@runuai/host 0.2.3 → 0.2.4

package/lib/github-tokens.ts

@@ -238,15 +238,19 @@ async function runRefresh(taskId: string, userId: string): Promise<void> {
   } catch (err) {
     const reason = err instanceof Error ? err.message : String(err);
     // A revoked / expired refresh token can't recover — drop it to force a
-    // clean re-grant through the OAuth flow.
+    // clean re-grant through the OAuth flow. Anything else is likely transient
+    // (network / cloud blip) — self-heal with a bounded retry.
     if (/\b40[13]\b|revoked|invalid_grant|bad_refresh/i.test(reason)) {
       deleteToken(userId);
+    } else {
+      scheduleGithubRetry(taskId, userId, 0);
     }
     authExpiredHandler?.(taskId, userId, reason);
   }
 }
 
 export function clearRefresh(taskId: string): void {
+  clearGithubRetry(taskId);
   const t = timers.get(taskId);
   if (t) {
     clearTimeout(t);
@@ -254,6 +258,50 @@ export function clearRefresh(taskId: string): void {
   }
 }
 
+// --- self-heal: retry a failed gh setup with bounded backoff ----------------
+// A gh setup/refresh can fail transiently — e.g. the cloud token-exchange fetch
+// times out while the host is briefly unreachable (as happened when the bridge
+// was flapping). Without a retry the task is stranded on a stale/invalid token
+// until someone runs /retry-gh, because ensureSessions won't re-inject once a
+// channel's sessions are started. Retry with bounded backoff; cleared on
+// success (clearGithubRetry) and on task teardown (via clearRefresh).
+const RETRY_BACKOFF_MS = [5_000, 20_000, 60_000, 180_000, 300_000];
+const retryTimers = new Map<string, ReturnType<typeof setTimeout>>();
+
+function clearGithubRetry(taskId: string): void {
+  const t = retryTimers.get(taskId);
+  if (t) {
+    clearTimeout(t);
+    retryTimers.delete(taskId);
+  }
+}
+
+function scheduleGithubRetry(
+  taskId: string,
+  userId: string,
+  attempt: number,
+): void {
+  clearGithubRetry(taskId);
+  if (attempt >= RETRY_BACKOFF_MS.length) {
+    console.warn(
+      `[github] task ${taskId}: gh setup still failing after ${attempt} attempts — giving up until /retry-gh`,
+    );
+    return;
+  }
+  const delay = RETRY_BACKOFF_MS[attempt] ?? 300_000;
+  const timer = setTimeout(() => {
+    retryTimers.delete(taskId);
+    void setupTaskGithub(taskId, userId, {}, attempt + 1);
+  }, delay);
+  timer.unref?.();
+  retryTimers.set(taskId, timer);
+}
+
+/** Test/diagnostic: is a gh-setup retry currently armed for this task? */
+export function hasPendingGithubRetry(taskId: string): boolean {
+  return retryTimers.has(taskId);
+}
+
 /**
  * Injectable seams for {@link setupTaskGithub} (mocked in tests so the decision
  * logic needs neither the host DB nor docker). All default to the real impls.
@@ -283,6 +331,7 @@ export async function setupTaskGithub(
   taskId: string,
   userId: string,
   deps: SetupTaskDeps = {},
+  attempt = 0,
 ): Promise<boolean> {
   const _hasToken = deps.hasToken ?? hasToken;
   const _request = deps.requestAccessToken ?? requestAccessToken;
@@ -297,6 +346,7 @@ export async function setupTaskGithub(
       if (tok) {
         await _inject(taskId, tok.accessToken);
         _schedule(taskId, userId, tok.expiresAt);
+        clearGithubRetry(taskId);
         console.log(`[github] task ${taskId}: injected user access token`);
         return true;
       }
@@ -304,6 +354,7 @@ export async function setupTaskGithub(
     const pat = process.env.UAI_GH_PAT_FALLBACK;
     if (pat) {
       await _inject(taskId, pat);
+      clearGithubRetry(taskId);
       console.log(`[github] task ${taskId}: using PAT fallback (no refresh)`);
       return true;
     }
@@ -313,14 +364,19 @@ export async function setupTaskGithub(
     const reason = err instanceof Error ? err.message : String(err);
     console.warn(`[github] task ${taskId}: gh setup failed: ${reason}`);
     authExpiredHandler?.(taskId, userId, reason);
+    // Self-heal: a transient exchange/inject failure shouldn't strand the task
+    // on a stale token until a manual /retry-gh. Retry with bounded backoff.
+    scheduleGithubRetry(taskId, userId, attempt);
     return false;
   } finally {
     setupInFlight.delete(taskId);
   }
 }
 
-/** Test/maintenance hook: drop all scheduled refresh timers. */
+/** Test/maintenance hook: drop all scheduled refresh + retry timers. */
 export function clearAllRefreshTimers(): void {
   for (const t of timers.values()) clearTimeout(t);
   timers.clear();
+  for (const t of retryTimers.values()) clearTimeout(t);
+  retryTimers.clear();
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runuai/host",
-  "version": "0.2.3",
+  "version": "0.2.4",
   "description": "Uai host — runs ephemeral AI coding tasks in Docker on a machine you control.",
   "license": "MIT",
   "author": "Diogo Perillo <diogo.perillo@gmail.com>",