Compare commits

..

4 Commits

Author SHA1 Message Date
2f55f628f5 v6.2.0
Some checks failed
Default (tags) / security (push) Failing after 16s
Default (tags) / test (push) Failing after 17s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2025-12-10 14:18:40 +00:00
5dda689b4c feat(watchers): Improve Node watcher robustness: file-level inode tracking, abortable restarts, restart race guards, and untracked-file handling 2025-12-10 14:18:40 +00:00
8677f61da1 v6.1.1
Some checks failed
Default (tags) / security (push) Failing after 17s
Default (tags) / test (push) Failing after 19s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2025-12-08 19:31:48 +00:00
0c236d44d3 fix(watchers/watcher.node): Improve Node watcher robustness: inode tracking, ENOSPC detection, enhanced health checks and temp-file handling 2025-12-08 19:31:48 +00:00
5 changed files with 260 additions and 58 deletions

View File

@@ -1,5 +1,25 @@
# Changelog # Changelog
## 2025-12-10 - 6.2.0 - feat(watchers)
Improve Node watcher robustness: file-level inode tracking, abortable restarts, restart race guards, and untracked-file handling
- Add file-level inode tracking to detect delete+recreate (editor atomic saves) and emit correct 'change'/'add' events
- Make restart delays abortable via AbortController so stop() cancels pending restarts and prevents orphan watchers
- Guard against concurrent/dual restarts with restartingPaths to avoid race conditions between health checks and error handlers
- Emit 'unlink' for deletions of previously untracked files (files created after initial scan) and clean up inode state
- Track file inodes during initial scan and update/cleanup inode state on events
- Improve logging for restart/inode/delete+recreate scenarios and update documentation/readme hints to v6.2.0+
## 2025-12-08 - 6.1.1 - fix(watchers/watcher.node)
Improve Node watcher robustness: inode tracking, ENOSPC detection, enhanced health checks and temp-file handling
- Track directory inodes (watchedInodes) and restart watchers if inode changes are detected (addresses stale watchers when directories are replaced).
- Health check now validates inode stability and explicitly detects ENOSPC (inotify max_user_watches) errors, emitting errors and logging a recommended fix command.
- Detect ENOSPC in FSWatcher error events and log guidance to increase inotify limits.
- Clear inode tracking state on watcher stop to avoid stale state across restarts.
- Improve temporary file handling and logging to avoid dropping events for atomic writes (only skip pure temp files and log skipped temp events).
- Documentation (readme.hints.md) updated with robustness notes, known fs.watch limitations, and example logs.
## 2025-12-08 - 6.1.0 - feat(watcher.node) ## 2025-12-08 - 6.1.0 - feat(watcher.node)
Add automatic restart, periodic health checks, and safe event emission to Node watcher; improve logging and stat handling Add automatic restart, periodic health checks, and safe event emission to Node watcher; improve logging and stat handling

View File

@@ -1,6 +1,6 @@
{ {
"name": "@push.rocks/smartwatch", "name": "@push.rocks/smartwatch",
"version": "6.1.0", "version": "6.2.0",
"private": false, "private": false,
"description": "A cross-runtime file watcher with glob pattern support for Node.js, Deno, and Bun.", "description": "A cross-runtime file watcher with glob pattern support for Node.js, Deno, and Bun.",
"main": "dist_ts/index.js", "main": "dist_ts/index.js",

View File

@@ -73,22 +73,40 @@ The `WriteStabilizer` class replaces chokidar's built-in write stabilization:
### Robustness Features (v6.1.0+) ### Robustness Features (v6.1.0+)
The Node.js watcher includes automatic recovery mechanisms: The Node.js watcher includes automatic recovery mechanisms based on learnings from [chokidar](https://github.com/paulmillr/chokidar) and known [fs.watch issues](https://github.com/nodejs/node/issues/47058):
**Auto-restart on failure:** **Auto-restart on failure:**
- Watchers automatically restart when errors occur - Watchers automatically restart when errors occur
- Exponential backoff (1s → 30s max) - Exponential backoff (1s → 30s max)
- Maximum 3 retry attempts before giving up - Maximum 3 retry attempts before giving up
- **v6.2.0+**: Race condition guards prevent orphan watchers when `stop()` is called during restart
**Inode tracking (critical for long-running watchers):**
- `fs.watch()` watches the **inode**, not the path!
- When directories are replaced (git checkout, atomic saves), the inode changes
- Health check detects inode changes and restarts the watcher
- **v6.2.0+**: File-level inode tracking detects delete+recreate (common editor save pattern)
- This is the most common cause of "watcher stops working after some time"
**Health check monitoring:** **Health check monitoring:**
- 30-second periodic health checks - 30-second periodic health checks
- Detects when watched paths disappear - Detects when watched paths disappear
- Triggers automatic restart when issues detected - Detects inode changes (directory replacement)
- Detects ENOSPC errors (inotify limit exceeded)
- **v6.2.0+**: Protected against dual-restart race conditions (health check + error handler)
**ENOSPC detection (Linux inotify limit):**
- Detects when `/proc/sys/fs/inotify/max_user_watches` is exceeded
- Logs fix command: `echo fs.inotify.max_user_watches=524288 | sudo tee -a /etc/sysctl.conf && sudo sysctl -p`
**Error isolation:** **Error isolation:**
- Subscriber errors don't crash the watcher - Subscriber errors don't crash the watcher
- All events emitted via `safeEmit()` with try-catch - All events emitted via `safeEmit()` with try-catch
**Untracked file handling (v6.2.0+):**
- Files created after initial scan are properly detected
- Untracked file deletions emit `unlink` events instead of being silently dropped
**Verbose logging:** **Verbose logging:**
- All lifecycle events logged with `[smartwatch]` prefix - All lifecycle events logged with `[smartwatch]` prefix
- Helps debug watcher issues in production - Helps debug watcher issues in production
@@ -100,8 +118,18 @@ Example log output:
[smartwatch] Starting health check (every 30s) [smartwatch] Starting health check (every 30s)
[smartwatch] Watcher started with 1 active watcher(s) [smartwatch] Watcher started with 1 active watcher(s)
[smartwatch] Health check: 1 watchers active [smartwatch] Health check: 1 watchers active
[smartwatch] Inode changed for ./src: 12345 -> 67890
[smartwatch] fs.watch watches inode, not path - restarting watcher
[smartwatch] File inode changed (delete+recreate): ./src/file.ts
``` ```
### Known fs.watch Limitations
1. **Watches inode, not path** - If a directory is replaced, watcher goes stale
2. **inotify limits on Linux** - Default `max_user_watches` (8192) may be too low
3. **No events for some atomic writes** - Some editors' save patterns may not trigger events
4. **Platform differences** - Linux uses inotify, macOS uses FSEvents/kqueue
### Testing ### Testing
```bash ```bash

View File

@@ -3,6 +3,6 @@
*/ */
export const commitinfo = { export const commitinfo = {
name: '@push.rocks/smartwatch', name: '@push.rocks/smartwatch',
version: '6.1.0', version: '6.2.0',
description: 'A cross-runtime file watcher with glob pattern support for Node.js, Deno, and Bun.' description: 'A cross-runtime file watcher with glob pattern support for Node.js, Deno, and Bun.'
} }

View File

@@ -19,6 +19,21 @@ export class NodeWatcher implements IWatcher {
private restartAttempts: Map<string, number> = new Map(); private restartAttempts: Map<string, number> = new Map();
private healthCheckInterval: NodeJS.Timeout | null = null; private healthCheckInterval: NodeJS.Timeout | null = null;
// Inode tracking - detect when directories are replaced (atomic saves, etc.)
// fs.watch watches the inode, not the path. If inode changes, we need to restart.
private watchedInodes: Map<string, bigint> = new Map();
// File inode tracking - detect when individual files are deleted and recreated
// This is critical: editors delete+recreate files, fs.watch watches OLD inode!
// See: https://github.com/paulmillr/chokidar/issues/972
private fileInodes: Map<string, bigint> = new Map();
// Abort controllers for pending restart delays - prevents orphan watchers on stop()
private restartAbortControllers: Map<string, AbortController> = new Map();
// Prevent concurrent restarts for the same path (health check + error can race)
private restartingPaths: Set<string> = new Set();
// Configuration constants // Configuration constants
private static readonly MAX_RETRIES = 3; private static readonly MAX_RETRIES = 3;
private static readonly INITIAL_RESTART_DELAY = 1000; private static readonly INITIAL_RESTART_DELAY = 1000;
@@ -43,8 +58,19 @@ export class NodeWatcher implements IWatcher {
/** /**
* Restart a watcher after an error with exponential backoff * Restart a watcher after an error with exponential backoff
* Includes guards against:
* - Dual restart race condition (health check + error handler calling simultaneously)
* - Orphan watchers when stop() is called during restart delay
*/ */
private async restartWatcher(basePath: string, error: Error): Promise<void> { private async restartWatcher(basePath: string, error: Error): Promise<void> {
// Guard: Prevent concurrent restarts for the same path
if (this.restartingPaths.has(basePath)) {
console.log(`[smartwatch] Restart already in progress for ${basePath}, skipping`);
return;
}
this.restartingPaths.add(basePath);
try {
const attempts = (this.restartAttempts.get(basePath) || 0) + 1; const attempts = (this.restartAttempts.get(basePath) || 0) + 1;
this.restartAttempts.set(basePath, attempts); this.restartAttempts.set(basePath, attempts);
@@ -72,10 +98,34 @@ export class NodeWatcher implements IWatcher {
this.watchers.delete(basePath); this.watchers.delete(basePath);
} }
// Exponential backoff // Exponential backoff with AbortController (so stop() can cancel)
const delay = this.restartDelays.get(basePath) || NodeWatcher.INITIAL_RESTART_DELAY; const delay = this.restartDelays.get(basePath) || NodeWatcher.INITIAL_RESTART_DELAY;
console.log(`[smartwatch] Waiting ${delay}ms before restart...`); console.log(`[smartwatch] Waiting ${delay}ms before restart...`);
await new Promise((resolve) => setTimeout(resolve, delay));
const abortController = new AbortController();
this.restartAbortControllers.set(basePath, abortController);
try {
await new Promise<void>((resolve, reject) => {
const timeout = setTimeout(resolve, delay);
abortController.signal.addEventListener('abort', () => {
clearTimeout(timeout);
reject(new Error('Restart aborted by stop()'));
});
});
} catch (abortError) {
console.log(`[smartwatch] Restart aborted for ${basePath}`);
return; // stop() was called, don't continue
} finally {
this.restartAbortControllers.delete(basePath);
}
// Double-check we're still watching after the delay
if (!this._isWatching) {
console.log(`[smartwatch] Watcher stopped during restart delay, aborting`);
return;
}
this.restartDelays.set(basePath, Math.min(delay * 2, NodeWatcher.MAX_RESTART_DELAY)); this.restartDelays.set(basePath, Math.min(delay * 2, NodeWatcher.MAX_RESTART_DELAY));
try { try {
@@ -85,20 +135,33 @@ export class NodeWatcher implements IWatcher {
this.restartAttempts.set(basePath, 0); this.restartAttempts.set(basePath, 0);
} catch (restartError) { } catch (restartError) {
console.error(`[smartwatch] Restart failed for ${basePath}:`, restartError); console.error(`[smartwatch] Restart failed for ${basePath}:`, restartError);
// Clear restartingPaths before recursive call
this.restartingPaths.delete(basePath);
this.restartWatcher(basePath, restartError as Error); // Recursive retry this.restartWatcher(basePath, restartError as Error); // Recursive retry
return; // Don't delete from restartingPaths again in finally
}
} finally {
this.restartingPaths.delete(basePath);
} }
} }
/** /**
* Start periodic health checks to detect silent failures * Start periodic health checks to detect silent failures
* Checks for:
* 1. Path no longer exists
* 2. Inode changed (directory was replaced - fs.watch watches inode, not path!)
*/ */
private startHealthCheck(): void { private startHealthCheck(): void {
console.log('[smartwatch] Starting health check (every 30s)'); console.log('[smartwatch] Starting health check (every 30s)');
this.healthCheckInterval = setInterval(async () => { this.healthCheckInterval = setInterval(async () => {
console.log(`[smartwatch] Health check: ${this.watchers.size} watchers active`); console.log(`[smartwatch] Health check: ${this.watchers.size} watchers active`);
for (const [basePath] of this.watchers) { for (const [basePath] of this.watchers) {
const stats = await this.statSafe(basePath); try {
if (!stats && this._isWatching) { const stats = await fs.promises.stat(basePath);
const currentInode = stats.ino;
const previousInode = this.watchedInodes.get(basePath);
if (!stats) {
console.error(`[smartwatch] Health check failed: ${basePath} no longer exists`); console.error(`[smartwatch] Health check failed: ${basePath} no longer exists`);
this.safeEmit({ this.safeEmit({
type: 'error', type: 'error',
@@ -106,6 +169,25 @@ export class NodeWatcher implements IWatcher {
error: new Error('Watched path no longer exists') error: new Error('Watched path no longer exists')
}); });
this.restartWatcher(basePath, new Error('Watched path disappeared')); this.restartWatcher(basePath, new Error('Watched path disappeared'));
} else if (previousInode !== undefined && BigInt(currentInode) !== previousInode) {
// CRITICAL: Inode changed! fs.watch is now watching a stale inode.
// This happens when the directory is replaced (atomic operations, git checkout, etc.)
console.warn(`[smartwatch] Inode changed for ${basePath}: ${previousInode} -> ${currentInode}`);
console.warn('[smartwatch] fs.watch watches inode, not path - restarting watcher');
this.restartWatcher(basePath, new Error('Inode changed - directory was replaced'));
}
} catch (error: any) {
if (error.code === 'ENOENT') {
console.error(`[smartwatch] Health check failed: ${basePath} no longer exists`);
this.restartWatcher(basePath, new Error('Watched path disappeared'));
} else if (error.code === 'ENOSPC') {
// inotify watch limit exceeded - critical system issue
console.error(`[smartwatch] ENOSPC: inotify watch limit exceeded!`);
console.error('[smartwatch] Fix: echo fs.inotify.max_user_watches=524288 | sudo tee -a /etc/sysctl.conf && sudo sysctl -p');
this.safeEmit({ type: 'error', path: basePath, error });
} else {
console.error(`[smartwatch] Health check error for ${basePath}:`, error);
}
} }
} }
}, NodeWatcher.HEALTH_CHECK_INTERVAL); }, NodeWatcher.HEALTH_CHECK_INTERVAL);
@@ -179,6 +261,13 @@ export class NodeWatcher implements IWatcher {
// Stop health check monitoring // Stop health check monitoring
this.stopHealthCheck(); this.stopHealthCheck();
// Abort all pending restart delays (prevents orphan watchers)
for (const [path, controller] of this.restartAbortControllers) {
console.log(`[smartwatch] Aborting pending restart for: ${path}`);
controller.abort();
}
this.restartAbortControllers.clear();
// Cancel all pending debounced emits // Cancel all pending debounced emits
for (const timeout of this.pendingEmits.values()) { for (const timeout of this.pendingEmits.values()) {
clearTimeout(timeout); clearTimeout(timeout);
@@ -193,9 +282,12 @@ export class NodeWatcher implements IWatcher {
this.watchers.clear(); this.watchers.clear();
this.watchedFiles.clear(); this.watchedFiles.clear();
// Clear restart tracking state // Clear all tracking state
this.restartDelays.clear(); this.restartDelays.clear();
this.restartAttempts.clear(); this.restartAttempts.clear();
this.watchedInodes.clear();
this.fileInodes.clear();
this.restartingPaths.clear();
console.log('[smartwatch] Watcher stopped'); console.log('[smartwatch] Watcher stopped');
} }
@@ -215,6 +307,10 @@ export class NodeWatcher implements IWatcher {
} }
if (stats.isDirectory()) { if (stats.isDirectory()) {
// Store inode for health check - fs.watch watches inode, not path!
// If inode changes (directory replaced), watcher becomes stale
this.watchedInodes.set(watchPath, BigInt(stats.ino));
// Watch the directory with recursive option (Node.js 20+ supports this on all platforms) // Watch the directory with recursive option (Node.js 20+ supports this on all platforms)
const watcher = fs.watch( const watcher = fs.watch(
watchPath, watchPath,
@@ -226,8 +322,15 @@ export class NodeWatcher implements IWatcher {
} }
); );
watcher.on('error', (error) => { watcher.on('error', (error: NodeJS.ErrnoException) => {
console.error(`[smartwatch] FSWatcher error event on ${watchPath}:`, error); console.error(`[smartwatch] FSWatcher error event on ${watchPath}:`, error);
// Detect inotify watch limit exceeded - common cause of "stops working"
if (error.code === 'ENOSPC') {
console.error('[smartwatch] CRITICAL: inotify watch limit exceeded!');
console.error('[smartwatch] Fix with: echo fs.inotify.max_user_watches=524288 | sudo tee -a /etc/sysctl.conf && sudo sysctl -p');
}
this.safeEmit({ type: 'error', path: watchPath, error }); this.safeEmit({ type: 'error', path: watchPath, error });
if (this._isWatching) { if (this._isWatching) {
this.restartWatcher(watchPath, error); this.restartWatcher(watchPath, error);
@@ -261,8 +364,14 @@ export class NodeWatcher implements IWatcher {
): void { ): void {
const fullPath = path.join(basePath, filename); const fullPath = path.join(basePath, filename);
// Skip temporary files created by editors (atomic saves) // Skip temporary files - but ONLY pure temp files, not the target of atomic writes
// Atomic writes: editor writes to file.tmp.xxx then renames to file
// We need to detect the final file, so only skip files that ARE temp files
// and haven't been renamed to the real file yet
if (this.isTemporaryFile(fullPath)) { if (this.isTemporaryFile(fullPath)) {
// For temp files, we still want to track if they get renamed TO a real file
// The 'rename' event fires for both source and target, so we'll catch the real file
console.log(`[smartwatch] Skipping temp file event: ${filename}`);
return; return;
} }
@@ -283,6 +392,12 @@ export class NodeWatcher implements IWatcher {
/** /**
* Emit the actual file event after debounce * Emit the actual file event after debounce
*
* Handles file inode tracking to detect delete+recreate scenarios:
* - fs.watch watches the inode, not the path
* - When editors delete+recreate files, the inode changes
* - Without inode tracking, events for the new file would be missed
* - See: https://github.com/paulmillr/chokidar/issues/972
*/ */
private async emitFileEvent( private async emitFileEvent(
fullPath: string, fullPath: string,
@@ -302,38 +417,75 @@ export class NodeWatcher implements IWatcher {
} }
} else { } else {
const wasWatched = this.watchedFiles.has(fullPath); const wasWatched = this.watchedFiles.has(fullPath);
const currentInode = BigInt(stats.ino);
const previousInode = this.fileInodes.get(fullPath);
// Track file inode for delete+recreate detection
this.fileInodes.set(fullPath, currentInode);
this.watchedFiles.add(fullPath); this.watchedFiles.add(fullPath);
// Check if file was recreated with different inode (delete+recreate scenario)
if (wasWatched && previousInode !== undefined && previousInode !== currentInode) {
console.log(`[smartwatch] File inode changed (delete+recreate): ${fullPath}`);
console.log(`[smartwatch] Previous inode: ${previousInode}, current: ${currentInode}`);
// Emit as 'change' since the file content likely changed
this.safeEmit({ type: 'change', path: fullPath, stats });
} else {
this.safeEmit({ this.safeEmit({
type: wasWatched ? 'change' : 'add', type: wasWatched ? 'change' : 'add',
path: fullPath, path: fullPath,
stats stats
}); });
} }
}
} else { } else {
// File doesn't exist - it was deleted // File doesn't exist - it was deleted
if (this.watchedFiles.has(fullPath)) { if (this.watchedFiles.has(fullPath)) {
const wasDir = this.isKnownDirectory(fullPath); const wasDir = this.isKnownDirectory(fullPath);
this.watchedFiles.delete(fullPath); this.watchedFiles.delete(fullPath);
this.fileInodes.delete(fullPath); // Clean up inode tracking
this.safeEmit({ this.safeEmit({
type: wasDir ? 'unlinkDir' : 'unlink', type: wasDir ? 'unlinkDir' : 'unlink',
path: fullPath path: fullPath
}); });
} else {
// Fix #4: File wasn't tracked but was deleted - still emit event
// This handles files created after initial scan that we may have missed
console.log(`[smartwatch] Untracked file deleted: ${fullPath}`);
this.fileInodes.delete(fullPath);
this.safeEmit({ type: 'unlink', path: fullPath });
} }
} }
} else if (eventType === 'change') { } else if (eventType === 'change') {
// File was modified // File was modified
if (stats && !stats.isDirectory()) { if (stats && !stats.isDirectory()) {
const wasWatched = this.watchedFiles.has(fullPath); const wasWatched = this.watchedFiles.has(fullPath);
const currentInode = BigInt(stats.ino);
const previousInode = this.fileInodes.get(fullPath);
// Track file inode
this.fileInodes.set(fullPath, currentInode);
if (!wasWatched) { if (!wasWatched) {
// This is actually an 'add' - file wasn't being watched before // This is actually an 'add' - file wasn't being watched before
this.watchedFiles.add(fullPath); this.watchedFiles.add(fullPath);
this.safeEmit({ type: 'add', path: fullPath, stats }); this.safeEmit({ type: 'add', path: fullPath, stats });
} else if (previousInode !== undefined && previousInode !== currentInode) {
// Inode changed during 'change' event - file was replaced
console.log(`[smartwatch] File replaced (inode change on modify): ${fullPath}`);
this.safeEmit({ type: 'change', path: fullPath, stats });
} else { } else {
this.safeEmit({ type: 'change', path: fullPath, stats }); this.safeEmit({ type: 'change', path: fullPath, stats });
} }
} else if (!stats && this.watchedFiles.has(fullPath)) { } else if (!stats && this.watchedFiles.has(fullPath)) {
// File was deleted // File was deleted
this.watchedFiles.delete(fullPath); this.watchedFiles.delete(fullPath);
this.fileInodes.delete(fullPath);
this.safeEmit({ type: 'unlink', path: fullPath });
} else if (!stats && !this.watchedFiles.has(fullPath)) {
// Fix #4: Untracked file deleted during 'change' event
console.log(`[smartwatch] Untracked file deleted (change event): ${fullPath}`);
this.fileInodes.delete(fullPath);
this.safeEmit({ type: 'unlink', path: fullPath }); this.safeEmit({ type: 'unlink', path: fullPath });
} }
} }
@@ -373,6 +525,8 @@ export class NodeWatcher implements IWatcher {
await this.scanDirectory(fullPath, depth + 1); await this.scanDirectory(fullPath, depth + 1);
} else if (entry.isFile()) { } else if (entry.isFile()) {
this.watchedFiles.add(fullPath); this.watchedFiles.add(fullPath);
// Track file inode for delete+recreate detection
this.fileInodes.set(fullPath, BigInt(stats.ino));
this.safeEmit({ type: 'add', path: fullPath, stats }); this.safeEmit({ type: 'add', path: fullPath, stats });
} }
} }