func TestPipelineWalkerWithSplit(t *testing.T) {
if TestWalkerPath == "" {
t.Skipf("walkerpath not set, skipping TestPipelineWalker")
}
var err error
if !filepath.IsAbs(TestWalkerPath) {
TestWalkerPath, err = filepath.Abs(TestWalkerPath)
OK(t, err)
}
before, err := statPath(TestWalkerPath)
OK(t, err)
t.Logf("walking path %s with %d dirs, %d files", TestWalkerPath,
before.dirs, before.files)
// account for top level dir
before.dirs++
after := stats{}
m := sync.Mutex{}
worker := func(wg *sync.WaitGroup, done <-chan struct{}, entCh <-chan pipe.Entry, dirCh <-chan pipe.Dir) {
defer wg.Done()
for {
select {
case e, ok := <-entCh:
if !ok {
// channel is closed
return
}
m.Lock()
after.files++
m.Unlock()
e.Result() <- true
case dir, ok := <-dirCh:
if !ok {
// channel is closed
return
}
// wait for all content
for _, ch := range dir.Entries {
<-ch
}
m.Lock()
after.dirs++
m.Unlock()
dir.Result() <- true
case <-done:
// pipeline was cancelled
return
}
}
}
var wg sync.WaitGroup
done := make(chan struct{})
entCh := make(chan pipe.Entry)
dirCh := make(chan pipe.Dir)
for i := 0; i < maxWorkers; i++ {
wg.Add(1)
go worker(&wg, done, entCh, dirCh)
}
jobs := make(chan pipe.Job, 200)
wg.Add(1)
go func() {
pipe.Split(jobs, dirCh, entCh)
close(entCh)
close(dirCh)
wg.Done()
}()
resCh := make(chan pipe.Result, 1)
pipe.Walk([]string{TestWalkerPath}, acceptAll, done, jobs, resCh)
// wait for all workers to terminate
wg.Wait()
// wait for top-level blob
<-resCh
t.Logf("walked path %s with %d dirs, %d files", TestWalkerPath,
after.dirs, after.files)
Assert(t, before == after, "stats do not match, expected %v, got %v", before, after)
}
// Snapshot creates a snapshot of the given paths. If parentID is set, this is
// used to compare the files to the ones archived at the time this snapshot was
// taken.
func (arch *Archiver) Snapshot(p *Progress, paths []string, parentID *backend.ID) (*Snapshot, backend.ID, error) {
paths = unique(paths)
sort.Sort(baseNameSlice(paths))
debug.Log("Archiver.Snapshot", "start for %v", paths)
debug.RunHook("Archiver.Snapshot", nil)
// signal the whole pipeline to stop
done := make(chan struct{})
var err error
p.Start()
defer p.Done()
// create new snapshot
sn, err := NewSnapshot(paths)
if err != nil {
return nil, backend.ID{}, err
}
sn.Excludes = arch.Excludes
jobs := archivePipe{}
// use parent snapshot (if some was given)
if parentID != nil {
sn.Parent = parentID
// load parent snapshot
parent, err := LoadSnapshot(arch.repo, *parentID)
if err != nil {
return nil, backend.ID{}, err
}
// start walker on old tree
ch := make(chan WalkTreeJob)
go WalkTree(arch.repo, *parent.Tree, done, ch)
jobs.Old = ch
} else {
// use closed channel
ch := make(chan WalkTreeJob)
close(ch)
jobs.Old = ch
}
// start walker
pipeCh := make(chan pipe.Job)
resCh := make(chan pipe.Result, 1)
go func() {
pipe.Walk(paths, arch.SelectFilter, done, pipeCh, resCh)
debug.Log("Archiver.Snapshot", "pipe.Walk done")
}()
jobs.New = pipeCh
ch := make(chan pipe.Job)
go jobs.compare(done, ch)
var wg sync.WaitGroup
entCh := make(chan pipe.Entry)
dirCh := make(chan pipe.Dir)
// split
wg.Add(1)
go func() {
pipe.Split(ch, dirCh, entCh)
debug.Log("Archiver.Snapshot", "split done")
close(dirCh)
close(entCh)
wg.Done()
}()
// run workers
for i := 0; i < maxConcurrency; i++ {
wg.Add(2)
go arch.fileWorker(&wg, p, done, entCh)
go arch.dirWorker(&wg, p, done, dirCh)
}
// run index saver
var wgIndexSaver sync.WaitGroup
stopIndexSaver := make(chan struct{})
wgIndexSaver.Add(1)
go arch.saveIndexes(&wgIndexSaver, stopIndexSaver)
// wait for all workers to terminate
debug.Log("Archiver.Snapshot", "wait for workers")
wg.Wait()
// stop index saver
close(stopIndexSaver)
wgIndexSaver.Wait()
debug.Log("Archiver.Snapshot", "workers terminated")
// receive the top-level tree
root := (<-resCh).(*Node)
debug.Log("Archiver.Snapshot", "root node received: %v", root.Subtree.Str())
sn.Tree = root.Subtree
// save snapshot
id, err := arch.repo.SaveJSONUnpacked(backend.Snapshot, sn)
if err != nil {
return nil, backend.ID{}, err
}
// store ID in snapshot struct
sn.id = &id
debug.Log("Archiver.Snapshot", "saved snapshot %v", id.Str())
// flush repository
err = arch.repo.Flush()
if err != nil {
return nil, backend.ID{}, err
}
// save index
err = arch.repo.SaveIndex()
if err != nil {
debug.Log("Archiver.Snapshot", "error saving index: %v", err)
return nil, backend.ID{}, err
}
debug.Log("Archiver.Snapshot", "saved indexes")
return sn, id, nil
}
func BenchmarkPipelineWalker(b *testing.B) {
if TestWalkerPath == "" {
b.Skipf("walkerpath not set, skipping BenchPipelineWalker")
}
var max time.Duration
m := sync.Mutex{}
fileWorker := func(wg *sync.WaitGroup, done <-chan struct{}, ch <-chan pipe.Entry) {
defer wg.Done()
for {
select {
case e, ok := <-ch:
if !ok {
// channel is closed
return
}
// simulate backup
//time.Sleep(10 * time.Millisecond)
e.Result() <- true
case <-done:
// pipeline was cancelled
return
}
}
}
dirWorker := func(wg *sync.WaitGroup, done <-chan struct{}, ch <-chan pipe.Dir) {
defer wg.Done()
for {
select {
case dir, ok := <-ch:
if !ok {
// channel is closed
return
}
start := time.Now()
// wait for all content
for _, ch := range dir.Entries {
<-ch
}
d := time.Since(start)
m.Lock()
if d > max {
max = d
}
m.Unlock()
dir.Result() <- true
case <-done:
// pipeline was cancelled
return
}
}
}
for i := 0; i < b.N; i++ {
max = 0
done := make(chan struct{})
entCh := make(chan pipe.Entry, 200)
dirCh := make(chan pipe.Dir, 200)
var wg sync.WaitGroup
b.Logf("starting %d workers", maxWorkers)
for i := 0; i < maxWorkers; i++ {
wg.Add(2)
go dirWorker(&wg, done, dirCh)
go fileWorker(&wg, done, entCh)
}
jobs := make(chan pipe.Job, 200)
wg.Add(1)
go func() {
pipe.Split(jobs, dirCh, entCh)
close(entCh)
close(dirCh)
wg.Done()
}()
resCh := make(chan pipe.Result, 1)
pipe.Walk([]string{TestWalkerPath}, acceptAll, done, jobs, resCh)
// wait for all workers to terminate
wg.Wait()
// wait for final result
<-resCh
b.Logf("max duration for a dir: %v", max)
}
}