fix(mail): rate-limit direct and fallback sends

src/govoplan_mail/backend/sending/rate_limit.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import time
+import threading
 from dataclasses import dataclass
 
 from redis import Redis
@@ -17,6 +18,10 @@ class RateLimitDecision:
     waited_seconds: float
 
 
+_local_rate_limit_lock = threading.Lock()
+_local_next_allowed: dict[str, float] = {}
+
+
 def _redis_client() -> Redis:
     return Redis.from_url(settings.redis_url, decode_responses=True)
 
@@ -30,15 +35,18 @@ def wait_for_rate_limit(*, key: str, messages_per_minute: int, enabled: bool = T
 
     The implementation stores the next allowed send timestamp per key. A Redis
     lock keeps multiple Celery processes from reading/updating the timestamp at
-    the same time. Direct local development runs do not have a broker, so Redis
-    is only used when Celery/worker mode is explicitly enabled. If Redis is
-    unavailable, it falls back to no distributed wait.
+    the same time. Direct local development runs do not have a broker, so they
+    use a process-local limiter. If Redis is unavailable in worker mode, the
+    process-local fallback still protects single-process sends.
     """
 
     messages_per_minute = max(1, int(messages_per_minute or 1))
     gap = 60.0 / messages_per_minute
-    if not enabled or not _distributed_rate_limit_enabled():
+    if not enabled:
         return RateLimitDecision(key=key, messages_per_minute=messages_per_minute, gap_seconds=gap, waited_seconds=0.0)
+    if not _distributed_rate_limit_enabled():
+        waited = _wait_for_local_rate_limit(key, gap)
+        return RateLimitDecision(key=key, messages_per_minute=messages_per_minute, gap_seconds=gap, waited_seconds=waited)
 
     redis_key = f"govoplan:ratelimit:{key}:next_allowed"
     lock_key = f"govoplan:ratelimit:{key}:lock"
@@ -56,7 +64,17 @@ def wait_for_rate_limit(*, key: str, messages_per_minute: int, enabled: bool = T
                 now = time.time()
             client.set(redis_key, now + gap, ex=max(60, int(gap * 10)))
     except (RedisError, TimeoutError, ValueError):
-        # Development fallback: do not fail sending because Redis is absent.
-        waited = 0.0
+        waited = _wait_for_local_rate_limit(key, gap)
 
     return RateLimitDecision(key=key, messages_per_minute=messages_per_minute, gap_seconds=gap, waited_seconds=waited)
+
+
+def _wait_for_local_rate_limit(key: str, gap: float) -> float:
+    with _local_rate_limit_lock:
+        now = time.time()
+        next_allowed = _local_next_allowed.get(key, now)
+        waited = max(0.0, next_allowed - now)
+        _local_next_allowed[key] = max(now, next_allowed) + gap
+    if waited > 0:
+        time.sleep(waited)
+    return waited