ticket #103 attempt 1

devplacepy/services/news/service.py

@@ -18,7 +18,9 @@ from devplacepy.services.openai_gateway.usage import (
     usage_metric_cards,
 )
 from devplacepy.utils import generate_uid, make_combined_slug
+from devplacepy.utils.notifications import create_notification
 from devplacepy.services.audit import record as audit
+from devplacepy.services.openai_gateway.reliability import retry_send
 from devplacepy.services.seo_meta import schedule_seo_meta
 
 from . import _get_ai_key
@@ -151,6 +153,32 @@ class NewsService(BaseService):
             ),
             group="AI formatting",
         ),
+        ConfigField(
+            "news_fetch_retries",
+            "News API fetch retries",
+            type="int",
+            default=3,
+            minimum=0,
+            maximum=10,
+            help=(
+                "How many times to retry the upstream news API on failure. "
+                "Each retry waits longer (linear backoff). Set 0 for no retries."
+            ),
+            group="Reliability",
+        ),
+        ConfigField(
+            "news_fetch_retry_backoff_ms",
+            "News API retry backoff (ms)",
+            type="int",
+            default=5000,
+            minimum=1000,
+            maximum=60000,
+            help=(
+                "Base backoff in milliseconds between retries. The actual "
+                "delay is backoff * attempt number."
+            ),
+            group="Reliability",
+        ),
     ]
 
     def __init__(self):
@@ -165,13 +193,38 @@ class NewsService(BaseService):
         format_enabled = config["news_format_enabled"]
 
         self.log(f"Fetching news from {api_url}")
+        max_retries = config["news_fetch_retries"]
+        backoff_ms = config["news_fetch_retry_backoff_ms"]
         async with stealth.stealth_async_client(timeout=30.0) as client:
             try:
-                resp = await client.get(api_url)
-                resp.raise_for_status()
+                resp, exc, attempts = await retry_send(
+                    do_call=lambda: client.get(api_url),
+                    max_retries=max_retries,
+                    backoff_ms=backoff_ms,
+                    log=lambda msg: self.log(msg),
+                )
+                if exc is not None:
+                    raise exc
+                if resp is None or resp.status_code >= 400:
+                    raise httpx.HTTPError(f"status {resp.status_code if resp else 0}")
                 data = resp.json()
             except Exception as e:
-                self.log(f"Failed to fetch news API: {e}")
+                self.log(f"Failed to fetch news API after {attempts} attempts: {e}")
+                audit.record_system(
+                    "news.service.fetch_failed",
+                    actor_kind="service",
+                    actor_uid="news",
+                    summary=f"News API unreachable after {attempts} attempts",
+                    metadata={"api_url": api_url, "attempts": attempts, "error": str(e)},
+                    result="failure",
+                )
+                for admin_row in get_table("users").find(role="Admin"):
+                    create_notification(
+                        admin_row["uid"],
+                        "system",
+                        f"News API unreachable after {attempts} attempts",
+                        related_uid="",
+                    )
                 return
 
             articles = data.get("articles", [])