mirror of https://github.com/perkeep/perkeep.git
pkg/blobserver/localdisk: simplify code, limit stat concurrency
Don't create an unbounded number of stat goroutines. Change-Id: Ie66cc9c680bd83e649966258a8e7ef09c8af5c62
This commit is contained in:
parent
59ac14caa5
commit
956a0a810b
|
@ -24,12 +24,12 @@ import (
|
|||
"path/filepath"
|
||||
"sort"
|
||||
"strings"
|
||||
"sync"
|
||||
|
||||
"camlistore.org/pkg/blob"
|
||||
)
|
||||
|
||||
type readBlobRequest struct {
|
||||
done <-chan struct{}
|
||||
ch chan<- blob.SizedRef
|
||||
after string
|
||||
remain *int // limit countdown
|
||||
|
@ -45,14 +45,15 @@ type enumerateError struct {
|
|||
}
|
||||
|
||||
func (ee *enumerateError) Error() string {
|
||||
return fmt.Sprintf("Enumerate error: %s: %v", ee.msg, ee.err)
|
||||
return fmt.Sprintf("localdisk enumerate error: %s: %v", ee.msg, ee.err)
|
||||
}
|
||||
|
||||
func (ds *DiskStorage) readBlobs(opts readBlobRequest) error {
|
||||
// readBlobs implements EnumerateBlobs. It calls itself recursively on subdirectories.
|
||||
func (ds *DiskStorage) readBlobs(ctx context.Context, opts readBlobRequest) error {
|
||||
dirFullPath := filepath.Join(opts.dirRoot, opts.pathInto)
|
||||
dir, err := os.Open(dirFullPath)
|
||||
if err != nil {
|
||||
return &enumerateError{"localdisk: opening directory " + dirFullPath, err}
|
||||
return &enumerateError{"opening directory " + dirFullPath, err}
|
||||
}
|
||||
names, err := dir.Readdirnames(-1)
|
||||
dir.Close()
|
||||
|
@ -65,27 +66,40 @@ func (ds *DiskStorage) readBlobs(opts readBlobRequest) error {
|
|||
return nil
|
||||
}
|
||||
if err != nil {
|
||||
return &enumerateError{"localdisk: readdirnames of " + dirFullPath, err}
|
||||
return &enumerateError{"readdirnames of " + dirFullPath, err}
|
||||
}
|
||||
sort.Strings(names)
|
||||
stat := make(map[string]chan interface{}) // gets sent error or os.FileInfo
|
||||
stat := make(map[string]*future) // name -> future<os.FileInfo>
|
||||
|
||||
var toStat []func()
|
||||
for _, name := range names {
|
||||
if skipDir(name) || isShardDir(name) {
|
||||
continue
|
||||
}
|
||||
ch := make(chan interface{}, 1) // 1 in case it's not read
|
||||
name := name
|
||||
stat[name] = ch
|
||||
go func() {
|
||||
fi, err := os.Stat(filepath.Join(dirFullPath, name))
|
||||
fullFile := filepath.Join(dirFullPath, name)
|
||||
f := newFuture(func() (os.FileInfo, error) {
|
||||
fi, err := os.Stat(fullFile)
|
||||
if err != nil {
|
||||
ch <- err
|
||||
} else {
|
||||
ch <- fi
|
||||
return nil, &enumerateError{"stat", err}
|
||||
}
|
||||
}()
|
||||
return fi, nil
|
||||
})
|
||||
stat[name] = f
|
||||
toStat = append(toStat, f.run)
|
||||
}
|
||||
|
||||
// Start pre-statting things.
|
||||
go func() {
|
||||
for _, f := range toStat {
|
||||
ds.statGate.Start()
|
||||
f := f
|
||||
go func() {
|
||||
ds.statGate.Done()
|
||||
f()
|
||||
}()
|
||||
}
|
||||
}()
|
||||
|
||||
for _, name := range names {
|
||||
if *opts.remain == 0 {
|
||||
return nil
|
||||
|
@ -93,30 +107,17 @@ func (ds *DiskStorage) readBlobs(opts readBlobRequest) error {
|
|||
if skipDir(name) {
|
||||
continue
|
||||
}
|
||||
var (
|
||||
fi os.FileInfo
|
||||
err error
|
||||
didStat bool
|
||||
)
|
||||
stat := func() {
|
||||
if didStat {
|
||||
return
|
||||
|
||||
isDir := isShardDir(name)
|
||||
if !isDir {
|
||||
fi, err := stat[name].Get()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
didStat = true
|
||||
fiv := <-stat[name]
|
||||
var ok bool
|
||||
if err, ok = fiv.(error); ok {
|
||||
err = &enumerateError{"localdisk: stat of file " + filepath.Join(dirFullPath, name), err}
|
||||
} else {
|
||||
fi = fiv.(os.FileInfo)
|
||||
}
|
||||
}
|
||||
isDir := func() bool {
|
||||
stat()
|
||||
return fi != nil && fi.IsDir()
|
||||
isDir = fi.IsDir()
|
||||
}
|
||||
|
||||
if isShardDir(name) || isDir() {
|
||||
if isDir {
|
||||
var newBlobPrefix string
|
||||
if opts.blobPrefix == "" {
|
||||
newBlobPrefix = name + "-"
|
||||
|
@ -135,18 +136,22 @@ func (ds *DiskStorage) readBlobs(opts readBlobRequest) error {
|
|||
ropts := opts
|
||||
ropts.blobPrefix = newBlobPrefix
|
||||
ropts.pathInto = opts.pathInto + "/" + name
|
||||
if err := ds.readBlobs(ropts); err != nil {
|
||||
if err := ds.readBlobs(ctx, ropts); err != nil {
|
||||
return err
|
||||
}
|
||||
continue
|
||||
}
|
||||
|
||||
stat()
|
||||
if !strings.HasSuffix(name, ".dat") {
|
||||
continue
|
||||
}
|
||||
|
||||
fi, err := stat[name].Get()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
if !fi.IsDir() && strings.HasSuffix(name, ".dat") {
|
||||
if !fi.IsDir() {
|
||||
blobName := strings.TrimSuffix(name, ".dat")
|
||||
if blobName <= opts.after {
|
||||
continue
|
||||
|
@ -154,12 +159,11 @@ func (ds *DiskStorage) readBlobs(opts readBlobRequest) error {
|
|||
if blobRef, ok := blob.Parse(blobName); ok {
|
||||
select {
|
||||
case opts.ch <- blob.SizedRef{Ref: blobRef, Size: uint32(fi.Size())}:
|
||||
case <-opts.done:
|
||||
return context.Canceled
|
||||
(*opts.remain)--
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
}
|
||||
(*opts.remain)--
|
||||
}
|
||||
continue
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -173,8 +177,7 @@ func (ds *DiskStorage) EnumerateBlobs(ctx context.Context, dest chan<- blob.Size
|
|||
}
|
||||
|
||||
limitMutable := limit
|
||||
return ds.readBlobs(readBlobRequest{
|
||||
done: ctx.Done(),
|
||||
return ds.readBlobs(ctx, readBlobRequest{
|
||||
ch: dest,
|
||||
dirRoot: ds.root,
|
||||
after: after,
|
||||
|
@ -197,3 +200,22 @@ func isShardDir(name string) bool {
|
|||
func isHex(b byte) bool {
|
||||
return ('0' <= b && b <= '9') || ('a' <= b && b <= 'f')
|
||||
}
|
||||
|
||||
// future is an os.FileInfo future.
|
||||
type future struct {
|
||||
once sync.Once
|
||||
f func() (os.FileInfo, error)
|
||||
v os.FileInfo
|
||||
err error
|
||||
}
|
||||
|
||||
func newFuture(f func() (os.FileInfo, error)) *future {
|
||||
return &future{f: f}
|
||||
}
|
||||
|
||||
func (f *future) Get() (os.FileInfo, error) {
|
||||
f.once.Do(f.run)
|
||||
return f.v, f.err
|
||||
}
|
||||
|
||||
func (f *future) run() { f.v, f.err = f.f() }
|
||||
|
|
|
@ -64,6 +64,9 @@ type DiskStorage struct {
|
|||
// systems (Windows) where we don't know the maximum number of open
|
||||
// file descriptors.
|
||||
tmpFileGate *syncutil.Gate
|
||||
|
||||
// statGate limits how many pending Stat calls we have in flight.
|
||||
statGate *syncutil.Gate
|
||||
}
|
||||
|
||||
func (ds *DiskStorage) String() string {
|
||||
|
@ -112,6 +115,7 @@ func New(root string) (*DiskStorage, error) {
|
|||
root: root,
|
||||
dirLockMu: new(sync.RWMutex),
|
||||
gen: local.NewGenerationer(root),
|
||||
statGate: syncutil.NewGate(10), // arbitrary, but bounded; be more clever later?
|
||||
}
|
||||
if err := ds.migrate3to2(); err != nil {
|
||||
return nil, fmt.Errorf("Error updating localdisk format: %v", err)
|
||||
|
|
Loading…
Reference in New Issue