lib/fs: More efficient casefs cache (#6974)

This changes the cache to cache less things, yet retain the required efficiency for our walk usecase. This uses less memory. Specifically, instead of keeping result and child caches for each path level, only keep a single cached child. In practice our operations are depth-first, or almost depth-first, and then we retain the same hit ratio for a smaller cache size. I improved the benchmark so that it counts the Lstat and DirNames operations performed, and they do not change significantly. The amount of allocated memory is reduced by 20% and the walk itself is actually slightly faster. This also removes the clear based on number of cached names (as that is not a thing any more) and the timer based clear (which was unused). This means we'll retain the last cache state forever until it's cleared by a write operation, but we did that before too and that state is now a lot smaller... The overhead compared to not using a casefs, for our typical "double walk" (walk the tree then stat everything again) is 2x the dirnames we would otherwise call, and no overhead on the stats (unchanged from old implementation) ``` name old time/op new time/op delta WalkCaseFakeFS100k/rawfs-8 306ms ± 1% 305ms ± 2% ~ (p=0.182 n=9+10) WalkCaseFakeFS100k/casefs-8 579ms ± 5% 557ms ± 1% -3.77% (p=0.000 n=10+10) name old B/entry new B/entry delta WalkCaseFakeFS100k/rawfs-8 590 ± 0% 590 ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 1.09k ± 0% 0.87k ± 0% -19.98% (p=0.000 n=10+10) name old DirNames/entry new DirNames/entry delta WalkCaseFakeFS100k/rawfs-8 0.51 ± 0% 0.51 ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 1.02 ± 0% 1.02 ± 0% ~ (all equal) name old DirNames/op new DirNames/op delta WalkCaseFakeFS100k/rawfs-8 51.2k ± 0% 51.2k ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 102k ± 0% 102k ± 0% ~ (all equal) name old Lstat/entry new Lstat/entry delta WalkCaseFakeFS100k/rawfs-8 3.02 ± 0% 3.02 ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 3.02 ± 0% 3.02 ± 0% ~ (all equal) name old Lstat/op new Lstat/op delta WalkCaseFakeFS100k/rawfs-8 302k ± 0% 302k ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 302k ± 0% 302k ± 0% ~ (all equal) name old allocs/entry new allocs/entry delta WalkCaseFakeFS100k/rawfs-8 15.7 ± 0% 15.7 ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 27.5 ± 0% 26.1 ± 0% -5.09% (p=0.000 n=10+10) name old ns/entry new ns/entry delta WalkCaseFakeFS100k/rawfs-8 2.02k ± 1% 2.02k ± 2% ~ (p=0.163 n=9+10) WalkCaseFakeFS100k/casefs-8 3.83k ± 5% 3.68k ± 1% -3.77% (p=0.000 n=10+10) name old alloc/op new alloc/op delta WalkCaseFakeFS100k/rawfs-8 89.2MB ± 0% 89.2MB ± 0% ~ (p=0.364 n=9+10) WalkCaseFakeFS100k/casefs-8 164MB ± 0% 131MB ± 0% -19.97% (p=0.000 n=10+10) name old allocs/op new allocs/op delta WalkCaseFakeFS100k/rawfs-8 2.38M ± 0% 2.38M ± 0% ~ (all equal) WalkCaseFakeFS100k/casefs-8 4.16M ± 0% 3.95M ± 0% -5.05% (p=0.000 n=10+10) ```
2020-09-09 14:38:39 +02:00
parent e3cd9219b8
commit 780fb3bac1
5 changed files with 157 additions and 121 deletions
@@ -16,12 +16,9 @@ import (
 	"time"
 )

-// Both values were chosen by magic.
 const (
+	// How long to consider cached dirnames valid
 	caseCacheTimeout = time.Second
-	// When the number of names (all lengths of []string from DirNames)
-	// exceeds this, we drop the cache.
-	caseMaxCachedNames = 1 << 20
 )

 type ErrCaseConflict struct {
@@ -47,10 +44,45 @@ type fskey struct {
 	uri    string
 }

-var (
-	caseFilesystems    = make(map[fskey]Filesystem)
-	caseFilesystemsMut sync.Mutex
-)
+// caseFilesystemRegistry caches caseFilesystems and runs a routine to drop
+// their cache every now and then.
+type caseFilesystemRegistry struct {
+	fss          map[fskey]*caseFilesystem
+	mut          sync.Mutex
+	startCleaner sync.Once
+}
+
+func (r *caseFilesystemRegistry) get(fs Filesystem) *caseFilesystem {
+	r.mut.Lock()
+	defer r.mut.Unlock()
+
+	k := fskey{fs.Type(), fs.URI()}
+	caseFs, ok := r.fss[k]
+	if !ok {
+		caseFs = &caseFilesystem{
+			Filesystem: fs,
+			realCaser:  newDefaultRealCaser(fs),
+		}
+		r.fss[k] = caseFs
+		r.startCleaner.Do(func() {
+			go r.cleaner()
+		})
+	}
+
+	return caseFs
+}
+
+func (r *caseFilesystemRegistry) cleaner() {
+	for range time.NewTicker(time.Minute).C {
+		r.mut.Lock()
+		for _, caseFs := range r.fss {
+			caseFs.dropCache()
+		}
+		r.mut.Unlock()
+	}
+}
+
+var globalCaseFilesystemRegistry = caseFilesystemRegistry{fss: make(map[fskey]*caseFilesystem)}

 // caseFilesystem is a BasicFilesystem with additional checks to make a
 // potentially case insensitive underlying FS behave like it's case-sensitive.
@@ -66,18 +98,7 @@ type caseFilesystem struct {
 // case-sensitive one. However it will add some overhead and thus shouldn't be
 // used if the filesystem is known to already behave case-sensitively.
 func NewCaseFilesystem(fs Filesystem) Filesystem {
-	caseFilesystemsMut.Lock()
-	defer caseFilesystemsMut.Unlock()
-	k := fskey{fs.Type(), fs.URI()}
-	if caseFs, ok := caseFilesystems[k]; ok {
-		return caseFs
-	}
-	caseFs := &caseFilesystem{
-		Filesystem: fs,
-		realCaser:  newDefaultRealCaser(fs),
-	}
-	caseFilesystems[k] = caseFs
-	return caseFs
+	return globalCaseFilesystemRegistry.get(fs)
 }

 func (f *caseFilesystem) Chmod(name string, mode FileMode) error {
@@ -308,21 +329,16 @@ func (f *caseFilesystem) checkCaseExisting(name string) error {
 }

 type defaultRealCaser struct {
-	fs        Filesystem
-	root      *caseNode
-	count     int
-	timer     *time.Timer
-	timerStop chan struct{}
-	mut       sync.RWMutex
+	fs   Filesystem
+	root *caseNode
+	mut  sync.RWMutex
 }

 func newDefaultRealCaser(fs Filesystem) *defaultRealCaser {
 	caser := &defaultRealCaser{
-		fs:    fs,
-		root:  &caseNode{name: "."},
-		timer: time.NewTimer(0),
+		fs:   fs,
+		root: &caseNode{name: "."},
 	}
-	<-caser.timer.C
 	return caser
 }

@@ -333,84 +349,49 @@ func (r *defaultRealCaser) realCase(name string) (string, error) {
 	}

 	r.mut.Lock()
-	defer func() {
-		if r.count > caseMaxCachedNames {
-			select {
-			case r.timerStop <- struct{}{}:
-			default:
-			}
-			r.dropCacheLocked()
-		}
-		r.mut.Unlock()
-	}()
+	defer r.mut.Unlock()

 	node := r.root
 	for _, comp := range strings.Split(name, string(PathSeparator)) {
-		if node.dirNames == nil {
-			// Haven't called DirNames yet
+		if node.dirNames == nil || node.expires.Before(time.Now()) {
+			// Haven't called DirNames yet, or the node has expired
+
 			var err error
 			node.dirNames, err = r.fs.DirNames(out)
 			if err != nil {
 				return "", err
 			}
+
 			node.dirNamesLower = make([]string, len(node.dirNames))
 			for i, n := range node.dirNames {
 				node.dirNamesLower[i] = UnicodeLowercase(n)
 			}
-			node.children = make(map[string]*caseNode)
-			node.results = make(map[string]*caseNode)
-			r.count += len(node.dirNames)
-		} else if child, ok := node.results[comp]; ok {
-			// Check if this exact name has been queried before to shortcut
-			node = child
-			out = filepath.Join(out, child.name)
-			continue
+
+			node.expires = time.Now().Add(caseCacheTimeout)
+			node.child = nil
 		}
-		// Actually loop dirNames to search for a match
-		n, err := findCaseInsensitiveMatch(comp, node.dirNames, node.dirNamesLower)
-		if err != nil {
-			return "", err
+
+		// If we don't already have a correct cached child, try to find it.
+		if node.child == nil || node.child.name != comp {
+			// Actually loop dirNames to search for a match.
+			n, err := findCaseInsensitiveMatch(comp, node.dirNames, node.dirNamesLower)
+			if err != nil {
+				return "", err
+			}
+			node.child = &caseNode{name: n}
 		}
-		child, ok := node.children[n]
-		if !ok {
-			child = &caseNode{name: n}
-		}
-		node.results[comp] = child
-		node.children[n] = child
-		node = child
-		out = filepath.Join(out, n)
+
+		node = node.child
+		out = filepath.Join(out, node.name)
 	}

 	return out, nil
 }

-func (r *defaultRealCaser) startCaseResetTimerLocked() {
-	r.timerStop = make(chan struct{})
-	r.timer.Reset(caseCacheTimeout)
-	go func() {
-		select {
-		case <-r.timer.C:
-			r.dropCache()
-		case <-r.timerStop:
-			if !r.timer.Stop() {
-				<-r.timer.C
-			}
-			r.mut.Lock()
-			r.timerStop = nil
-			r.mut.Unlock()
-		}
-	}()
-}
-
 func (r *defaultRealCaser) dropCache() {
 	r.mut.Lock()
-	r.dropCacheLocked()
-	r.mut.Unlock()
-}
-
-func (r *defaultRealCaser) dropCacheLocked() {
 	r.root = &caseNode{name: "."}
-	r.count = 0
+	r.mut.Unlock()
 }

 // Both name and the key to children are "Real", case resolved names of the path
@@ -419,10 +400,10 @@ func (r *defaultRealCaser) dropCacheLocked() {
 // case resolved.
 type caseNode struct {
 	name          string
+	expires       time.Time
 	dirNames      []string
 	dirNamesLower []string
-	children      map[string]*caseNode
-	results       map[string]*caseNode
+	child         *caseNode
 }

 func findCaseInsensitiveMatch(name string, names, namesLower []string) (string, error) {