Fix silent compose failures: poll actual job status in the compose modal instead of a fire-and-forget dispatch toast; add Groq 429 retry-with-backoff using its own rate-limit-reset header (12k tokens/min tier gets exhausted by gmail_triage alone)

deploy/reactor.py

@@ -173,17 +173,40 @@ async def _claude_call(messages: list, system: str = "") -> str:
                 raise RuntimeError(f"Claude API error {resp.status}: {data.get('error',{}).get('message','')}")
             return data["content"][0]["text"]
 
+def _parse_groq_reset(val: str) -> float:
+    """Parse Groq's Go-style duration strings ('229ms', '2.5s', '2m52.8s') into seconds."""
+    import re as _re
+    if not val:
+        return 0.0
+    total = 0.0
+    for num, unit in _re.findall(r'([\d.]+)(ms|s|m|h)', val):
+        n = float(num)
+        total += n/1000 if unit == 'ms' else n*60 if unit == 'm' else n*3600 if unit == 'h' else n
+    return total
+
 async def _groq_call(messages: list, system: str = "") -> str:
+    # Added 2026-07-07: retry once on 429 (rate limit) using Groq's own reset-time
+    # header, capped short — this account's tier has a low 12k-tokens/min ceiling that
+    # gmail_triage alone can exhaust, so transient contention here is common and often
+    # clears in well under a second. Capped so a genuinely long reset just falls
+    # through to Ollama instead of blocking the whole compose flow.
     all_msgs = ([{"role": "system", "content": system}] if system else []) + messages
     payload = {"model": GROQ_MODEL, "messages": all_msgs, "max_tokens": 4096}
     headers = {"Authorization": f"Bearer {GROQ_API_KEY}", "Content-Type": "application/json"}
-    async with aiohttp.ClientSession() as session:
-        async with session.post("https://api.groq.com/openai/v1/chat/completions", json=payload,
-                                headers=headers, timeout=aiohttp.ClientTimeout(total=45)) as resp:
-            data = await resp.json()
-            if resp.status != 200:
-                raise RuntimeError(f"Groq error {resp.status}")
-            return data["choices"][0]["message"]["content"]
+    for attempt in range(2):
+        async with aiohttp.ClientSession() as session:
+            async with session.post("https://api.groq.com/openai/v1/chat/completions", json=payload,
+                                    headers=headers, timeout=aiohttp.ClientTimeout(total=45)) as resp:
+                if resp.status == 429 and attempt == 0:
+                    wait = min(_parse_groq_reset(resp.headers.get("x-ratelimit-reset-tokens", "")) or
+                               _parse_groq_reset(resp.headers.get("x-ratelimit-reset-requests", "")) or 2.0, 8.0)
+                    log.info(f"[LLM] Groq 429 — retrying in {wait:.1f}s")
+                    await asyncio.sleep(wait)
+                    continue
+                data = await resp.json()
+                if resp.status != 200:
+                    raise RuntimeError(f"Groq error {resp.status}")
+                return data["choices"][0]["message"]["content"]
 
 async def _ollama_call(messages: list, system: str = "") -> str:
     prompt = (system + "\n\n" if system else "") + "\n".join(f"{m['role'].upper()}: {m['content']}" for m in messages)