return filters.apply(list), nil
default:
return scanPaths(paths, sort, cache, formats)
list, err := scanPaths(paths, sort, cache, formats)
if err != nil {
return []string{}, err
}
return list, nil
}
}
func pickFile(args []string, filters *filters, sort string, cache *fileCache, formats *types.Types) (string, error) {
list := fileList(args, filters, sort, cache, formats)
list, err := fileList(args, filters, sort, cache, formats)
if err != nil {
return "", err
}
fileCount := len(list)

View File

@ -11,7 +11,7 @@ import (
)
const (
ReleaseVersion string = "0.87.0"
ReleaseVersion string = "0.88.0"
)
var (
@ -27,8 +27,8 @@ var (
Handlers bool
Images bool
Info bool
MaximumFileCount uint32
MinimumFileCount uint32
MaximumFileCount int
MinimumFileCount int
PageLength uint32
Port uint16
Prefix string
@ -77,8 +77,8 @@ func init() {
rootCmd.Flags().BoolVar(&Handlers, "handlers", false, "display registered handlers (for debugging)")
rootCmd.Flags().BoolVar(&Images, "images", false, "enable support for image files")
rootCmd.Flags().BoolVarP(&Info, "info", "i", false, "expose informational endpoints")
rootCmd.Flags().Uint32Var(&MaximumFileCount, "maximum-files", 1<<32-1, "skip directories with file counts above this value")
rootCmd.Flags().Uint32Var(&MinimumFileCount, "minimum-files", 1, "skip directories with file counts below this value")
rootCmd.Flags().IntVar(&MaximumFileCount, "maximum-files", 1<<32-1, "skip directories with file counts above this value")
rootCmd.Flags().IntVar(&MinimumFileCount, "minimum-files", 1, "skip directories with file counts below this value")
rootCmd.Flags().Uint32Var(&PageLength, "page-length", 0, "pagination length for info pages")
rootCmd.Flags().Uint16VarP(&Port, "port", "p", 8080, "port to listen on")
rootCmd.Flags().StringVar(&Prefix, "prefix", "/", "root path for http handlers (for reverse proxying)")

Binary file not shown.