fix(cache): stop quarantining a healthy database when the native module fails to load

CHANGELOG.md

@@ -17,6 +17,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- **cache**: A cache backend that fails to load (e.g. `better-sqlite3` ABI mismatch after a Node major upgrade, `ERR_DLOPEN_FAILED`) is no longer misclassified as database corruption. Previously the constructor's catch-all renamed the user's healthy `cache.db` to `cache.db.corrupt-<timestamp>` and recreated an empty database — verified to quarantine a 2,646-entry cache that passed `integrity_check`. Native-module load failures now leave the database and its `-wal`/`-shm` sidecars untouched; genuine corruption still triggers the rename-aside recovery. `deepl translate` and `deepl write` degrade to running without a cache (single warning per process, exit code 0) instead of crashing, since the cache backend is loaded lazily behind a warn-once latch; `deepl cache …` subcommands, which cannot run cacheless, fail with an actionable error suggesting a reinstall or matching Node version.
 - **sync**: `deepl sync --frozen` now reports an accurate key count when drift is caused by a newly-added target locale. Previously the message read `Sync drift detected: 0 new, 0 stale keys.` because the frozen branch in `processBucket` short-circuited before promoting current-status keys missing a target-locale translation into `newKeys` — even though `--dry-run` against the same state correctly reported the backfill count. The drift exit code (10) is unchanged. The drift message now also surfaces `deletedKeys` and only mentions nonzero categories, mirroring the success-path summary format.
 
 ### Security

docs/TROUBLESHOOTING.md

@@ -417,16 +417,20 @@ rm ~/.cache/deepl-cli/cache.db   # or ~/.deepl-cli/cache.db for legacy installat
 deepl cache enable
 ```
 
-### "Cache database corrupted" with NODE_MODULE_VERSION mismatch
+### "Translation cache backend failed to load" with NODE_MODULE_VERSION mismatch
 
 **Cause:** The `better-sqlite3` native addon was compiled for a different Node.js version than the one currently running. This happens when you switch Node.js versions (e.g., via `nvm use`, `fnm`, or a system upgrade) without rebuilding native modules.
 
+Translation and write commands keep working with caching disabled for the run; your cache database is not modified. `deepl cache` subcommands fail until the backend loads again.
+
 **Solution:**
 
 ```bash
 npm rebuild better-sqlite3
 ```
 
+or reinstall the CLI / switch back to the Node.js version it was installed with.
+
 ---
 
 ## Document Translation Issues

src/cli/cache-loader.ts

@@ -0,0 +1,47 @@
+/**
+ * Lazy cache-service loader for the CLI entry point.
+ * The cache backend (a native module today, node:sqlite tomorrow) can
+ * fail to load on runtimes it wasn't built for. Commands that merely
+ * benefit from a cache must degrade to running without one instead of
+ * crashing, with a single warning per process.
+ */
+
+import type { CacheService, CacheServiceOptions } from '../storage/cache.js';
+import { Logger } from '../utils/logger.js';
+import { errorMessage } from '../utils/error-message.js';
+import { isNativeModuleLoadError } from '../utils/native-module-error.js';
+
+type CacheModule = Pick<typeof import('../storage/cache.js'), 'CacheService'>;
+
+export function createCacheServiceGetter(
+  getOptions: () => CacheServiceOptions,
+  importCacheModule: () => Promise<CacheModule> = () => import('../storage/cache.js'),
+): () => Promise<CacheService | undefined> {
+  let instance: CacheService | undefined;
+  let unavailable = false;
+
+  return async (): Promise<CacheService | undefined> => {
+    if (instance || unavailable) {
+      return instance;
+    }
+    try {
+      const { CacheService: CacheSvc } = await importCacheModule();
+      instance = CacheSvc.getInstance(getOptions());
+    } catch (error) {
+      unavailable = true;
+      const detail = errorMessage(error);
+      if (isNativeModuleLoadError(error)) {
+        Logger.warn(
+          `Translation cache backend failed to load (${detail}). ` +
+          'Your cache database has not been modified. Caching is disabled for this run. ' +
+          'Reinstall the CLI, or run it with the Node.js version it was installed with, to restore caching.',
+        );
+      } else {
+        Logger.warn(
+          `Translation cache is unavailable (${detail}). Caching is disabled for this run.`,
+        );
+      }
+    }
+    return instance;
+  };
+}

src/cli/commands/register-cache.ts

@@ -3,17 +3,29 @@ import chalk from 'chalk';
 import type { ConfigService } from '../../storage/config.js';
 import type { CacheService } from '../../storage/cache.js';
 import { Logger } from '../../utils/logger.js';
+import { ConfigError } from '../../utils/errors.js';
 
 export function registerCache(
   program: Command,
   deps: {
     getConfigService: () => ConfigService;
-    getCacheService: () => Promise<CacheService>;
+    getCacheService: () => Promise<CacheService | undefined>;
     handleError: (error: unknown) => never;
   },
 ): void {
   const { getConfigService, getCacheService, handleError } = deps;
 
+  async function requireCacheService(): Promise<CacheService> {
+    const cacheService = await getCacheService();
+    if (!cacheService) {
+      throw new ConfigError(
+        'Cache backend is unavailable, so cache commands cannot run.',
+        'Reinstall the CLI, or run it with the Node.js version it was installed with.',
+      );
+    }
+    return cacheService;
+  }
+
   program
     .command('cache')
     .description('Manage translation cache')
@@ -32,7 +44,7 @@ Examples:
         .action(async (options: { format?: string }) => {
           try {
             const { CacheCommand } = await import('./cache.js');
-            const cacheCommand = new CacheCommand(await getCacheService(), getConfigService());
+            const cacheCommand = new CacheCommand(await requireCacheService(), getConfigService());
             const stats = await cacheCommand.stats();
             if (options.format === 'json') {
               Logger.output(JSON.stringify(stats, null, 2));
@@ -61,7 +73,7 @@ Examples:
           try {
             if (options.dryRun) {
               const { CacheCommand } = await import('./cache.js');
-              const cacheCommand = new CacheCommand(await getCacheService(), getConfigService());
+              const cacheCommand = new CacheCommand(await requireCacheService(), getConfigService());
               const stats = await cacheCommand.stats();
               const totalSizeMB = (stats.totalSize / (1024 * 1024)).toFixed(2);
               const lines = [
@@ -83,7 +95,7 @@ Examples:
             }
 
             const { CacheCommand } = await import('./cache.js');
-            const cacheCommand = new CacheCommand(await getCacheService(), getConfigService());
+            const cacheCommand = new CacheCommand(await requireCacheService(), getConfigService());
             await cacheCommand.clear();
             Logger.success(chalk.green('\u2713 Cache cleared successfully'));
           } catch (error) {
@@ -106,7 +118,7 @@ Examples:
             }
 
             const { CacheCommand } = await import('./cache.js');
-            const cacheCommand = new CacheCommand(await getCacheService(), getConfigService());
+            const cacheCommand = new CacheCommand(await requireCacheService(), getConfigService());
             await cacheCommand.enable(maxSizeBytes);
             Logger.success(chalk.green('\u2713 Cache enabled'));
 
@@ -126,7 +138,7 @@ Examples:
         .action(async () => {
           try {
             const { CacheCommand } = await import('./cache.js');
-            const cacheCommand = new CacheCommand(await getCacheService(), getConfigService());
+            const cacheCommand = new CacheCommand(await requireCacheService(), getConfigService());
             await cacheCommand.disable();
             Logger.success(chalk.green('\u2713 Cache disabled'));
           } catch (error) {

src/cli/commands/service-factory.ts

@@ -22,7 +22,7 @@ export interface ServiceDeps {
   createDeepLClient: CreateDeepLClient;
   getApiKeyAndOptions: GetApiKeyAndOptions;
   getConfigService: () => ConfigService;
-  getCacheService: () => Promise<CacheService>;
+  getCacheService: () => Promise<CacheService | undefined>;
   handleError: (error: unknown) => never;
 }
 

src/cli/index.ts

@@ -11,7 +11,7 @@ import { readFileSync } from 'fs';
 import { fileURLToPath } from 'url';
 import { dirname, join, resolve, isAbsolute, extname } from 'path';
 import { ConfigService } from '../storage/config.js';
-import type { CacheService } from '../storage/cache.js';
+import { createCacheServiceGetter } from './cache-loader.js';
 import { resolvePaths } from '../utils/paths.js';
 import type { DeepLClient } from '../api/deepl-client.js';
 import { Logger } from '../utils/logger.js';
@@ -57,22 +57,17 @@ const paths = resolvePaths();
 
 // Create config service - can be overridden by --config flag
 let configService = new ConfigService(paths.configFile);
-let cacheService: CacheService | null = null;
-
-async function getCacheService(): Promise<CacheService> {
-  if (!cacheService) {
-    const { CacheService: CacheSvc } = await import('../storage/cache.js');
-    const configTtl = configService.getValue<number>('cache.ttl');
-    const configMaxSize = configService.getValue<number>('cache.maxSize');
-    cacheService = CacheSvc.getInstance({
-      dbPath: paths.cacheFile,
-      // Config TTL is in seconds, CacheService expects milliseconds
-      ttl: configTtl !== undefined ? configTtl * 1000 : undefined,
-      maxSize: configMaxSize,
-    });
-  }
-  return cacheService;
-}
+
+const getCacheService = createCacheServiceGetter(() => {
+  const configTtl = configService.getValue<number>('cache.ttl');
+  const configMaxSize = configService.getValue<number>('cache.maxSize');
+  return {
+    dbPath: paths.cacheFile,
+    // Config TTL is in seconds, CacheService expects milliseconds
+    ttl: configTtl !== undefined ? configTtl * 1000 : undefined,
+    maxSize: configMaxSize,
+  };
+});
 
 /**
  * Handle error and exit with appropriate exit code

src/services/translation.ts

@@ -6,7 +6,7 @@
 import * as crypto from 'crypto';
 import { DeepLClient, TranslationResult, isTranslationResult, UsageInfo, LanguageInfo } from '../api/deepl-client.js';
 import { ConfigService } from '../storage/config.js';
-import { CacheService } from '../storage/cache.js';
+import type { CacheService } from '../storage/cache.js';
 import { TranslationOptions, Language, TranslationMemory } from '../types/index.js';
 import { Logger } from '../utils/logger.js';
 import { mapWithConcurrency, MULTI_TARGET_CONCURRENCY } from '../utils/concurrency.js';
@@ -40,14 +40,16 @@ export const TRANSLATE_BATCH_SIZE = 50; // DeepL API max texts per request
 export class TranslationService {
   private client: DeepLClient;
   private config: ConfigService;
-  private cache: CacheService;
+  // No cache means "run cacheless" — the CLI passes undefined when the
+  // cache backend is unavailable (see cli/cache-loader.ts).
+  private cache?: CacheService;
   private languageCache: Map<'source' | 'target', { data: LanguageInfo[]; timestamp: number }> = new Map();
   private readonly LANGUAGE_CACHE_TTL = 24 * 60 * 60 * 1000; // 24 hours
 
   constructor(client: DeepLClient, config: ConfigService, cache?: CacheService) {
     this.client = client;
     this.config = config;
-    this.cache = cache ?? CacheService.getInstance();
+    this.cache = cache;
   }
 
   /**
@@ -112,7 +114,7 @@ export class TranslationService {
     const cacheKey = this.generateCacheKey(processedText, translationOptions);
 
     if (shouldUseCache) {
-      const cachedResult = this.cache.get(cacheKey, isTranslationResult);
+      const cachedResult = this.cache?.get(cacheKey, isTranslationResult);
       if (cachedResult) {
         Logger.verbose('[verbose] Cache hit');
         return {
@@ -131,7 +133,7 @@ export class TranslationService {
 
     // Store in cache
     if (shouldUseCache) {
-      this.cache.set(cacheKey, result);
+      this.cache?.set(cacheKey, result);
     }
 
     return {
@@ -198,7 +200,7 @@ export class TranslationService {
       const cacheKey = this.generateCacheKey(text, translationOptions);
 
       if (cacheEnabled) {
-        const cachedResult = this.cache.get(cacheKey, isTranslationResult);
+        const cachedResult = this.cache?.get(cacheKey, isTranslationResult);
         if (cachedResult) {
           results[i] = cachedResult;
           continue;
@@ -281,7 +283,7 @@ export class TranslationService {
         // Cache the result (only once per unique text)
         if (cacheEnabled) {
           const cacheKey = this.generateCacheKey(text, translationOptions);
-          this.cache.set(cacheKey, result);
+          this.cache?.set(cacheKey, result);
         }
       }
     }

src/services/write.ts

@@ -6,7 +6,7 @@
 import * as crypto from 'crypto';
 import { DeepLClient } from '../api/deepl-client.js';
 import { ConfigService } from '../storage/config.js';
-import { CacheService } from '../storage/cache.js';
+import type { CacheService } from '../storage/cache.js';
 import { WriteOptions, WriteImprovement, isWriteImprovementArray } from '../types/index.js';
 import { Logger } from '../utils/logger.js';
 import { ValidationError, ConfigError } from '../utils/errors.js';
@@ -18,7 +18,9 @@ export interface WriteServiceOptions {
 export class WriteService {
   private client: DeepLClient;
   private config: ConfigService;
-  private cache: CacheService;
+  // No cache means "run cacheless" — the CLI passes undefined when the
+  // cache backend is unavailable (see cli/cache-loader.ts).
+  private cache?: CacheService;
 
   constructor(client: DeepLClient, config: ConfigService, cache?: CacheService) {
     if (!client) {
@@ -31,7 +33,7 @@ export class WriteService {
 
     this.client = client;
     this.config = config;
-    this.cache = cache ?? CacheService.getInstance();
+    this.cache = cache;
   }
 
   /**
@@ -62,7 +64,7 @@ export class WriteService {
     const cacheKey = this.generateCacheKey(text, options);
 
     if (shouldUseCache) {
-      const cachedResult = this.cache.get(cacheKey, isWriteImprovementArray);
+      const cachedResult = this.cache?.get(cacheKey, isWriteImprovementArray);
       if (cachedResult) {
         Logger.verbose('[verbose] Cache hit');
         return cachedResult;
@@ -73,7 +75,7 @@ export class WriteService {
     const improvements = await this.client.improveText(text, options);
 
     if (shouldUseCache) {
-      this.cache.set(cacheKey, improvements);
+      this.cache?.set(cacheKey, improvements);
     }
 
     return improvements;

src/storage/cache.ts

@@ -10,6 +10,7 @@ import { resolvePaths } from '../utils/paths.js';
 import { ConfigError } from '../utils/errors.js';
 import { Logger } from '../utils/logger.js';
 import { errorMessage } from '../utils/error-message.js';
+import { isNativeModuleLoadError } from '../utils/native-module-error.js';
 
 export interface CacheServiceOptions {
   dbPath?: string;
@@ -69,6 +70,12 @@ export class CacheService {
     try {
       this.openDatabase(dbPath);
     } catch (error) {
+      // A backend that cannot load (ABI mismatch after a Node upgrade,
+      // missing binding) is not corruption: the database on disk is
+      // healthy, so renaming it aside would throw away a warm cache.
+      if (isNativeModuleLoadError(error)) {
+        throw error;
+      }
       // Corrupted DB: rename-aside rather than unlink, so the user keeps
       // 30 days of cache history (and a forensic artifact) instead of
       // losing both silently. Suffix with a timestamp so repeated

src/utils/native-module-error.ts

@@ -0,0 +1,24 @@
+/**
+ * Classifies errors thrown while loading the cache's storage backend.
+ * These indicate the module itself cannot load (ABI mismatch after a
+ * Node upgrade, missing binding, runtime without node:sqlite) — the
+ * database file on disk is healthy and must not be touched.
+ */
+
+const NATIVE_LOAD_ERROR_CODES = new Set([
+  'ERR_DLOPEN_FAILED',
+  'ERR_UNKNOWN_BUILTIN_MODULE',
+  'MODULE_NOT_FOUND',
+  'ERR_MODULE_NOT_FOUND',
+]);
+
+export function isNativeModuleLoadError(error: unknown): boolean {
+  if (!(error instanceof Error)) {
+    return false;
+  }
+  const code = (error as NodeJS.ErrnoException).code;
+  if (code && NATIVE_LOAD_ERROR_CODES.has(code)) {
+    return true;
+  }
+  return error.message.includes('NODE_MODULE_VERSION');
+}