fix(i18n): forget a batch only when it is gone, not on an auth failure

Author: berry-13

lib/i18n/batch.test.ts

@@ -328,8 +328,12 @@ describe('prefetchTranslations', () => {
     expect(posts).toBe(0)
     expect(polled).toEqual(['gone'])
     expect(pending.size).toBe(0)
-    // The other model's batch is kept for when that model is used again.
-    expect(Object.keys(JSON.parse(await readFile(ledgerPath, 'utf8')))).toEqual(['other'])
+    // The other model's batch is kept for when that model is used again, and the
+    // forbidden one for a run whose key can read it.
+    expect(Object.keys(JSON.parse(await readFile(ledgerPath, 'utf8'))).sort()).toEqual([
+      'gone',
+      'other',
+    ])
   })
 
   it('does not throw when a submit fails', async () => {

lib/i18n/batch.ts

@@ -244,11 +244,12 @@ export async function prefetchTranslations(
       let batch: BatchObject
       try {
         const res = await doFetch(`${BATCHES_URL}/${id}`, { headers, signal: timeout() })
-        // Deleted, never ours, or no longer readable with this key (any 4xx but a
-        // timeout or rate limit): nothing will ever arrive, so stop waiting on it.
+        // Any 4xx but a timeout or rate limit will not change by asking again, so stop
+        // waiting and let the requests run live. Only a 404 means the batch is gone;
+        // an auth failure keeps it in the ledger for a run with a working key.
         if (res.status >= 400 && res.status < 500 && res.status !== 408 && res.status !== 429) {
           batches.delete(id)
-          known.delete(id)
+          if (res.status === 404) known.delete(id)
           opts.log(`batch ${id} unreadable (HTTP ${res.status}); its requests will run live`)
           continue
         }