@purposeinplay/payload-ai-translate 0.1.0 → 0.1.2

package/dist/lib/per-doc-claim.d.ts

@@ -75,6 +75,7 @@ export type LegacyLockResult = {
 } | {
     acquired: false;
     release?: undefined;
+    reason: 'sibling_running' | 'not_pending';
 } | {
     acquired: 'noop';
     release: () => Promise<void>;

package/dist/lib/per-doc-claim.js

@@ -131,7 +131,8 @@ export async function tryClaimAsLock(args) {
         };
     }
     return {
-        acquired: false
+        acquired: false,
+        reason: r.reason
     };
 }
 // Re-export the units collection slug used by callers (kept here so the

package/dist/tasks/bulk-translate-doc-task.js

@@ -134,7 +134,11 @@ export async function runWorkerForUnit(params) {
     //     batches: a janitor-reset unit whose batch already ended must not
     //     run again.
     // -------------------------------------------------------------------
-    if (batchStatus === 'cancelling' || batchStatus === 'cancelled' || batchStatus === 'reverted') {
+    if (batchStatus === 'cancelling' || batchStatus === 'cancelled' || batchStatus === 'reverted' || batchStatus === 'success' || batchStatus === 'failed' || batchStatus === 'partial') {
+        // Preserve the existing `cancelled_by_admin` reason for the cancel /
+        // revert path (callers + tests rely on it); a batch that finished on
+        // its own (`success`/`failed`/`partial`) gets a distinct reason.
+        const skipReason = batchStatus === 'success' || batchStatus === 'failed' || batchStatus === 'partial' ? 'batch_already_terminal' : 'cancelled_by_admin';
         // Only rewrite units that are still open — a terminal unit (e.g.
         // `failed`) re-fired by a duplicate job must keep its real outcome.
         if (unit.status === 'pending' || unit.status === 'running') {
@@ -143,7 +147,7 @@ export async function runWorkerForUnit(params) {
                 id: unitId,
                 data: {
                     status: 'skipped',
-                    failureMessage: 'cancelled_by_admin',
+                    failureMessage: skipReason,
                     completedAt: now().toISOString()
                 },
                 overrideAccess: true
@@ -155,7 +159,7 @@ export async function runWorkerForUnit(params) {
         await maybeTransitionBatch(payload, batchesSlug, unitsSlug, unit.batchId, callbacks);
         return {
             status: 'skipped',
-            failureMessage: 'cancelled_by_admin'
+            failureMessage: skipReason
         };
     }
     // -------------------------------------------------------------------
@@ -242,11 +246,38 @@ export async function runWorkerForUnit(params) {
         documentId: unit.documentId
     });
     if (claim.acquired === false) {
-        // Couldn't claim — sibling unit for the same doc is running, OR
-        // the unit was already moved out of `pending` by a competing fire
-        // of the same job. Re-enqueue this unit so a later cron tick
-        // retries; don't bump attempts (deferral isn't a translation
-        // failure).
+        // Two very different failure modes, previously conflated into an
+        // unconditional re-enqueue — which is what caused the 2026-06-16
+        // outage: a cancelled/finished batch's leftover jobs re-enqueued
+        // themselves every cron tick forever, each re-reading the source doc.
+        //
+        //  - `not_pending`: this unit is no longer `pending` (it's terminal,
+        //    or another fire of the SAME job claimed it). There is no work to
+        //    retry. Complete the job WITHOUT re-enqueuing — re-enqueuing here
+        //    is the infinite loop. Return the unit's real current status so
+        //    the job output is truthful; never write the unit row (a running
+        //    sibling fire owns its outcome).
+        //  - `sibling_running`: a DIFFERENT locale of the same doc is running
+        //    and this unit is still `pending` and genuinely needs to retry on
+        //    a later tick. This is the only case that warrants a re-enqueue.
+        if (claim.reason === 'not_pending') {
+            const fresh = await payload.findByID({
+                collection: unitsSlug,
+                id: unitId,
+                overrideAccess: true,
+                depth: 0
+            });
+            const freshStatus = fresh?.status;
+            const mapped = freshStatus === 'success' ? 'success' : freshStatus === 'failed' ? 'failed' : 'skipped';
+            earlyLog.event('info', 'bulk.worker.unit.terminal-unit-no-requeue', {
+                unitStatus: freshStatus ?? 'unknown'
+            });
+            return {
+                status: mapped
+            };
+        }
+        // sibling_running — re-enqueue this unit so a later cron tick retries;
+        // don't bump attempts (deferral isn't a translation failure).
         try {
             await payload.jobs.queue({
                 task: BULK_TRANSLATE_DOC_TASK_SLUG,

package/dist/tasks/bulk-translate-janitor.js

@@ -85,12 +85,72 @@ export async function runJanitorSweep(params) {
     const stale = result.docs;
     const batchesSlug = params.batchesSlug ?? DEFAULT_BULK_TRANSLATE_BATCHES_COLLECTION_SLUG;
     const affectedBatchIds = new Set();
-    const noteBatch = (u)=>{
+    const extractBatchId = (u)=>{
         const raw = typeof u.batchId === 'object' && u.batchId !== null ? u.batchId.id : u.batchId;
-        if (raw !== undefined && raw !== null && String(raw).length > 0) {
-            affectedBatchIds.add(String(raw));
+        return raw !== undefined && raw !== null && String(raw).length > 0 ? String(raw) : null;
+    };
+    const noteBatch = (u)=>{
+        const id = extractBatchId(u);
+        if (id) {
+            affectedBatchIds.add(id);
         }
     };
+    // A batch in a terminal state must never have its units re-queued: the
+    // run is over, and a re-queued job would only re-read the source doc
+    // before being skipped. Re-queueing terminal-batch units (left behind by
+    // a cancel/finish) is what produced the 2026-06-16 zombie-job storm.
+    // `queued`/`running`/`cancelling` are still live and safe to re-queue
+    // into (the worker's cancel gate skips cancelling units on entry).
+    const REQUEUEABLE_BATCH_STATUSES = new Set([
+        'queued',
+        'running',
+        'cancelling'
+    ]);
+    const batchStatusCache = new Map();
+    const getBatchStatus = async (batchId)=>{
+        if (batchStatusCache.has(batchId)) {
+            return batchStatusCache.get(batchId);
+        }
+        let status;
+        try {
+            const batch = await payload.findByID({
+                collection: batchesSlug,
+                id: batchId,
+                overrideAccess: true,
+                depth: 0
+            });
+            status = batch?.status;
+        } catch  {
+            status = undefined;
+        }
+        batchStatusCache.set(batchId, status);
+        return status;
+    };
+    // Whether a unit's batch is still live enough to re-queue into. Tolerant
+    // on a missing/unreadable batch (undefined) — preserve prior behaviour
+    // and let the worker's own guards handle it.
+    const batchIsRequeueable = async (u)=>{
+        const batchId = extractBatchId(u);
+        if (!batchId) {
+            return true;
+        }
+        const status = await getBatchStatus(batchId);
+        return status === undefined || REQUEUEABLE_BATCH_STATUSES.has(status);
+    };
+    // Mark a unit terminal (skipped) so a dead/terminal-batch orphan stops
+    // being rescanned every sweep, without creating a replacement job.
+    const skipTerminalBatchUnit = async (unitId)=>{
+        await payload.update({
+            collection: unitsSlug,
+            id: unitId,
+            data: {
+                status: 'skipped',
+                failureMessage: 'batch_already_terminal',
+                completedAt: new Date(now()).toISOString()
+            },
+            overrideAccess: true
+        });
+    };
     // A unit's worker job is queued exactly once (at enumeration, or by the
     // worker's own deferral re-queue). The worker task has `retries: 0`, and
     // the coordinator never re-runs after enumeration finishes — so a reset
@@ -111,6 +171,12 @@ export async function runJanitorSweep(params) {
     for (const u of stale){
         const attempts = u.attempts ?? 0;
         try {
+            // Don't resurrect a unit whose batch already ended — mark it
+            // terminal instead so it leaves the stale scan without spawning a job.
+            if (!await batchIsRequeueable(u)) {
+                await skipTerminalBatchUnit(u.id);
+                continue;
+            }
             if (attempts < maxAttempts) {
                 await payload.update({
                     collection: unitsSlug,
@@ -205,6 +271,12 @@ export async function runJanitorSweep(params) {
         });
         for (const u of orphans.docs){
             try {
+                // Skip — don't re-queue — units whose batch already ended; mark
+                // them terminal so they drop out of the orphan scan.
+                if (!await batchIsRequeueable(u)) {
+                    await skipTerminalBatchUnit(u.id);
+                    continue;
+                }
                 await requeueUnit(u.id);
                 await payload.update({
                     collection: unitsSlug,

package/package.json

@@ -1,46 +1,98 @@
 {
   "name": "@purposeinplay/payload-ai-translate",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "AI translation plugin for Payload CMS 3 — multi-provider (OpenAI, Anthropic, Gemini, custom), bulk translation, Lexical-aware, with an admin Translation Hub.",
   "license": "MIT",
   "type": "module",
-  "main": "./dist/exports/index.js",
-  "types": "./dist/exports/index.d.ts",
+  "packageManager": "pnpm@9.15.0",
+  "main": "./src/exports/index.ts",
+  "types": "./src/exports/index.ts",
   "exports": {
     ".": {
-      "types": "./dist/exports/index.d.ts",
-      "import": "./dist/exports/index.js",
-      "default": "./dist/exports/index.js"
+      "types": "./src/exports/index.ts",
+      "import": "./src/exports/index.ts",
+      "default": "./src/exports/index.ts"
     },
     "./providers": {
-      "types": "./dist/exports/providers.d.ts",
-      "import": "./dist/exports/providers.js",
-      "default": "./dist/exports/providers.js"
+      "types": "./src/exports/providers.ts",
+      "import": "./src/exports/providers.ts",
+      "default": "./src/exports/providers.ts"
     },
     "./views": {
-      "types": "./dist/exports/views.d.ts",
-      "import": "./dist/exports/views.js",
-      "default": "./dist/exports/views.js"
+      "types": "./src/exports/views.ts",
+      "import": "./src/exports/views.ts",
+      "default": "./src/exports/views.ts"
     },
     "./views-client": {
-      "types": "./dist/exports/views-client.d.ts",
-      "import": "./dist/exports/views-client.js",
-      "default": "./dist/exports/views-client.js"
+      "types": "./src/exports/views-client.ts",
+      "import": "./src/exports/views-client.ts",
+      "default": "./src/exports/views-client.ts"
     },
     "./client": {
-      "types": "./dist/exports/client.d.ts",
-      "import": "./dist/exports/client.js",
-      "default": "./dist/exports/client.js"
+      "types": "./src/exports/client.ts",
+      "import": "./src/exports/client.ts",
+      "default": "./src/exports/client.ts"
     },
     "./components": {
-      "types": "./dist/exports/components.d.ts",
-      "import": "./dist/exports/components.js",
-      "default": "./dist/exports/components.js"
+      "types": "./src/exports/components.ts",
+      "import": "./src/exports/components.ts",
+      "default": "./src/exports/components.ts"
     }
   },
   "files": [
     "dist"
   ],
+  "publishConfig": {
+    "main": "./dist/exports/index.js",
+    "types": "./dist/exports/index.d.ts",
+    "exports": {
+      ".": {
+        "types": "./dist/exports/index.d.ts",
+        "import": "./dist/exports/index.js",
+        "default": "./dist/exports/index.js"
+      },
+      "./providers": {
+        "types": "./dist/exports/providers.d.ts",
+        "import": "./dist/exports/providers.js",
+        "default": "./dist/exports/providers.js"
+      },
+      "./views": {
+        "types": "./dist/exports/views.d.ts",
+        "import": "./dist/exports/views.js",
+        "default": "./dist/exports/views.js"
+      },
+      "./views-client": {
+        "types": "./dist/exports/views-client.d.ts",
+        "import": "./dist/exports/views-client.js",
+        "default": "./dist/exports/views-client.js"
+      },
+      "./client": {
+        "types": "./dist/exports/client.d.ts",
+        "import": "./dist/exports/client.js",
+        "default": "./dist/exports/client.js"
+      },
+      "./components": {
+        "types": "./dist/exports/components.d.ts",
+        "import": "./dist/exports/components.js",
+        "default": "./dist/exports/components.js"
+      }
+    }
+  },
+  "scripts": {
+    "build": "pnpm run build:types && pnpm run build:swc && pnpm run build:fix-esm",
+    "build:types": "tsc -p tsconfig.build.json --emitDeclarationOnly --outDir dist",
+    "build:swc": "swc ./src -d ./dist --config-file ./.swcrc --strip-leading-paths --ignore \"**/__tests__/**\"",
+    "build:fix-esm": "node ./scripts/fix-dist-extensions.mjs dist && node ./scripts/check-dist-esm.mjs dist",
+    "typecheck": "tsc -p tsconfig.json --noEmit",
+    "test": "vitest run",
+    "lint": "biome check .",
+    "format": "biome format --write .",
+    "check": "biome check . && pnpm typecheck && pnpm build && pnpm test",
+    "clean": "rm -rf dist *.tsbuildinfo",
+    "changeset": "changeset",
+    "version-packages": "changeset version",
+    "release": "pnpm build && changeset publish"
+  },
   "peerDependencies": {
     "@ai-sdk/anthropic": "^3.0.0",
     "@ai-sdk/google": "^3.0.0",
@@ -122,20 +174,5 @@
     "openai",
     "anthropic",
     "gemini"
-  ],
-  "scripts": {
-    "build": "pnpm run build:types && pnpm run build:swc && pnpm run build:fix-esm",
-    "build:types": "tsc -p tsconfig.build.json --emitDeclarationOnly --outDir dist",
-    "build:swc": "swc ./src -d ./dist --config-file ./.swcrc --strip-leading-paths --ignore \"**/__tests__/**\"",
-    "build:fix-esm": "node ./scripts/fix-dist-extensions.mjs dist && node ./scripts/check-dist-esm.mjs dist",
-    "typecheck": "tsc -p tsconfig.json --noEmit",
-    "test": "vitest run",
-    "lint": "biome check .",
-    "format": "biome format --write .",
-    "check": "biome check . && pnpm typecheck && pnpm build && pnpm test",
-    "clean": "rm -rf dist *.tsbuildinfo",
-    "changeset": "changeset",
-    "version-packages": "changeset version",
-    "release": "pnpm build && changeset publish"
-  }
-}
+  ]
+}