fix: add Primary-Wait Prometheus counters + conservative defaults — BIZ-60 review

P0 changes per 4-reviewer consensus (严维序/陆怀瑾/沈路明/梁思筑):

1. Prometheus metrics counters (proxy.py + server.py):
   - sidecar_primary_wait_enter_total: requests entering Primary-Wait
   - sidecar_primary_wait_recovery_total: successful primary recoveries
   - sidecar_primary_wait_exhausted_total: wait exhausted → emergency

2. Conservative default (config.py):
   - primary_wait_max_retries: 6 → 3 (15s total wait, safe start)
   - Observe recovery rate before increasing to 6

Counters form complete funnel: enter - recovery = exhausted,
enabling Grafana monitoring and ROI validation per COO/PM/Ops.

config.py

@@ -75,7 +75,7 @@ class Config:
 
     # Primary-Wait: when all primary backends are cooling, wait before fallback
     primary_wait_ms: int = 5000
-    primary_wait_max_retries: int = 6
+    primary_wait_max_retries: int = 3
 
     # Request timeout
     default_request_timeout_seconds: int = 120

proxy.py

@@ -21,11 +21,28 @@ from storage.usage_store import record_usage
 # Emergency activation counter (read by metrics endpoint)
 _emergency_count: int = 0
 
+# Primary-Wait metrics counters (read by metrics endpoint)
+_primary_wait_enter_count: int = 0
+_primary_wait_recovery_count: int = 0
+_primary_wait_exhausted_count: int = 0
+
 
 def get_emergency_count() -> int:
     return _emergency_count
 
 
+def get_primary_wait_enter_count() -> int:
+    return _primary_wait_enter_count
+
+
+def get_primary_wait_recovery_count() -> int:
+    return _primary_wait_recovery_count
+
+
+def get_primary_wait_exhausted_count() -> int:
+    return _primary_wait_exhausted_count
+
+
 logger: structlog.stdlib.BoundLogger = structlog.get_logger("sidecar_v2.proxy")
 
 
@@ -309,6 +326,9 @@ async def handle_proxy_request(
             continue
 
     # --- Primary-Wait: wait for primary pool recovery before fallback/emergency ---
+    global _primary_wait_enter_count, _primary_wait_recovery_count, _primary_wait_exhausted_count
+    _primary_wait_enter_count += 1
+
     pwl = logger.bind(phase="primary_wait")
     for pw_attempt in range(config.primary_wait_max_retries):
         await asyncio.sleep(config.primary_wait_ms / 1000.0)
@@ -356,6 +376,7 @@ async def handle_proxy_request(
                 continue
 
             # Primary recovered — success
+            _primary_wait_recovery_count += 1
             resp_json: dict[str, Any] = {}
             try:
                 if not is_stream and resp.content:
@@ -417,6 +438,9 @@ async def handle_proxy_request(
             )
         continue
 
+    # Primary-Wait all retries exhausted
+    _primary_wait_exhausted_count += 1
+
     # All pools exhausted (including primary-wait retries) — emergency rate-limited passthrough
     emergency_rpm = int(config.default_rpm_limit * config.emergency_rpm_fraction)
     if emergency_rpm < 1:

server.py

@@ -20,7 +20,13 @@ from crypto import init_crypto, is_initialized
 from pool_manager import PoolManager
 from rate_limiter import PerBackendRateLimiter
 from router import Router
-from proxy import handle_proxy_request, get_emergency_count
+from proxy import (
+    handle_proxy_request,
+    get_emergency_count,
+    get_primary_wait_enter_count,
+    get_primary_wait_recovery_count,
+    get_primary_wait_exhausted_count,
+)
 
 from storage.db import init_db, create_tables, run_integrity_check, get_connection, _DB_PATH
 from storage.backend_store import (
@@ -383,6 +389,11 @@ async def metrics() -> Response:
     # Emergency count (from proxy module)
     lines.append(f"sidecar_emergency_count {get_emergency_count()}")
 
+    # Primary-Wait metrics
+    lines.append(f"sidecar_primary_wait_enter_total {get_primary_wait_enter_count()}")
+    lines.append(f"sidecar_primary_wait_recovery_total {get_primary_wait_recovery_count()}")
+    lines.append(f"sidecar_primary_wait_exhausted_total {get_primary_wait_exhausted_count()}")
+
     # DB sizes
     from storage.db import get_db_sizes
     sizes = get_db_sizes()