Upload folder using huggingface_hub

app/core/usage_tracker.py

@@ -621,6 +621,40 @@ class UsageTracker:
             self._mark_idempotent_key_used(idempotency_key)
         return True, None
 
+    def _insert_audit_log(self, record: UsageRecord) -> None:
+        """Insert a standalone usage_log row for a call whose quota was
+        already consumed at request time (see consume_quota_and_log) --
+        used by routes_governance.py's background tasks to record the
+        response body once it's known, under a distinct endpoint suffix
+        (e.g. ".../response"). Best-effort and logged, not raised, for the
+        same reason _record_pg_monthly_count is: this runs after the
+        response has already been sent to the caller, so it must not
+        surface as a request failure -- a background task exception here
+        is otherwise swallowed silently. `record.tier` is None at both real
+        call sites (tier only matters for quota consumption, already done
+        by the earlier consume_quota_and_log call for the same request),
+        but usage_log.tier is NOT NULL, so an absent tier is recorded as
+        "unknown" rather than raising or silently guessing a real tier."""
+        tier_value = record.tier.value if record.tier else "unknown"
+        try:
+            with self._get_conn() as conn:
+                conn.execute(
+                    """INSERT INTO usage_log
+                       (api_key, tier, timestamp, endpoint, request_body, response, error,
+                        processing_ms, idempotency_key)
+                       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+                    (record.api_key, tier_value, record.timestamp, record.endpoint,
+                     json.dumps(record.request_body) if record.request_body else None,
+                     json.dumps(record.response) if record.response else None,
+                     record.error, record.processing_ms, None)
+                )
+                conn.commit()
+        except Exception:
+            logger.error(
+                "Failed to insert audit log for api_key=%s endpoint=%s",
+                record.api_key, record.endpoint, exc_info=True,
+            )
+
     # --------------------------------------------------------------------------
     # Legacy interface (kept for compatibility)
     # --------------------------------------------------------------------------

tests/test_usage_tracker.py

@@ -1,3 +1,4 @@
+import json
 import os
 
 import psycopg2
@@ -149,3 +150,30 @@ def test_increment_usage_sync_succeeds_even_if_postgres_mirror_fails(tracker, mo
     result = tracker.increment_usage_sync(record)
     assert result is True
     assert tracker.get_remaining_quota("mirror-fail-key", Tier.FREE) == 999
+
+
+def test_insert_audit_log_writes_response_row(tracker):
+    """routes_governance.py schedules current_tracker._insert_audit_log as
+    a background task (background_tasks.add_task) to record the response
+    body once it's known, at app/api/routes_governance.py:407 and :738 --
+    always with tier=None, since quota was already consumed by an earlier
+    consume_quota_and_log call for the same request. The real UsageTracker
+    had no such method (only tests/conftest.py's MockTracker did), so every
+    real call raised AttributeError inside the background task (arf-api-002)."""
+    record = UsageRecord(
+        api_key="audit-log-key",
+        tier=None,
+        timestamp=time.time(),
+        endpoint="/api/v1/intents/evaluate/response",
+        request_body=None,
+        response={"recommended_action": "approve"},
+        processing_ms=12.5,
+    )
+
+    tracker._insert_audit_log(record)
+
+    logs = tracker.get_audit_logs("audit-log-key", limit=10)
+    assert len(logs) == 1
+    assert logs[0]["endpoint"] == "/api/v1/intents/evaluate/response"
+    assert logs[0]["tier"] == "unknown"
+    assert json.loads(logs[0]["response"]) == {"recommended_action": "approve"}