Add data directory locking to NodeFS to prevent multi-process corruption

.changeset/add-nodefs-lock-file.md

@@ -0,0 +1,5 @@
+---
+'@electric-sql/pglite': patch
+---
+
+Add data directory locking to NodeFS to prevent multi-process corruption

packages/pglite/src/fs/nodefs.ts

@@ -4,8 +4,11 @@ import { EmscriptenBuiltinFilesystem, PGDATA } from './base.js'
 import type { PostgresMod } from '../postgresMod.js'
 import { PGlite } from '../pglite.js'
 
+// TODO: Add locking for browser backends via Web Locks API
+
 export class NodeFS extends EmscriptenBuiltinFilesystem {
   protected rootDir: string
+  #lockFd: number | null = null
 
   constructor(dataDir: string) {
     super(dataDir)
@@ -17,6 +20,9 @@ export class NodeFS extends EmscriptenBuiltinFilesystem {
 
   async init(pg: PGlite, opts: Partial<PostgresMod>) {
     this.pg = pg
+
+    this.#acquireLock()
+
     const options: Partial<PostgresMod> = {
       ...opts,
       preRun: [
@@ -31,7 +37,66 @@ export class NodeFS extends EmscriptenBuiltinFilesystem {
     return { emscriptenOpts: options }
   }
 
+  // Lock file is a sibling (mydb.lock) to avoid polluting the PG data dir
+  #acquireLock() {
+    const lockPath = this.rootDir + '.lock'
+
+    if (fs.existsSync(lockPath)) {
+      try {
+        const content = fs.readFileSync(lockPath, 'utf-8').trim()
+        const lines = content.split('\n')
+        const pid = parseInt(lines[0], 10)
+
+        if (pid && !isNaN(pid) && this.#isProcessAlive(pid)) {
+          throw new Error(
+            `PGlite data directory "${this.rootDir}" is already in use by another instance (PID ${pid}). ` +
+              `Close the other instance or use a different data directory. ` +
+              `Delete "${lockPath}" if PID ${pid} is no longer running.`,
+          )
+        }
+        // Stale lock from a dead process — safe to take over
+      } catch (e) {
+        // Re-throw lock errors, ignore parse errors (corrupt lock file = stale)
+        if (e instanceof Error && e.message.includes('already in use')) {
+          throw e
+        }
+      }
+    }
+
+    // Write our PID to the lock file and keep the fd open
+    this.#lockFd = fs.openSync(lockPath, 'w')
+    fs.writeSync(this.#lockFd, `${process.pid}\n${Date.now()}\n`)
+  }
+
+  #releaseLock() {
+    if (this.#lockFd !== null) {
+      try {
+        fs.closeSync(this.#lockFd)
+      } catch {
+        // Ignore errors on close
+      }
+      this.#lockFd = null
+
+      const lockPath = this.rootDir + '.lock'
+      try {
+        fs.unlinkSync(lockPath)
+      } catch {
+        // Ignore errors on unlink (dir may already be cleaned up)
+      }
+    }
+  }
+
+  #isProcessAlive(pid: number): boolean {
+    try {
+      process.kill(pid, 0) // signal 0 = check if process exists
+      return true
+    } catch {
+      return false // ESRCH = process doesn't exist
+    }
+  }
+
   async closeFs(): Promise<void> {
+    this.#releaseLock()
     this.pg!.Module.FS.quit()
   }
 }

packages/pglite/tests/nodefs-lock.test.js

@@ -0,0 +1,71 @@
+import { describe, it, expect, afterAll } from 'vitest'
+import { existsSync, writeFileSync, rmSync } from 'node:fs'
+
+const dataDir = `/tmp/pglite-lock-test-${Date.now()}`
+
+afterAll(async () => {
+  if (!process.env.RETAIN_DATA) {
+    for (const p of [dataDir, dataDir + '.lock']) {
+      if (existsSync(p)) rmSync(p, { recursive: true, force: true })
+    }
+  }
+})
+
+describe('NodeFS data directory locking', () => {
+  it('should block a second instance from opening the same data directory', async () => {
+    const { PGlite } = await import('../dist/index.js')
+
+    const db1 = new PGlite(dataDir)
+    await db1.waitReady
+
+    // Lock file should exist while db1 is open
+    expect(existsSync(dataDir + '.lock')).toBe(true)
+
+    // Second instance on same dir must throw
+    let lockError = null
+    try {
+      const db2 = new PGlite(dataDir)
+      await db2.waitReady
+      await db2.close()
+    } catch (err) {
+      lockError = err
+    }
+
+    expect(lockError).not.toBeNull()
+    expect(lockError.message).toContain('already in use')
+    expect(lockError.message).toContain(String(process.pid))
+
+    // First instance should still work fine
+    const result = await db1.query('SELECT 1 as ok')
+    expect(result.rows[0].ok).toBe(1)
+
+    await db1.close()
+  }, 30000)
+
+  it('should allow reopening after the first instance is closed', async () => {
+    const { PGlite } = await import('../dist/index.js')
+
+    // Lock file should be cleaned up after close
+    expect(existsSync(dataDir + '.lock')).toBe(false)
+
+    const db = new PGlite(dataDir)
+    await db.waitReady
+    const result = await db.query('SELECT 1 as ok')
+    expect(result.rows[0].ok).toBe(1)
+    await db.close()
+  }, 30000)
+
+  it('should override a stale lock from a dead process', async () => {
+    const { PGlite } = await import('../dist/index.js')
+
+    // Write a fake lock file with a PID that doesn't exist
+    writeFileSync(dataDir + '.lock', '999999\n0\n')
+
+    // Should succeed — stale lock gets overridden
+    const db = new PGlite(dataDir)
+    await db.waitReady
+    const result = await db.query('SELECT 1 as ok')
+    expect(result.rows[0].ok).toBe(1)
+    await db.close()
+  }, 30000)
+})