add hot reloading for fs config

Signed-off-by: abushwang <abushwang@tencent.com>

Author: abushwang

cmd/containerd-stargz-grpc/cfgwatcher.go

@@ -0,0 +1,152 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package main
+
+import (
+	"context"
+	"path/filepath"
+	"reflect"
+	"sync"
+	"time"
+
+	"github.com/containerd/containerd/v2/core/snapshots"
+	"github.com/containerd/log"
+	fsconfig "github.com/containerd/stargz-snapshotter/fs/config"
+	"github.com/fsnotify/fsnotify"
+	"github.com/pelletier/go-toml"
+)
+
+// WatchConfig monitors the specified configuration file for changes.
+// It triggers the config reload when a change is detected.
+func WatchConfig(
+	ctx context.Context,
+	filePath string,
+	rs snapshots.Snapshotter,
+	initialConfig *fsconfig.Config,
+) error {
+	absFilePath, err := filepath.Abs(filePath)
+	if err != nil {
+		return err
+	}
+
+	watchDir := filepath.Dir(absFilePath)
+
+	watcher, err := fsnotify.NewWatcher()
+	if err != nil {
+		return err
+	}
+
+	if err := watcher.Add(watchDir); err != nil {
+		watcher.Close()
+		return err
+	}
+
+	log.G(ctx).Infof("started monitoring config file: %s", absFilePath)
+
+	cw := &configWatcher{
+		lastConfig: initialConfig,
+	}
+
+	go func() {
+		defer watcher.Close()
+
+		var (
+			debounceTimer *time.Timer
+			mu            sync.Mutex
+		)
+
+		for {
+			select {
+			case event, ok := <-watcher.Events:
+				if !ok {
+					return
+				}
+
+				if event.Name == absFilePath {
+					// Trigger on Write, Create, Rename, or Chmod events
+					// such as vim, nano, etc.
+					if event.Has(fsnotify.Write) || event.Has(fsnotify.Create) ||
+						event.Has(fsnotify.Rename) || event.Has(fsnotify.Chmod) {
+
+						mu.Lock()
+						if debounceTimer != nil {
+							debounceTimer.Stop()
+						}
+						// Debounce changes with a 50ms delay
+						debounceTimer = time.AfterFunc(50*time.Millisecond, func() {
+							log.G(ctx).Infof("config file modification detected: %s", absFilePath)
+							cw.reload(ctx, absFilePath, rs)
+						})
+						mu.Unlock()
+					}
+				}
+
+			case err, ok := <-watcher.Errors:
+				if !ok {
+					return
+				}
+				log.G(ctx).WithError(err).Error("config watcher encountered an error")
+			}
+		}
+	}()
+
+	return nil
+}
+
+type configWatcher struct {
+	lastConfig *fsconfig.Config
+	mu         sync.Mutex
+}
+
+func (w *configWatcher) reload(ctx context.Context, configPath string, rs snapshots.Snapshotter) {
+	log.G(ctx).Infof("Config file %s changed, reloading...", configPath)
+	var newConfig snapshotterConfig
+	tree, err := toml.LoadFile(configPath)
+	if err != nil {
+		log.G(ctx).WithError(err).Error("failed to reload config file")
+		return
+	}
+	if err := tree.Unmarshal(&newConfig); err != nil {
+		log.G(ctx).WithError(err).Error("failed to unmarshal config")
+		return
+	}
+
+	newFsConfig := newConfig.Config.Config
+
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	if w.lastConfig != nil && reflect.DeepEqual(*w.lastConfig, newFsConfig) {
+		log.G(ctx).Info("Config content unchanged, skipping update")
+		return
+	}
+
+	if updater, ok := rs.(interface {
+		UpdateConfig(context.Context, fsconfig.Config) error
+	}); ok {
+		log.G(ctx).Debugf("applying new config: %+v", newFsConfig)
+		if err := updater.UpdateConfig(ctx, newFsConfig); err != nil {
+			log.G(ctx).WithError(err).Error("failed to update config")
+		} else {
+			log.G(ctx).Info("Config updated successfully")
+			cfgCopy := newFsConfig
+			w.lastConfig = &cfgCopy
+		}
+	} else {
+		log.G(ctx).Warn("snapshotter does not support config update")
+	}
+}

cmd/containerd-stargz-grpc/main.go

@@ -260,6 +260,10 @@ func main() {
 		}
 	}
 
+	if err := WatchConfig(ctx, *configPath, rs, &config.Config.Config); err != nil {
+		log.G(ctx).WithError(err).Warn("failed to start config watcher")
+	}
+
 	cleanup, err := serve(ctx, rpc, *address, rs, config)
 	if err != nil {
 		log.G(ctx).WithError(err).Fatalf("failed to serve snapshotter")

cmd/go.mod

@@ -14,6 +14,7 @@ require (
 	github.com/containerd/stargz-snapshotter/ipfs v0.18.1
 	github.com/coreos/go-systemd/v22 v22.6.0
 	github.com/docker/go-metrics v0.0.1
+	github.com/fsnotify/fsnotify v1.9.0
 	github.com/goccy/go-json v0.10.5
 	github.com/klauspost/compress v1.18.2
 	github.com/opencontainers/go-digest v1.0.0
@@ -56,7 +57,6 @@ require (
 	github.com/docker/go-units v0.5.0 // indirect
 	github.com/emicklei/go-restful/v3 v3.13.0 // indirect
 	github.com/felixge/httpsnoop v1.0.4 // indirect
-	github.com/fsnotify/fsnotify v1.9.0 // indirect
 	github.com/fxamacker/cbor/v2 v2.9.0 // indirect
 	github.com/go-logr/logr v1.4.3 // indirect
 	github.com/go-logr/stdr v1.2.2 // indirect

docs/overview.md

@@ -260,6 +260,14 @@ insecure = true
 
 The config file can be passed to stargz snapshotter using `containerd-stargz-grpc`'s `--config` option.
 
+## Configuration hot reload
+
+[Fs configurations](/fs/config/config.go) supports hot reloading. When the configuration file is modified, the snapshotter detects the change and applies the new configuration without restarting the process.
+This enables instant performance tuning (e.g. concurrency, timeouts) without I/O suspension, and allows updating FUSE parameters that cannot be changed by simply restarting the main process when FUSE manager is enabled.
+
+Note that other configurations (e.g. `proxy_plugins`, `fuse_manager`, `resolver`, `mount_options`) require a restart to take effect.
+Also, some specific fields in `[stargz]` section (e.g. `no_prometheus`) do not support hot reloading and changes to them will be ignored until restart.
+
 ## Make your remote snapshotter
 
 It isn't difficult for you to implement your remote snapshotter using [our general snapshotter package](/snapshot) without considering the protocol between that and containerd.

fs/fs.go

@@ -77,6 +77,18 @@ var (
 	metricsCtr *layermetrics.Controller
 )
 
+func applyDefaults(cfg *config.Config) {
+	if cfg.MaxConcurrency == 0 {
+		cfg.MaxConcurrency = defaultMaxConcurrency
+	}
+	if cfg.AttrTimeout == 0 {
+		cfg.AttrTimeout = int64(defaultFuseTimeout.Seconds())
+	}
+	if cfg.EntryTimeout == 0 {
+		cfg.EntryTimeout = int64(defaultFuseTimeout.Seconds())
+	}
+}
+
 type Option func(*options)
 
 type options struct {
@@ -132,20 +144,11 @@ func NewFilesystem(root string, cfg config.Config, opts ...Option) (_ snapshot.F
 	for _, o := range opts {
 		o(&fsOpts)
 	}
-	maxConcurrency := cfg.MaxConcurrency
-	if maxConcurrency == 0 {
-		maxConcurrency = defaultMaxConcurrency
-	}
+	applyDefaults(&cfg)
 
+	maxConcurrency := cfg.MaxConcurrency
 	attrTimeout := time.Duration(cfg.AttrTimeout) * time.Second
-	if attrTimeout == 0 {
-		attrTimeout = defaultFuseTimeout
-	}
-
 	entryTimeout := time.Duration(cfg.EntryTimeout) * time.Second
-	if entryTimeout == 0 {
-		entryTimeout = defaultFuseTimeout
-	}
 
 	metadataStore := fsOpts.metadataStore
 	if metadataStore == nil {
@@ -205,6 +208,7 @@ type filesystem struct {
 	debug                 bool
 	layer                 map[string]layer.Layer
 	layerMu               sync.Mutex
+	configMu              sync.RWMutex
 	backgroundTaskManager *task.BackgroundTaskManager
 	allowNoVerification   bool
 	disableVerification   bool
@@ -214,6 +218,23 @@ type filesystem struct {
 	entryTimeout          time.Duration
 }
 
+func (fs *filesystem) UpdateConfig(ctx context.Context, cfg config.Config) error {
+	applyDefaults(&cfg)
+	fs.configMu.Lock()
+	fs.prefetchSize = cfg.PrefetchSize
+	fs.noprefetch = cfg.NoPrefetch
+	fs.noBackgroundFetch = cfg.NoBackgroundFetch
+	fs.debug = cfg.Debug
+	fs.allowNoVerification = cfg.AllowNoVerification
+	fs.disableVerification = cfg.DisableVerification
+	fs.attrTimeout = time.Duration(cfg.AttrTimeout) * time.Second
+	fs.entryTimeout = time.Duration(cfg.EntryTimeout) * time.Second
+	fs.configMu.Unlock()
+
+	fs.resolver.UpdateConfig(ctx, cfg)
+	return nil
+}
+
 func (fs *filesystem) Mount(ctx context.Context, mountpoint string, labels map[string]string) (retErr error) {
 	// Setting the start time to measure the Mount operation duration.
 	start := time.Now()
@@ -233,7 +254,9 @@ func (fs *filesystem) Mount(ctx context.Context, mountpoint string, labels map[s
 		return fmt.Errorf("source must be passed")
 	}
 
+	fs.configMu.RLock()
 	defaultPrefetchSize := fs.prefetchSize
+	fs.configMu.RUnlock()
 	if psStr, ok := labels[config.TargetPrefetchSizeLabel]; ok {
 		if ps, err := strconv.ParseInt(psStr, 10, 64); err == nil {
 			defaultPrefetchSize = ps
@@ -297,7 +320,15 @@ func (fs *filesystem) Mount(ctx context.Context, mountpoint string, labels map[s
 	}()
 
 	// Verify layer's content
-	if fs.disableVerification {
+	fs.configMu.RLock()
+	disableVerification := fs.disableVerification
+	allowNoVerification := fs.allowNoVerification
+	attrTimeout := fs.attrTimeout
+	entryTimeout := fs.entryTimeout
+	debug := fs.debug
+	fs.configMu.RUnlock()
+
+	if disableVerification {
 		// Skip if verification is disabled completely
 		l.SkipVerify()
 		log.G(ctx).Infof("Verification forcefully skipped")
@@ -313,7 +344,7 @@ func (fs *filesystem) Mount(ctx context.Context, mountpoint string, labels map[s
 			return fmt.Errorf("invalid stargz layer: %w", err)
 		}
 		log.G(ctx).Debugf("verified")
-	} else if _, ok := labels[config.TargetSkipVerifyLabel]; ok && fs.allowNoVerification {
+	} else if _, ok := labels[config.TargetSkipVerifyLabel]; ok && allowNoVerification {
 		// If unverified layer is allowed, use it with warning.
 		// This mode is for legacy stargz archives which don't contain digests
 		// necessary for layer verification.
@@ -342,14 +373,14 @@ func (fs *filesystem) Mount(ctx context.Context, mountpoint string, labels map[s
 	// mount the node to the specified mountpoint
 	// TODO: bind mount the state directory as a read-only fs on snapshotter's side
 	rawFS := fusefs.NewNodeFS(node, &fusefs.Options{
-		AttrTimeout:     &fs.attrTimeout,
-		EntryTimeout:    &fs.entryTimeout,
+		AttrTimeout:     &attrTimeout,
+		EntryTimeout:    &entryTimeout,
 		NullPermissions: true,
 	})
 	mountOpts := &fuse.MountOptions{
 		AllowOther:  true,     // allow users other than root&mounter to access fs
 		FsName:      "stargz", // name this filesystem as "stargz"
-		Debug:       fs.debug,
+		Debug:       debug,
 		DirectMount: true,
 	}
 	server, err := fuse.NewServer(rawFS, mountpoint, mountOpts)
@@ -391,7 +422,10 @@ func (fs *filesystem) Check(ctx context.Context, mountpoint string, labels map[s
 	}
 
 	// Wait for prefetch compeletion
-	if !fs.noprefetch {
+	fs.configMu.RLock()
+	noprefetch := fs.noprefetch
+	fs.configMu.RUnlock()
+	if !noprefetch {
 		if err := l.WaitForPrefetchCompletion(); err != nil {
 			log.G(ctx).WithError(err).Warn("failed to sync with prefetch completion")
 		}
@@ -473,12 +507,17 @@ func unmount(target string, flags int) error {
 
 func (fs *filesystem) prefetch(ctx context.Context, l layer.Layer, defaultPrefetchSize int64, start time.Time) {
 	// Prefetch a layer. The first Check() for this layer waits for the prefetch completion.
-	if !fs.noprefetch {
+	fs.configMu.RLock()
+	noprefetch := fs.noprefetch
+	noBackgroundFetch := fs.noBackgroundFetch
+	fs.configMu.RUnlock()
+
+	if !noprefetch {
 		go l.Prefetch(defaultPrefetchSize)
 	}
 
 	// Fetch whole layer aggressively in background.
-	if !fs.noBackgroundFetch {
+	if !noBackgroundFetch {
 		go func() {
 			if err := l.BackgroundFetch(); err == nil {
 				// write log record for the latency between mount start and last on demand fetch