fix: retry DNS resolution in validate_tokens() with backoff (FB_DNS_RETRY_ATTEMPTS)

app.py

@@ -154,13 +154,29 @@ if __name__ == "__main__":
             ok = False
         else:
             # quick DNS resolution check to avoid noisy stack traces when DNS fails
-            try:
-                socket.getaddrinfo("graph.facebook.com", 443)
-            except Exception as e:
-                logger.error("DNS resolution failed for graph.facebook.com: %s", e)
+            # Retry DNS resolution a few times with exponential backoff to handle transient name-resolution blips
+            dns_attempts = int(os.getenv("FB_DNS_RETRY_ATTEMPTS", "3"))
+            dns_backoff_base = int(os.getenv("FB_DNS_BACKOFF_BASE", "1"))
+            dns_ok = False
+            last_dns_exc = None
+            for attempt in range(1, dns_attempts + 1):
+                try:
+                    socket.getaddrinfo("graph.facebook.com", 443)
+                    dns_ok = True
+                    break
+                except Exception as e:
+                    last_dns_exc = e
+                    logger.warning("DNS resolution attempt %s/%s failed: %s", attempt, dns_attempts, e)
+                    if attempt < dns_attempts:
+                        try:
+                            time.sleep(dns_backoff_base * (2 ** (attempt - 1)))
+                        except Exception:
+                            pass
+            if not dns_ok:
+                logger.error("DNS resolution failed for graph.facebook.com after %s attempts: %s", dns_attempts, last_dns_exc)
                 try:
                     with open("log.txt", "a", encoding="utf-8") as lf:
-                        lf.write(f"[{__import__('time').strftime('%Y-%m-%d %H:%M:%S')}] FB_DNS_RESOLUTION_FAILED {e}\n")
+                        lf.write(f"[{__import__('time').strftime('%Y-%m-%d %H:%M:%S')}] FB_DNS_RESOLUTION_FAILED attempts={dns_attempts} error={last_dns_exc}\n")
                 except Exception:
                     logger.exception("Failed to write DNS failure to log.txt")
                 ok = False