Spaces:
Build error
Build error
Commit Β·
3b45581
1
Parent(s): 30bcdd5
Upload folder using huggingface_hub
Browse files- app/api/routes_admin.py +21 -20
- app/api/routes_governance.py +40 -0
- app/api/routes_incidents.py +6 -5
- app/api/routes_payments.py +5 -4
- app/api/routes_users.py +4 -3
- app/core/usage_tracker.py +11 -0
- app/main.py +31 -33
- tests/test_tracker_binding.py +52 -0
app/api/routes_admin.py
CHANGED
|
@@ -11,7 +11,8 @@ import secrets
|
|
| 11 |
import uuid
|
| 12 |
from sqlalchemy.orm import Session
|
| 13 |
from app.api.deps import get_db
|
| 14 |
-
from app.core
|
|
|
|
| 15 |
from app.database.models_intents import TenantDB
|
| 16 |
|
| 17 |
router = APIRouter(prefix="/admin", tags=["admin"])
|
|
@@ -67,7 +68,7 @@ async def create_api_key(req: CreateKeyRequest, db: Session = Depends(get_db)):
|
|
| 67 |
db.commit()
|
| 68 |
|
| 69 |
new_key = f"sk_live_{uuid.uuid4().hex[:24]}"
|
| 70 |
-
tracker.get_or_create_api_key(new_key, tenant_id=tenant_id, tier=tier_enum)
|
| 71 |
return {"api_key": new_key, "tenant_id": tenant_id, "tier": req.tier}
|
| 72 |
|
| 73 |
|
|
@@ -82,8 +83,8 @@ async def list_api_keys(limit: int = 100, offset: int = 0):
|
|
| 82 |
longer has -- query `/admin/keys/{api_key}/audit` with the real key for
|
| 83 |
per-key usage/audit history instead.
|
| 84 |
"""
|
| 85 |
-
with tracker._get_pg_conn() as conn:
|
| 86 |
-
rows = tracker._pg_execute(
|
| 87 |
conn,
|
| 88 |
"SELECT lookup_hash, tier, created_at, last_used_at, is_active FROM api_keys "
|
| 89 |
"ORDER BY created_at DESC LIMIT %s OFFSET %s",
|
|
@@ -111,13 +112,13 @@ async def update_key_tier(
|
|
| 111 |
if req.tier not in [t.value for t in Tier]:
|
| 112 |
raise HTTPException(
|
| 113 |
status_code=400, detail=f"Invalid tier. Must be one of {[t.value for t in Tier]}")
|
| 114 |
-
with tracker._get_pg_conn() as conn:
|
| 115 |
-
row = tracker._pg_execute(
|
| 116 |
conn, "SELECT lookup_hash FROM api_keys WHERE lookup_hash = %s", (key_id,)).fetchone()
|
| 117 |
if not row:
|
| 118 |
conn.rollback()
|
| 119 |
raise HTTPException(status_code=404, detail="API key not found")
|
| 120 |
-
tracker._pg_execute(
|
| 121 |
conn, "UPDATE api_keys SET tier = %s WHERE lookup_hash = %s", (req.tier, key_id))
|
| 122 |
conn.commit()
|
| 123 |
return {"message": f"Tier updated to {req.tier}"}
|
|
@@ -133,8 +134,8 @@ async def rotate_api_key(
|
|
| 133 |
before this endpoint existed) risks losing track of the tenant_id
|
| 134 |
partway through, or leaving the old key active if a later step fails.
|
| 135 |
The new plaintext key is returned exactly once, like create_api_key's."""
|
| 136 |
-
with tracker._get_pg_conn() as conn:
|
| 137 |
-
old_row = tracker._pg_execute(
|
| 138 |
conn, "SELECT tenant_id, tier FROM api_keys WHERE lookup_hash = %s", (key_id,)
|
| 139 |
).fetchone()
|
| 140 |
if not old_row:
|
|
@@ -143,15 +144,15 @@ async def rotate_api_key(
|
|
| 143 |
|
| 144 |
new_key = f"sk_live_{uuid.uuid4().hex[:24]}"
|
| 145 |
salt = secrets.token_hex(16)
|
| 146 |
-
tracker._pg_execute(
|
| 147 |
conn, "UPDATE api_keys SET is_active = false WHERE lookup_hash = %s", (key_id,))
|
| 148 |
-
tracker._pg_execute(
|
| 149 |
conn,
|
| 150 |
"INSERT INTO api_keys "
|
| 151 |
"(tenant_id, tier, created_at, is_active, salt, key_hash, lookup_hash) "
|
| 152 |
"VALUES (%s, %s, %s, %s, %s, %s, %s)",
|
| 153 |
(old_row["tenant_id"], old_row["tier"], datetime.utcnow(), True,
|
| 154 |
-
salt, tracker._salted_hash(new_key, salt), tracker._lookup_hash(new_key)),
|
| 155 |
)
|
| 156 |
conn.commit()
|
| 157 |
|
|
@@ -166,13 +167,13 @@ async def rotate_api_key(
|
|
| 166 |
@router.delete("/keys/{key_id}", dependencies=[Depends(verify_admin)])
|
| 167 |
async def deactivate_api_key(
|
| 168 |
key_id: str = Path(..., description="The key_id from GET /admin/keys (not the raw API key)")):
|
| 169 |
-
with tracker._get_pg_conn() as conn:
|
| 170 |
-
row = tracker._pg_execute(
|
| 171 |
conn, "SELECT lookup_hash FROM api_keys WHERE lookup_hash = %s", (key_id,)).fetchone()
|
| 172 |
if not row:
|
| 173 |
conn.rollback()
|
| 174 |
raise HTTPException(status_code=404, detail="API key not found")
|
| 175 |
-
tracker._pg_execute(
|
| 176 |
conn, "UPDATE api_keys SET is_active = false WHERE lookup_hash = %s", (key_id,))
|
| 177 |
conn.commit()
|
| 178 |
return {"message": "API key deactivated"}
|
|
@@ -187,23 +188,23 @@ async def get_audit_logs(
|
|
| 187 |
):
|
| 188 |
start = datetime.fromisoformat(start_date) if start_date else None
|
| 189 |
end = datetime.fromisoformat(end_date) if end_date else None
|
| 190 |
-
logs = tracker.get_audit_logs(api_key, start, end, limit)
|
| 191 |
return {"api_key": api_key, "logs": logs}
|
| 192 |
|
| 193 |
|
| 194 |
@router.get("/stats", dependencies=[Depends(verify_admin)])
|
| 195 |
async def get_global_stats():
|
| 196 |
-
with tracker._get_pg_conn() as pg_conn:
|
| 197 |
-
total_keys = tracker._pg_execute(
|
| 198 |
pg_conn, "SELECT COUNT(*) FROM api_keys WHERE is_active = true").fetchone()["count"]
|
| 199 |
pg_conn.commit()
|
| 200 |
-
with tracker._get_conn() as conn:
|
| 201 |
total_requests = conn.execute(
|
| 202 |
"SELECT COUNT(*) FROM usage_log").fetchone()[0]
|
| 203 |
by_tier = conn.execute(
|
| 204 |
"SELECT tier, COUNT(*) as count FROM usage_log GROUP BY tier"
|
| 205 |
).fetchall()
|
| 206 |
-
month = tracker._get_month_key()
|
| 207 |
current_month_requests = conn.execute(
|
| 208 |
"SELECT SUM(count) FROM monthly_counts WHERE year_month = ?", (month,)
|
| 209 |
).fetchone()[0] or 0
|
|
|
|
| 11 |
import uuid
|
| 12 |
from sqlalchemy.orm import Session
|
| 13 |
from app.api.deps import get_db
|
| 14 |
+
from app.core import usage_tracker
|
| 15 |
+
from app.core.usage_tracker import Tier
|
| 16 |
from app.database.models_intents import TenantDB
|
| 17 |
|
| 18 |
router = APIRouter(prefix="/admin", tags=["admin"])
|
|
|
|
| 68 |
db.commit()
|
| 69 |
|
| 70 |
new_key = f"sk_live_{uuid.uuid4().hex[:24]}"
|
| 71 |
+
usage_tracker.tracker.get_or_create_api_key(new_key, tenant_id=tenant_id, tier=tier_enum)
|
| 72 |
return {"api_key": new_key, "tenant_id": tenant_id, "tier": req.tier}
|
| 73 |
|
| 74 |
|
|
|
|
| 83 |
longer has -- query `/admin/keys/{api_key}/audit` with the real key for
|
| 84 |
per-key usage/audit history instead.
|
| 85 |
"""
|
| 86 |
+
with usage_tracker.tracker._get_pg_conn() as conn:
|
| 87 |
+
rows = usage_tracker.tracker._pg_execute(
|
| 88 |
conn,
|
| 89 |
"SELECT lookup_hash, tier, created_at, last_used_at, is_active FROM api_keys "
|
| 90 |
"ORDER BY created_at DESC LIMIT %s OFFSET %s",
|
|
|
|
| 112 |
if req.tier not in [t.value for t in Tier]:
|
| 113 |
raise HTTPException(
|
| 114 |
status_code=400, detail=f"Invalid tier. Must be one of {[t.value for t in Tier]}")
|
| 115 |
+
with usage_tracker.tracker._get_pg_conn() as conn:
|
| 116 |
+
row = usage_tracker.tracker._pg_execute(
|
| 117 |
conn, "SELECT lookup_hash FROM api_keys WHERE lookup_hash = %s", (key_id,)).fetchone()
|
| 118 |
if not row:
|
| 119 |
conn.rollback()
|
| 120 |
raise HTTPException(status_code=404, detail="API key not found")
|
| 121 |
+
usage_tracker.tracker._pg_execute(
|
| 122 |
conn, "UPDATE api_keys SET tier = %s WHERE lookup_hash = %s", (req.tier, key_id))
|
| 123 |
conn.commit()
|
| 124 |
return {"message": f"Tier updated to {req.tier}"}
|
|
|
|
| 134 |
before this endpoint existed) risks losing track of the tenant_id
|
| 135 |
partway through, or leaving the old key active if a later step fails.
|
| 136 |
The new plaintext key is returned exactly once, like create_api_key's."""
|
| 137 |
+
with usage_tracker.tracker._get_pg_conn() as conn:
|
| 138 |
+
old_row = usage_tracker.tracker._pg_execute(
|
| 139 |
conn, "SELECT tenant_id, tier FROM api_keys WHERE lookup_hash = %s", (key_id,)
|
| 140 |
).fetchone()
|
| 141 |
if not old_row:
|
|
|
|
| 144 |
|
| 145 |
new_key = f"sk_live_{uuid.uuid4().hex[:24]}"
|
| 146 |
salt = secrets.token_hex(16)
|
| 147 |
+
usage_tracker.tracker._pg_execute(
|
| 148 |
conn, "UPDATE api_keys SET is_active = false WHERE lookup_hash = %s", (key_id,))
|
| 149 |
+
usage_tracker.tracker._pg_execute(
|
| 150 |
conn,
|
| 151 |
"INSERT INTO api_keys "
|
| 152 |
"(tenant_id, tier, created_at, is_active, salt, key_hash, lookup_hash) "
|
| 153 |
"VALUES (%s, %s, %s, %s, %s, %s, %s)",
|
| 154 |
(old_row["tenant_id"], old_row["tier"], datetime.utcnow(), True,
|
| 155 |
+
salt, usage_tracker.tracker._salted_hash(new_key, salt), usage_tracker.tracker._lookup_hash(new_key)),
|
| 156 |
)
|
| 157 |
conn.commit()
|
| 158 |
|
|
|
|
| 167 |
@router.delete("/keys/{key_id}", dependencies=[Depends(verify_admin)])
|
| 168 |
async def deactivate_api_key(
|
| 169 |
key_id: str = Path(..., description="The key_id from GET /admin/keys (not the raw API key)")):
|
| 170 |
+
with usage_tracker.tracker._get_pg_conn() as conn:
|
| 171 |
+
row = usage_tracker.tracker._pg_execute(
|
| 172 |
conn, "SELECT lookup_hash FROM api_keys WHERE lookup_hash = %s", (key_id,)).fetchone()
|
| 173 |
if not row:
|
| 174 |
conn.rollback()
|
| 175 |
raise HTTPException(status_code=404, detail="API key not found")
|
| 176 |
+
usage_tracker.tracker._pg_execute(
|
| 177 |
conn, "UPDATE api_keys SET is_active = false WHERE lookup_hash = %s", (key_id,))
|
| 178 |
conn.commit()
|
| 179 |
return {"message": "API key deactivated"}
|
|
|
|
| 188 |
):
|
| 189 |
start = datetime.fromisoformat(start_date) if start_date else None
|
| 190 |
end = datetime.fromisoformat(end_date) if end_date else None
|
| 191 |
+
logs = usage_tracker.tracker.get_audit_logs(api_key, start, end, limit)
|
| 192 |
return {"api_key": api_key, "logs": logs}
|
| 193 |
|
| 194 |
|
| 195 |
@router.get("/stats", dependencies=[Depends(verify_admin)])
|
| 196 |
async def get_global_stats():
|
| 197 |
+
with usage_tracker.tracker._get_pg_conn() as pg_conn:
|
| 198 |
+
total_keys = usage_tracker.tracker._pg_execute(
|
| 199 |
pg_conn, "SELECT COUNT(*) FROM api_keys WHERE is_active = true").fetchone()["count"]
|
| 200 |
pg_conn.commit()
|
| 201 |
+
with usage_tracker.tracker._get_conn() as conn:
|
| 202 |
total_requests = conn.execute(
|
| 203 |
"SELECT COUNT(*) FROM usage_log").fetchone()[0]
|
| 204 |
by_tier = conn.execute(
|
| 205 |
"SELECT tier, COUNT(*) as count FROM usage_log GROUP BY tier"
|
| 206 |
).fetchall()
|
| 207 |
+
month = usage_tracker.tracker._get_month_key()
|
| 208 |
current_month_requests = conn.execute(
|
| 209 |
"SELECT SUM(count) FROM monthly_counts WHERE year_month = ?", (month,)
|
| 210 |
).fetchone()[0] or 0
|
app/api/routes_governance.py
CHANGED
|
@@ -81,6 +81,7 @@ except ImportError:
|
|
| 81 |
try:
|
| 82 |
from arf_enterprise.executor import EnterpriseExecutor
|
| 83 |
from arf_enterprise.actuators.fake import FakeCloudActuator
|
|
|
|
| 84 |
from arf_enterprise.store import ApprovalStore, PostgresStore
|
| 85 |
from arf_enterprise.exceptions import (
|
| 86 |
ExecutionError as EnterpriseExecutionError,
|
|
@@ -92,12 +93,31 @@ except ImportError:
|
|
| 92 |
ENTERPRISE_EXECUTOR_AVAILABLE = False
|
| 93 |
EnterpriseExecutor = None
|
| 94 |
FakeCloudActuator = None
|
|
|
|
| 95 |
ApprovalStore = None
|
| 96 |
PostgresStore = None
|
| 97 |
EnterpriseExecutionError = None
|
| 98 |
PendingApprovalError = None
|
| 99 |
EnterpriseSafetyError = None
|
| 100 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
ARF_ENABLE_EXECUTION = os.getenv("ARF_ENABLE_EXECUTION", "false").lower() == "true"
|
| 102 |
|
| 103 |
# ===== OPEN TELEMETRY =====
|
|
@@ -501,7 +521,27 @@ async def execute_intent_endpoint(
|
|
| 501 |
# mode is the documented fallback (see EnterpriseExecutor.execute).
|
| 502 |
approval_store = getattr(request.app.state, "approval_store", None)
|
| 503 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 504 |
executor = EnterpriseExecutor(
|
|
|
|
| 505 |
actuator=FakeCloudActuator(),
|
| 506 |
approval_store=approval_store,
|
| 507 |
on_verified_outcome=_on_verified_outcome,
|
|
|
|
| 81 |
try:
|
| 82 |
from arf_enterprise.executor import EnterpriseExecutor
|
| 83 |
from arf_enterprise.actuators.fake import FakeCloudActuator
|
| 84 |
+
from arf_enterprise.config import EnterpriseConfig
|
| 85 |
from arf_enterprise.store import ApprovalStore, PostgresStore
|
| 86 |
from arf_enterprise.exceptions import (
|
| 87 |
ExecutionError as EnterpriseExecutionError,
|
|
|
|
| 93 |
ENTERPRISE_EXECUTOR_AVAILABLE = False
|
| 94 |
EnterpriseExecutor = None
|
| 95 |
FakeCloudActuator = None
|
| 96 |
+
EnterpriseConfig = None
|
| 97 |
ApprovalStore = None
|
| 98 |
PostgresStore = None
|
| 99 |
EnterpriseExecutionError = None
|
| 100 |
PendingApprovalError = None
|
| 101 |
EnterpriseSafetyError = None
|
| 102 |
|
| 103 |
+
|
| 104 |
+
def _trusted_signing_keys() -> List[str]:
|
| 105 |
+
"""Parse ARF_TRUSTED_SIGNING_KEYS into a list of hex fingerprints.
|
| 106 |
+
|
| 107 |
+
Comma-separated, matching EnterpriseConfig.from_env's own parsing.
|
| 108 |
+
Splitting matters: the variable holds N fingerprints, and passing the
|
| 109 |
+
raw string as a single-element list would register the literal
|
| 110 |
+
"abc,def" as one key -- so neither real key would be trusted, and an
|
| 111 |
+
unset variable would register the empty string as trusted rather than
|
| 112 |
+
trusting nothing.
|
| 113 |
+
|
| 114 |
+
An empty result is the correct fail-closed state: the ladder then
|
| 115 |
+
rejects every signed intent, which is loud and safe.
|
| 116 |
+
"""
|
| 117 |
+
raw = os.getenv("ARF_TRUSTED_SIGNING_KEYS", "")
|
| 118 |
+
return [k.strip() for k in raw.split(",") if k.strip()]
|
| 119 |
+
|
| 120 |
+
|
| 121 |
ARF_ENABLE_EXECUTION = os.getenv("ARF_ENABLE_EXECUTION", "false").lower() == "true"
|
| 122 |
|
| 123 |
# ===== OPEN TELEMETRY =====
|
|
|
|
| 521 |
# mode is the documented fallback (see EnterpriseExecutor.execute).
|
| 522 |
approval_store = getattr(request.app.state, "approval_store", None)
|
| 523 |
|
| 524 |
+
# A narrow config carrying only the trust anchor. Deliberately NOT
|
| 525 |
+
# EnterpriseConfig.from_env(), which would also pick up ARF_CLOUD,
|
| 526 |
+
# ARF_MAX_BLAST_RADIUS, ARF_ENFORCE_BUSINESS_HOURS and the audit/safety
|
| 527 |
+
# toggles -- turning on guardrails and audit logging as a side effect of
|
| 528 |
+
# enabling signing is a behaviour change nobody asked for. Everything
|
| 529 |
+
# other than the trusted keys stays on today's defaults.
|
| 530 |
+
#
|
| 531 |
+
# Without this the executor got EnterpriseConfig() with an empty
|
| 532 |
+
# trusted_signing_keys, so the ladder trusted no keys and rejected every
|
| 533 |
+
# signed intent as "Untrusted signing key" -- the whole execute path was
|
| 534 |
+
# unreachable regardless of ARF_ENABLE_EXECUTION.
|
| 535 |
+
trusted_keys = _trusted_signing_keys()
|
| 536 |
+
if not trusted_keys:
|
| 537 |
+
logger.warning(
|
| 538 |
+
"ARF_TRUSTED_SIGNING_KEYS is unset or empty; the execution ladder "
|
| 539 |
+
"trusts no signing keys and will reject every signed intent. Set it "
|
| 540 |
+
"to the hex fingerprint(s) of the key(s) permitted to sign intents."
|
| 541 |
+
)
|
| 542 |
+
|
| 543 |
executor = EnterpriseExecutor(
|
| 544 |
+
config=EnterpriseConfig(trusted_signing_keys=trusted_keys),
|
| 545 |
actuator=FakeCloudActuator(),
|
| 546 |
approval_store=approval_store,
|
| 547 |
on_verified_outcome=_on_verified_outcome,
|
app/api/routes_incidents.py
CHANGED
|
@@ -33,7 +33,8 @@ from agentic_reliability_framework.core.models.event import (
|
|
| 33 |
from app.api.deps import verify_internal_key
|
| 34 |
from app.causal_explainer import CausalExplainer
|
| 35 |
from app.core.storage import incident_history
|
| 36 |
-
from app.core
|
|
|
|
| 37 |
|
| 38 |
logger = logging.getLogger(__name__)
|
| 39 |
|
|
@@ -227,7 +228,7 @@ async def evaluate_incident(
|
|
| 227 |
# ------------------------------------------------------------------
|
| 228 |
# Asynchronous usage logging
|
| 229 |
# ------------------------------------------------------------------
|
| 230 |
-
if tracker:
|
| 231 |
record = UsageRecord(
|
| 232 |
api_key=api_key,
|
| 233 |
tier=tier,
|
|
@@ -237,7 +238,7 @@ async def evaluate_incident(
|
|
| 237 |
response=response_data,
|
| 238 |
processing_ms=(time.time() - start_time) * 1000,
|
| 239 |
)
|
| 240 |
-
await tracker.increment_usage_async(record, background_tasks)
|
| 241 |
|
| 242 |
logger.warning(
|
| 243 |
"Deprecated endpoint /v1/incidents/evaluate called by key %s",
|
|
@@ -250,7 +251,7 @@ async def evaluate_incident(
|
|
| 250 |
except Exception as exc:
|
| 251 |
error_msg = str(exc)
|
| 252 |
logger.exception("Error in evaluate_incident (deprecated endpoint)")
|
| 253 |
-
if tracker:
|
| 254 |
record = UsageRecord(
|
| 255 |
api_key=api_key,
|
| 256 |
tier=tier,
|
|
@@ -260,5 +261,5 @@ async def evaluate_incident(
|
|
| 260 |
error=error_msg,
|
| 261 |
processing_ms=(time.time() - start_time) * 1000,
|
| 262 |
)
|
| 263 |
-
await tracker.increment_usage_async(record, background_tasks)
|
| 264 |
raise HTTPException(status_code=500, detail="Internal server error")
|
|
|
|
| 33 |
from app.api.deps import verify_internal_key
|
| 34 |
from app.causal_explainer import CausalExplainer
|
| 35 |
from app.core.storage import incident_history
|
| 36 |
+
from app.core import usage_tracker
|
| 37 |
+
from app.core.usage_tracker import UsageRecord, enforce_quota
|
| 38 |
|
| 39 |
logger = logging.getLogger(__name__)
|
| 40 |
|
|
|
|
| 228 |
# ------------------------------------------------------------------
|
| 229 |
# Asynchronous usage logging
|
| 230 |
# ------------------------------------------------------------------
|
| 231 |
+
if usage_tracker.tracker:
|
| 232 |
record = UsageRecord(
|
| 233 |
api_key=api_key,
|
| 234 |
tier=tier,
|
|
|
|
| 238 |
response=response_data,
|
| 239 |
processing_ms=(time.time() - start_time) * 1000,
|
| 240 |
)
|
| 241 |
+
await usage_tracker.tracker.increment_usage_async(record, background_tasks)
|
| 242 |
|
| 243 |
logger.warning(
|
| 244 |
"Deprecated endpoint /v1/incidents/evaluate called by key %s",
|
|
|
|
| 251 |
except Exception as exc:
|
| 252 |
error_msg = str(exc)
|
| 253 |
logger.exception("Error in evaluate_incident (deprecated endpoint)")
|
| 254 |
+
if usage_tracker.tracker:
|
| 255 |
record = UsageRecord(
|
| 256 |
api_key=api_key,
|
| 257 |
tier=tier,
|
|
|
|
| 261 |
error=error_msg,
|
| 262 |
processing_ms=(time.time() - start_time) * 1000,
|
| 263 |
)
|
| 264 |
+
await usage_tracker.tracker.increment_usage_async(record, background_tasks)
|
| 265 |
raise HTTPException(status_code=500, detail="Internal server error")
|
app/api/routes_payments.py
CHANGED
|
@@ -8,7 +8,8 @@ import stripe
|
|
| 8 |
from fastapi import APIRouter, HTTPException
|
| 9 |
from pydantic import BaseModel
|
| 10 |
|
| 11 |
-
from app.core
|
|
|
|
| 12 |
|
| 13 |
logger = logging.getLogger(__name__)
|
| 14 |
|
|
@@ -31,11 +32,11 @@ async def create_checkout_session(req: CheckoutRequest):
|
|
| 31 |
"""Create a Stripe Checkout session for the Pro tier."""
|
| 32 |
if not stripe.api_key:
|
| 33 |
raise HTTPException(status_code=500, detail="Stripe not configured")
|
| 34 |
-
if not tracker:
|
| 35 |
raise HTTPException(status_code=503, detail="Usage tracking service not initialised")
|
| 36 |
|
| 37 |
# Verify the API key exists and is free tier
|
| 38 |
-
tier = tracker.get_tier(req.api_key)
|
| 39 |
if tier != Tier.FREE:
|
| 40 |
raise HTTPException(status_code=400,
|
| 41 |
detail="Only free tier keys can be upgraded")
|
|
@@ -47,7 +48,7 @@ async def create_checkout_session(req: CheckoutRequest):
|
|
| 47 |
# third party with no reason to ever see it. tenant_id is an opaque,
|
| 48 |
# non-secret row identifier and is exactly what the webhook needs to
|
| 49 |
# look up which tenant's keys to retier.
|
| 50 |
-
tenant_id = tracker.get_tenant_id(req.api_key)
|
| 51 |
|
| 52 |
try:
|
| 53 |
checkout_session = stripe.checkout.Session.create(
|
|
|
|
| 8 |
from fastapi import APIRouter, HTTPException
|
| 9 |
from pydantic import BaseModel
|
| 10 |
|
| 11 |
+
from app.core import usage_tracker
|
| 12 |
+
from app.core.usage_tracker import Tier
|
| 13 |
|
| 14 |
logger = logging.getLogger(__name__)
|
| 15 |
|
|
|
|
| 32 |
"""Create a Stripe Checkout session for the Pro tier."""
|
| 33 |
if not stripe.api_key:
|
| 34 |
raise HTTPException(status_code=500, detail="Stripe not configured")
|
| 35 |
+
if not usage_tracker.tracker:
|
| 36 |
raise HTTPException(status_code=503, detail="Usage tracking service not initialised")
|
| 37 |
|
| 38 |
# Verify the API key exists and is free tier
|
| 39 |
+
tier = usage_tracker.tracker.get_tier(req.api_key)
|
| 40 |
if tier != Tier.FREE:
|
| 41 |
raise HTTPException(status_code=400,
|
| 42 |
detail="Only free tier keys can be upgraded")
|
|
|
|
| 48 |
# third party with no reason to ever see it. tenant_id is an opaque,
|
| 49 |
# non-secret row identifier and is exactly what the webhook needs to
|
| 50 |
# look up which tenant's keys to retier.
|
| 51 |
+
tenant_id = usage_tracker.tracker.get_tenant_id(req.api_key)
|
| 52 |
|
| 53 |
try:
|
| 54 |
checkout_session = stripe.checkout.Session.create(
|
app/api/routes_users.py
CHANGED
|
@@ -9,7 +9,8 @@ from sqlalchemy.orm import Session
|
|
| 9 |
from slowapi import Limiter
|
| 10 |
from slowapi.util import get_remote_address
|
| 11 |
|
| 12 |
-
from app.core
|
|
|
|
| 13 |
from app.api.deps import get_db
|
| 14 |
from app.database.models_intents import TenantDB # <-- NEW
|
| 15 |
|
|
@@ -30,7 +31,7 @@ async def register_user(
|
|
| 30 |
Public endpoint to create a new freeβtier API key and a new tenant.
|
| 31 |
Rateβlimited to 5 requests per hour per IP address.
|
| 32 |
"""
|
| 33 |
-
if tracker is None:
|
| 34 |
raise HTTPException(status_code=503, detail="Usage tracking service not initialised")
|
| 35 |
|
| 36 |
# 1. Create a new tenant in the main database
|
|
@@ -48,7 +49,7 @@ async def register_user(
|
|
| 48 |
|
| 49 |
# 2. Generate a new API key for this tenant
|
| 50 |
new_key = f"sk_free_{uuid.uuid4().hex[:24]}"
|
| 51 |
-
success = tracker.get_or_create_api_key(api_key=new_key, tenant_id=tenant_id, tier=Tier.FREE)
|
| 52 |
if not success:
|
| 53 |
# Rollback tenant creation if key creation fails
|
| 54 |
db.delete(new_tenant)
|
|
|
|
| 9 |
from slowapi import Limiter
|
| 10 |
from slowapi.util import get_remote_address
|
| 11 |
|
| 12 |
+
from app.core import usage_tracker
|
| 13 |
+
from app.core.usage_tracker import enforce_quota, Tier
|
| 14 |
from app.api.deps import get_db
|
| 15 |
from app.database.models_intents import TenantDB # <-- NEW
|
| 16 |
|
|
|
|
| 31 |
Public endpoint to create a new freeβtier API key and a new tenant.
|
| 32 |
Rateβlimited to 5 requests per hour per IP address.
|
| 33 |
"""
|
| 34 |
+
if usage_tracker.tracker is None:
|
| 35 |
raise HTTPException(status_code=503, detail="Usage tracking service not initialised")
|
| 36 |
|
| 37 |
# 1. Create a new tenant in the main database
|
|
|
|
| 49 |
|
| 50 |
# 2. Generate a new API key for this tenant
|
| 51 |
new_key = f"sk_free_{uuid.uuid4().hex[:24]}"
|
| 52 |
+
success = usage_tracker.tracker.get_or_create_api_key(api_key=new_key, tenant_id=tenant_id, tier=Tier.FREE)
|
| 53 |
if not success:
|
| 54 |
# Rollback tenant creation if key creation fails
|
| 55 |
db.delete(new_tenant)
|
app/core/usage_tracker.py
CHANGED
|
@@ -629,6 +629,17 @@ class UsageTracker:
|
|
| 629 |
# --------------------------------------------------------------------------
|
| 630 |
# Global instance and FastAPI dependency
|
| 631 |
# --------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 632 |
tracker: Optional[UsageTracker] = None
|
| 633 |
|
| 634 |
|
|
|
|
| 629 |
# --------------------------------------------------------------------------
|
| 630 |
# Global instance and FastAPI dependency
|
| 631 |
# --------------------------------------------------------------------------
|
| 632 |
+
# Rebound by init_tracker() during the app lifespan, which means consumers
|
| 633 |
+
# MUST reach it through the module -- `from app.core import usage_tracker`,
|
| 634 |
+
# then `usage_tracker.tracker`. Never `from app.core.usage_tracker import
|
| 635 |
+
# tracker`: that copies the *binding* (None) at import time, and init_tracker
|
| 636 |
+
# rebinding this global does not update the importer's copy. Five modules
|
| 637 |
+
# did exactly that (main, routes_admin, routes_incidents, routes_payments,
|
| 638 |
+
# routes_users) and every one of them saw None forever -- silently skipping
|
| 639 |
+
# metering and disabling signup/checkout, and crashing the Render deploy
|
| 640 |
+
# outright once main.py called a method on it. Functions defined *in* this
|
| 641 |
+
# module (enforce_quota, update_key_tier*) are safe to import by name: they
|
| 642 |
+
# resolve `tracker` here, at call time.
|
| 643 |
tracker: Optional[UsageTracker] = None
|
| 644 |
|
| 645 |
|
app/main.py
CHANGED
|
@@ -83,7 +83,8 @@ from agentic_reliability_framework.core.temporal_reliability import (
|
|
| 83 |
)
|
| 84 |
|
| 85 |
# ββ Usage tracker ββββββββββββββββββββββββββββββββββββββββββββ
|
| 86 |
-
from app.core
|
|
|
|
| 87 |
|
| 88 |
from app.api import (
|
| 89 |
routes_governance,
|
|
@@ -285,8 +286,8 @@ async def lifespan(app: FastAPI):
|
|
| 285 |
# the health endpoint answers, the deploy succeeds, and API
|
| 286 |
# requests get a clean 503 from enforce_quota until the database
|
| 287 |
# is reachable -- at which point they recover with no redeploy.
|
| 288 |
-
postgres_ready = tracker.warm_up()
|
| 289 |
-
if
|
| 290 |
logger.error(
|
| 291 |
"Usage tracker started WITHOUT a Postgres connection: api_keys "
|
| 292 |
"is unreachable, so API-key validation and quota enforcement "
|
|
@@ -302,7 +303,7 @@ async def lifespan(app: FastAPI):
|
|
| 302 |
# kills the process -- reintroducing the crash loop the warm-up
|
| 303 |
# above exists to prevent.
|
| 304 |
api_keys_json = os.getenv("ARF_API_KEYS", "{}")
|
| 305 |
-
if
|
| 306 |
logger.warning(
|
| 307 |
"Skipping ARF_API_KEYS seeding: Postgres is unreachable. "
|
| 308 |
"Seeded keys will not exist until the database recovers "
|
|
@@ -311,41 +312,38 @@ async def lifespan(app: FastAPI):
|
|
| 311 |
api_keys_json = "{}"
|
| 312 |
try:
|
| 313 |
api_keys = json.loads(api_keys_json)
|
| 314 |
-
|
| 315 |
-
|
| 316 |
-
|
| 317 |
-
|
| 318 |
-
|
| 319 |
-
|
| 320 |
-
|
| 321 |
-
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
|
| 326 |
-
|
| 327 |
-
|
| 328 |
-
|
| 329 |
-
|
| 330 |
-
|
| 331 |
-
|
| 332 |
-
|
| 333 |
-
|
| 334 |
-
|
| 335 |
-
|
| 336 |
-
|
| 337 |
-
)
|
| 338 |
except json.JSONDecodeError:
|
| 339 |
logger.warning(
|
| 340 |
"ARF_API_KEYS environment variable is not valid JSON; skipping seeding."
|
| 341 |
)
|
| 342 |
-
app.state.usage_tracker = tracker
|
| 343 |
-
if
|
| 344 |
logger.info("β
Usage tracker ready.")
|
| 345 |
-
elif tracker:
|
| 346 |
-
logger.warning("β οΈ Usage tracker started in degraded mode (no Postgres).")
|
| 347 |
else:
|
| 348 |
-
logger.
|
| 349 |
except Exception as e:
|
| 350 |
# Still fail closed on genuine configuration errors -- a missing
|
| 351 |
# or too-short ARF_KEY_PEPPER, an unset DATABASE_URL. Those never
|
|
|
|
| 83 |
)
|
| 84 |
|
| 85 |
# ββ Usage tracker ββββββββββββββββββββββββββββββββββββββββββββ
|
| 86 |
+
from app.core import usage_tracker
|
| 87 |
+
from app.core.usage_tracker import init_tracker, Tier
|
| 88 |
|
| 89 |
from app.api import (
|
| 90 |
routes_governance,
|
|
|
|
| 286 |
# the health endpoint answers, the deploy succeeds, and API
|
| 287 |
# requests get a clean 503 from enforce_quota until the database
|
| 288 |
# is reachable -- at which point they recover with no redeploy.
|
| 289 |
+
postgres_ready = usage_tracker.tracker.warm_up()
|
| 290 |
+
if not postgres_ready:
|
| 291 |
logger.error(
|
| 292 |
"Usage tracker started WITHOUT a Postgres connection: api_keys "
|
| 293 |
"is unreachable, so API-key validation and quota enforcement "
|
|
|
|
| 303 |
# kills the process -- reintroducing the crash loop the warm-up
|
| 304 |
# above exists to prevent.
|
| 305 |
api_keys_json = os.getenv("ARF_API_KEYS", "{}")
|
| 306 |
+
if not postgres_ready and api_keys_json not in ("", "{}"):
|
| 307 |
logger.warning(
|
| 308 |
"Skipping ARF_API_KEYS seeding: Postgres is unreachable. "
|
| 309 |
"Seeded keys will not exist until the database recovers "
|
|
|
|
| 312 |
api_keys_json = "{}"
|
| 313 |
try:
|
| 314 |
api_keys = json.loads(api_keys_json)
|
| 315 |
+
for key, tier_str in api_keys.items():
|
| 316 |
+
try:
|
| 317 |
+
tier = Tier(tier_str.lower())
|
| 318 |
+
# Previously called get_or_create_api_key(key, tier)
|
| 319 |
+
# -- tier was silently accepted as tenant_id, so
|
| 320 |
+
# every seeded key of the same tier collided onto
|
| 321 |
+
# one bogus tenant_id. These are demo/env-seeded
|
| 322 |
+
# keys with no real TenantDB row, but each still
|
| 323 |
+
# needs its own tenant_id to avoid cross-key
|
| 324 |
+
# contamination in tenant-scoped state elsewhere
|
| 325 |
+
# (BetaStateDB, IntentDB, decision audit log). A
|
| 326 |
+
# fixed-length key prefix isn't safe here: keys
|
| 327 |
+
# generated elsewhere in this codebase share an
|
| 328 |
+
# 8-char prefix ("sk_live_"/"sk_free_"), so a
|
| 329 |
+
# prefix-based id would collide the same way the
|
| 330 |
+
# original bug did. Hash the whole key instead.
|
| 331 |
+
tenant_id = "env-seed-" + hashlib.sha256(key.encode()).hexdigest()[:16]
|
| 332 |
+
usage_tracker.tracker.get_or_create_api_key(key, tenant_id=tenant_id, tier=tier)
|
| 333 |
+
logger.info(f"Seeded API key for tier {tier.value}")
|
| 334 |
+
except ValueError:
|
| 335 |
+
logger.warning(
|
| 336 |
+
f"Invalid tier '{tier_str}' for key {key}, skipping"
|
| 337 |
+
)
|
|
|
|
| 338 |
except json.JSONDecodeError:
|
| 339 |
logger.warning(
|
| 340 |
"ARF_API_KEYS environment variable is not valid JSON; skipping seeding."
|
| 341 |
)
|
| 342 |
+
app.state.usage_tracker = usage_tracker.tracker
|
| 343 |
+
if postgres_ready:
|
| 344 |
logger.info("β
Usage tracker ready.")
|
|
|
|
|
|
|
| 345 |
else:
|
| 346 |
+
logger.warning("β οΈ Usage tracker started in degraded mode (no Postgres).")
|
| 347 |
except Exception as e:
|
| 348 |
# Still fail closed on genuine configuration errors -- a missing
|
| 349 |
# or too-short ARF_KEY_PEPPER, an unset DATABASE_URL. Those never
|
tests/test_tracker_binding.py
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Guard against re-introducing the stale `tracker` from-import.
|
| 2 |
+
|
| 3 |
+
`app.core.usage_tracker.tracker` starts as None and is rebound by
|
| 4 |
+
`init_tracker()` during the lifespan. `from app.core.usage_tracker import
|
| 5 |
+
tracker` copies the *binding* -- None -- at import time, and the later
|
| 6 |
+
rebinding never reaches the importer. Five modules did this: main.py
|
| 7 |
+
crashed the Render deploy with "'NoneType' object has no attribute
|
| 8 |
+
'warm_up'", routes_admin's 20 unguarded uses would have 500'd, and
|
| 9 |
+
routes_incidents/payments/users each had a `if tracker` guard that could
|
| 10 |
+
only ever take the None branch -- so usage went unmetered and signup and
|
| 11 |
+
checkout were permanently disabled, silently.
|
| 12 |
+
|
| 13 |
+
This is a source-level check on purpose. Reproducing the bug at runtime
|
| 14 |
+
needs a real UsageTracker (Postgres, pepper) and would only cover the
|
| 15 |
+
modules the test happened to import; parsing every module catches the
|
| 16 |
+
next one too, and costs nothing.
|
| 17 |
+
"""
|
| 18 |
+
|
| 19 |
+
import ast
|
| 20 |
+
import pathlib
|
| 21 |
+
|
| 22 |
+
import pytest
|
| 23 |
+
|
| 24 |
+
APP = pathlib.Path(__file__).resolve().parent.parent / "app"
|
| 25 |
+
|
| 26 |
+
# Names safe to import directly: functions defined in usage_tracker.py
|
| 27 |
+
# resolve the module global at call time, so they always see the live
|
| 28 |
+
# instance. Only the mutable module-level object itself is unsafe.
|
| 29 |
+
UNSAFE_NAMES = {"tracker"}
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
def _source_files():
|
| 33 |
+
return sorted(p for p in APP.rglob("*.py"))
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
@pytest.mark.parametrize("path", _source_files(), ids=lambda p: p.name)
|
| 37 |
+
def test_tracker_is_never_imported_by_name(path):
|
| 38 |
+
tree = ast.parse(path.read_text(encoding="utf-8"), filename=str(path))
|
| 39 |
+
for node in ast.walk(tree):
|
| 40 |
+
if not isinstance(node, ast.ImportFrom):
|
| 41 |
+
continue
|
| 42 |
+
if node.module != "app.core.usage_tracker":
|
| 43 |
+
continue
|
| 44 |
+
offenders = sorted(
|
| 45 |
+
{a.name for a in node.names} & UNSAFE_NAMES
|
| 46 |
+
)
|
| 47 |
+
assert not offenders, (
|
| 48 |
+
f"{path.name}:{node.lineno} imports {offenders} by name from "
|
| 49 |
+
"app.core.usage_tracker. That binding is None at import time and "
|
| 50 |
+
"init_tracker() will not update it. Use `from app.core import "
|
| 51 |
+
"usage_tracker` and reference `usage_tracker.tracker` instead."
|
| 52 |
+
)
|