Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Enforce premium quota across sessions daily (#287)
Browse filesCo-authored-by: OpenAI Codex <codex@openai.com>
- agent/core/session_persistence.py +4 -0
- backend/routes/agent.py +13 -6
- backend/session_manager.py +39 -8
- backend/user_quotas.py +9 -4
- frontend/src/components/SessionChat.tsx +6 -6
- frontend/src/store/sessionStore.ts +8 -4
- tests/unit/test_agent_model_gating.py +60 -0
- tests/unit/test_session_manager_persistence.py +42 -0
agent/core/session_persistence.py
CHANGED
|
@@ -175,6 +175,7 @@ class MongoSessionStore(NoopSessionStore):
|
|
| 175 |
turn_count: int = 0,
|
| 176 |
pending_approval: list[dict[str, Any]] | None = None,
|
| 177 |
claude_counted: bool = False,
|
|
|
|
| 178 |
premium_user_billed: bool = False,
|
| 179 |
notification_destinations: list[str] | None = None,
|
| 180 |
auto_approval_enabled: bool = False,
|
|
@@ -207,6 +208,7 @@ class MongoSessionStore(NoopSessionStore):
|
|
| 207 |
"turn_count": turn_count,
|
| 208 |
"pending_approval": pending_approval or [],
|
| 209 |
"claude_counted": claude_counted,
|
|
|
|
| 210 |
"premium_user_billed": premium_user_billed,
|
| 211 |
"notification_destinations": notification_destinations or [],
|
| 212 |
"auto_approval_enabled": auto_approval_enabled,
|
|
@@ -230,6 +232,7 @@ class MongoSessionStore(NoopSessionStore):
|
|
| 230 |
turn_count: int = 0,
|
| 231 |
pending_approval: list[dict[str, Any]] | None = None,
|
| 232 |
claude_counted: bool = False,
|
|
|
|
| 233 |
premium_user_billed: bool = False,
|
| 234 |
created_at: datetime | None = None,
|
| 235 |
notification_destinations: list[str] | None = None,
|
|
@@ -255,6 +258,7 @@ class MongoSessionStore(NoopSessionStore):
|
|
| 255 |
turn_count=turn_count,
|
| 256 |
pending_approval=pending_approval,
|
| 257 |
claude_counted=claude_counted,
|
|
|
|
| 258 |
premium_user_billed=premium_user_billed,
|
| 259 |
notification_destinations=notification_destinations,
|
| 260 |
auto_approval_enabled=auto_approval_enabled,
|
|
|
|
| 175 |
turn_count: int = 0,
|
| 176 |
pending_approval: list[dict[str, Any]] | None = None,
|
| 177 |
claude_counted: bool = False,
|
| 178 |
+
claude_counted_day: str | None = None,
|
| 179 |
premium_user_billed: bool = False,
|
| 180 |
notification_destinations: list[str] | None = None,
|
| 181 |
auto_approval_enabled: bool = False,
|
|
|
|
| 208 |
"turn_count": turn_count,
|
| 209 |
"pending_approval": pending_approval or [],
|
| 210 |
"claude_counted": claude_counted,
|
| 211 |
+
"claude_counted_day": claude_counted_day,
|
| 212 |
"premium_user_billed": premium_user_billed,
|
| 213 |
"notification_destinations": notification_destinations or [],
|
| 214 |
"auto_approval_enabled": auto_approval_enabled,
|
|
|
|
| 232 |
turn_count: int = 0,
|
| 233 |
pending_approval: list[dict[str, Any]] | None = None,
|
| 234 |
claude_counted: bool = False,
|
| 235 |
+
claude_counted_day: str | None = None,
|
| 236 |
premium_user_billed: bool = False,
|
| 237 |
created_at: datetime | None = None,
|
| 238 |
notification_destinations: list[str] | None = None,
|
|
|
|
| 258 |
turn_count=turn_count,
|
| 259 |
pending_approval=pending_approval,
|
| 260 |
claude_counted=claude_counted,
|
| 261 |
+
claude_counted_day=claude_counted_day,
|
| 262 |
premium_user_billed=premium_user_billed,
|
| 263 |
notification_destinations=notification_destinations,
|
| 264 |
auto_approval_enabled=auto_approval_enabled,
|
backend/routes/agent.py
CHANGED
|
@@ -207,21 +207,23 @@ async def _enforce_premium_model_quota(
|
|
| 207 |
|
| 208 |
Runs at *message-submit* time, not session-create time — so spinning up a
|
| 209 |
premium-model session to look around doesn't burn quota. The
|
| 210 |
-
``
|
| 211 |
-
same session
|
|
|
|
| 212 |
|
| 213 |
Subsidizes the daily allowance (free = 2 for default premium, pro = 20
|
| 214 |
across premium models), organization-billed through the HF Router. Opus and
|
| 215 |
GPT-5.5 are pro-only before quota is charged. Past the allowance, premium
|
| 216 |
router models flip the session to ``premium_user_billed`` so the call bills
|
| 217 |
the user's own HF token instead of blocking. No-ops when the model isn't
|
| 218 |
-
premium or when this session's billing has already been decided.
|
| 219 |
"""
|
| 220 |
model_name = agent_session.session.config.model_name
|
| 221 |
if not _is_premium_model(model_name):
|
| 222 |
return
|
| 223 |
_reject_model_unavailable_for_plan(model_name, user)
|
| 224 |
-
|
|
|
|
| 225 |
return
|
| 226 |
user_id = user["user_id"]
|
| 227 |
plan = user.get("plan", "free")
|
|
@@ -241,7 +243,12 @@ async def _enforce_premium_model_quota(
|
|
| 241 |
# Past the subsidized allowance on a user-billable model: bill the
|
| 242 |
# user's own HF (OAuth) token for this session instead of blocking.
|
| 243 |
agent_session.session.premium_user_billed = True
|
|
|
|
|
|
|
|
|
|
|
|
|
| 244 |
agent_session.claude_counted = True
|
|
|
|
| 245 |
await session_manager.persist_session_snapshot(agent_session)
|
| 246 |
|
| 247 |
|
|
@@ -909,8 +916,8 @@ async def chat_sse(
|
|
| 909 |
approvals = body.get("approvals")
|
| 910 |
|
| 911 |
# Gate user-message sends against the daily premium-model quota. Approvals are
|
| 912 |
-
# continuations of an in-progress turn
|
| 913 |
-
#
|
| 914 |
if text is not None and not approvals:
|
| 915 |
try:
|
| 916 |
await _enforce_premium_model_quota(user, agent_session)
|
|
|
|
| 207 |
|
| 208 |
Runs at *message-submit* time, not session-create time — so spinning up a
|
| 209 |
premium-model session to look around doesn't burn quota. The
|
| 210 |
+
``claude_counted_day`` flag on ``AgentSession`` guards against re-counting
|
| 211 |
+
the same session on the same day, while still counting old sessions when
|
| 212 |
+
they are used again on a later day.
|
| 213 |
|
| 214 |
Subsidizes the daily allowance (free = 2 for default premium, pro = 20
|
| 215 |
across premium models), organization-billed through the HF Router. Opus and
|
| 216 |
GPT-5.5 are pro-only before quota is charged. Past the allowance, premium
|
| 217 |
router models flip the session to ``premium_user_billed`` so the call bills
|
| 218 |
the user's own HF token instead of blocking. No-ops when the model isn't
|
| 219 |
+
premium or when this session's billing has already been decided for today.
|
| 220 |
"""
|
| 221 |
model_name = agent_session.session.config.model_name
|
| 222 |
if not _is_premium_model(model_name):
|
| 223 |
return
|
| 224 |
_reject_model_unavailable_for_plan(model_name, user)
|
| 225 |
+
quota_day = user_quotas.current_quota_day()
|
| 226 |
+
if agent_session.claude_counted and agent_session.claude_counted_day == quota_day:
|
| 227 |
return
|
| 228 |
user_id = user["user_id"]
|
| 229 |
plan = user.get("plan", "free")
|
|
|
|
| 243 |
# Past the subsidized allowance on a user-billable model: bill the
|
| 244 |
# user's own HF (OAuth) token for this session instead of blocking.
|
| 245 |
agent_session.session.premium_user_billed = True
|
| 246 |
+
else:
|
| 247 |
+
# A session that overflowed on a previous day can use today's
|
| 248 |
+
# subsidized allowance again if quota is available.
|
| 249 |
+
agent_session.session.premium_user_billed = False
|
| 250 |
agent_session.claude_counted = True
|
| 251 |
+
agent_session.claude_counted_day = quota_day
|
| 252 |
await session_manager.persist_session_snapshot(agent_session)
|
| 253 |
|
| 254 |
|
|
|
|
| 916 |
approvals = body.get("approvals")
|
| 917 |
|
| 918 |
# Gate user-message sends against the daily premium-model quota. Approvals are
|
| 919 |
+
# continuations of an in-progress turn, so the relevant quota decision was
|
| 920 |
+
# made when that user message was submitted.
|
| 921 |
if text is not None and not approvals:
|
| 922 |
try:
|
| 923 |
await _enforce_premium_model_quota(user, agent_session)
|
backend/session_manager.py
CHANGED
|
@@ -6,7 +6,7 @@ import logging
|
|
| 6 |
import os
|
| 7 |
import uuid
|
| 8 |
from dataclasses import dataclass, field
|
| 9 |
-
from datetime import datetime, timedelta
|
| 10 |
from pathlib import Path
|
| 11 |
from typing import Any, Optional
|
| 12 |
|
|
@@ -113,9 +113,28 @@ class AgentSession:
|
|
| 113 |
is_reaping: bool = False
|
| 114 |
broadcaster: Any = None
|
| 115 |
title: str | None = None
|
| 116 |
-
# True once this session has been counted against
|
| 117 |
-
#
|
| 118 |
claude_counted: bool = False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
|
| 120 |
|
| 121 |
class SessionCapacityError(Exception):
|
|
@@ -600,6 +619,7 @@ class SessionManager:
|
|
| 600 |
agent_session.session
|
| 601 |
),
|
| 602 |
claude_counted=agent_session.claude_counted,
|
|
|
|
| 603 |
premium_user_billed=getattr(
|
| 604 |
agent_session.session, "premium_user_billed", False
|
| 605 |
),
|
|
@@ -696,6 +716,13 @@ class SessionManager:
|
|
| 696 |
premium_user_billed=bool(meta.get("premium_user_billed", False)),
|
| 697 |
claude_counted=bool(meta.get("claude_counted")),
|
| 698 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 699 |
event_queue: asyncio.Queue = asyncio.Queue()
|
| 700 |
submission_queue: asyncio.Queue = asyncio.Queue()
|
| 701 |
tool_router, session = await asyncio.to_thread(
|
|
@@ -779,6 +806,7 @@ class SessionManager:
|
|
| 779 |
is_active=True,
|
| 780 |
is_processing=False,
|
| 781 |
claude_counted=claude_counted,
|
|
|
|
| 782 |
title=meta.get("title"),
|
| 783 |
)
|
| 784 |
started = await self._start_agent_session(
|
|
@@ -1505,10 +1533,8 @@ class SessionManager:
|
|
| 1505 |
agent_session.session.notification_destinations
|
| 1506 |
),
|
| 1507 |
"auto_approval": self._auto_approval_summary(agent_session.session),
|
| 1508 |
-
"premium_user_billed":
|
| 1509 |
-
|
| 1510 |
-
),
|
| 1511 |
-
"premium_quota_counted": agent_session.claude_counted,
|
| 1512 |
}
|
| 1513 |
|
| 1514 |
def set_notification_destinations(
|
|
@@ -1562,6 +1588,10 @@ class SessionManager:
|
|
| 1562 |
else:
|
| 1563 |
created_at_str = str(created_at or datetime.utcnow().isoformat())
|
| 1564 |
pending = self._pending_docs_for_api(row.get("pending_approval") or [])
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1565 |
results.append(
|
| 1566 |
{
|
| 1567 |
"session_id": str(sid),
|
|
@@ -1575,8 +1605,9 @@ class SessionManager:
|
|
| 1575 |
"title": row.get("title"),
|
| 1576 |
"premium_user_billed": bool(
|
| 1577 |
row.get("premium_user_billed", False)
|
|
|
|
| 1578 |
),
|
| 1579 |
-
"premium_quota_counted":
|
| 1580 |
"notification_destinations": row.get(
|
| 1581 |
"notification_destinations"
|
| 1582 |
)
|
|
|
|
| 6 |
import os
|
| 7 |
import uuid
|
| 8 |
from dataclasses import dataclass, field
|
| 9 |
+
from datetime import UTC, datetime, timedelta
|
| 10 |
from pathlib import Path
|
| 11 |
from typing import Any, Optional
|
| 12 |
|
|
|
|
| 113 |
is_reaping: bool = False
|
| 114 |
broadcaster: Any = None
|
| 115 |
title: str | None = None
|
| 116 |
+
# True once this session has ever been counted against premium quota.
|
| 117 |
+
# claude_counted_day decides whether it has already consumed today's cap.
|
| 118 |
claude_counted: bool = False
|
| 119 |
+
claude_counted_day: str | None = None
|
| 120 |
+
|
| 121 |
+
|
| 122 |
+
def _quota_day_today() -> str:
|
| 123 |
+
return datetime.now(UTC).date().isoformat()
|
| 124 |
+
|
| 125 |
+
|
| 126 |
+
def _quota_counted_today(agent_session: AgentSession) -> bool:
|
| 127 |
+
return (
|
| 128 |
+
agent_session.claude_counted
|
| 129 |
+
and agent_session.claude_counted_day == _quota_day_today()
|
| 130 |
+
)
|
| 131 |
+
|
| 132 |
+
|
| 133 |
+
def _premium_user_billed_today(agent_session: AgentSession) -> bool:
|
| 134 |
+
return bool(
|
| 135 |
+
getattr(agent_session.session, "premium_user_billed", False)
|
| 136 |
+
and _quota_counted_today(agent_session)
|
| 137 |
+
)
|
| 138 |
|
| 139 |
|
| 140 |
class SessionCapacityError(Exception):
|
|
|
|
| 619 |
agent_session.session
|
| 620 |
),
|
| 621 |
claude_counted=agent_session.claude_counted,
|
| 622 |
+
claude_counted_day=agent_session.claude_counted_day,
|
| 623 |
premium_user_billed=getattr(
|
| 624 |
agent_session.session, "premium_user_billed", False
|
| 625 |
),
|
|
|
|
| 716 |
premium_user_billed=bool(meta.get("premium_user_billed", False)),
|
| 717 |
claude_counted=bool(meta.get("claude_counted")),
|
| 718 |
)
|
| 719 |
+
claude_counted_day = (
|
| 720 |
+
str(meta.get("claude_counted_day"))
|
| 721 |
+
if meta.get("claude_counted_day")
|
| 722 |
+
else None
|
| 723 |
+
)
|
| 724 |
+
if not claude_counted:
|
| 725 |
+
claude_counted_day = None
|
| 726 |
event_queue: asyncio.Queue = asyncio.Queue()
|
| 727 |
submission_queue: asyncio.Queue = asyncio.Queue()
|
| 728 |
tool_router, session = await asyncio.to_thread(
|
|
|
|
| 806 |
is_active=True,
|
| 807 |
is_processing=False,
|
| 808 |
claude_counted=claude_counted,
|
| 809 |
+
claude_counted_day=claude_counted_day,
|
| 810 |
title=meta.get("title"),
|
| 811 |
)
|
| 812 |
started = await self._start_agent_session(
|
|
|
|
| 1533 |
agent_session.session.notification_destinations
|
| 1534 |
),
|
| 1535 |
"auto_approval": self._auto_approval_summary(agent_session.session),
|
| 1536 |
+
"premium_user_billed": _premium_user_billed_today(agent_session),
|
| 1537 |
+
"premium_quota_counted": _quota_counted_today(agent_session),
|
|
|
|
|
|
|
| 1538 |
}
|
| 1539 |
|
| 1540 |
def set_notification_destinations(
|
|
|
|
| 1588 |
else:
|
| 1589 |
created_at_str = str(created_at or datetime.utcnow().isoformat())
|
| 1590 |
pending = self._pending_docs_for_api(row.get("pending_approval") or [])
|
| 1591 |
+
quota_counted_today = (
|
| 1592 |
+
bool(row.get("claude_counted", False))
|
| 1593 |
+
and row.get("claude_counted_day") == _quota_day_today()
|
| 1594 |
+
)
|
| 1595 |
results.append(
|
| 1596 |
{
|
| 1597 |
"session_id": str(sid),
|
|
|
|
| 1605 |
"title": row.get("title"),
|
| 1606 |
"premium_user_billed": bool(
|
| 1607 |
row.get("premium_user_billed", False)
|
| 1608 |
+
and quota_counted_today
|
| 1609 |
),
|
| 1610 |
+
"premium_quota_counted": quota_counted_today,
|
| 1611 |
"notification_destinations": row.get(
|
| 1612 |
"notification_destinations"
|
| 1613 |
)
|
backend/user_quotas.py
CHANGED
|
@@ -7,10 +7,10 @@ in-process dict remains the fallback for local/dev/test runs.
|
|
| 7 |
The public names still say ``claude`` because this quota bucket originally
|
| 8 |
only covered Claude and the persisted session field uses that name.
|
| 9 |
|
| 10 |
-
Unit: first premium-model submit
|
| 11 |
-
sends with an allowed premium model in
|
| 12 |
-
|
| 13 |
-
(`AgentSession.
|
| 14 |
``backend.routes.agent``; this module only tracks the per-plan daily cap.
|
| 15 |
|
| 16 |
Cap tiers:
|
|
@@ -40,6 +40,11 @@ def _today() -> str:
|
|
| 40 |
return datetime.now(UTC).date().isoformat()
|
| 41 |
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
def daily_cap_for(plan: str | None) -> int:
|
| 44 |
"""Return the daily Claude-session cap for the given plan."""
|
| 45 |
return CLAUDE_PRO_DAILY if plan == "pro" else CLAUDE_FREE_DAILY
|
|
|
|
| 7 |
The public names still say ``claude`` because this quota bucket originally
|
| 8 |
only covered Claude and the persisted session field uses that name.
|
| 9 |
|
| 10 |
+
Unit: first premium-model submit per session per UTC day, not raw messages. A
|
| 11 |
+
user who sends with an allowed premium model in any session consumes one quota
|
| 12 |
+
point for that day; continuing the same session on the same day doesn't
|
| 13 |
+
(`AgentSession.claude_counted_day` guards that). Model-level plan gates live in
|
| 14 |
``backend.routes.agent``; this module only tracks the per-plan daily cap.
|
| 15 |
|
| 16 |
Cap tiers:
|
|
|
|
| 40 |
return datetime.now(UTC).date().isoformat()
|
| 41 |
|
| 42 |
|
| 43 |
+
def current_quota_day() -> str:
|
| 44 |
+
"""Return the UTC date key used for today's premium-model quota bucket."""
|
| 45 |
+
return _today()
|
| 46 |
+
|
| 47 |
+
|
| 48 |
def daily_cap_for(plan: str | None) -> int:
|
| 49 |
"""Return the daily Claude-session cap for the given plan."""
|
| 50 |
return CLAUDE_PRO_DAILY if plan == "pro" else CLAUDE_FREE_DAILY
|
frontend/src/components/SessionChat.tsx
CHANGED
|
@@ -94,13 +94,13 @@ export default function SessionChat({ sessionId, isActive, onSessionDead }: Sess
|
|
| 94 |
// account (past the subsidized daily allowance). Re-read after each turn,
|
| 95 |
// since the backend flips it at submit time. Only premium-model sessions can
|
| 96 |
// ever be user-billed, so skip the fetch for free models.
|
| 97 |
-
const [premiumBilled, setPremiumBilled] = useState(
|
| 98 |
-
const [premiumQuotaCounted, setPremiumQuotaCounted] = useState(
|
| 99 |
const onPremiumModel = isPremiumPath(sessionMeta?.model ?? undefined);
|
| 100 |
useEffect(() => {
|
| 101 |
if (!isActive || !onPremiumModel) {
|
| 102 |
-
setPremiumBilled(
|
| 103 |
-
setPremiumQuotaCounted(
|
| 104 |
return;
|
| 105 |
}
|
| 106 |
if (busy) return;
|
|
@@ -119,9 +119,9 @@ export default function SessionChat({ sessionId, isActive, onSessionDead }: Sess
|
|
| 119 |
};
|
| 120 |
}, [busy, isActive, onPremiumModel, sessionId]);
|
| 121 |
|
| 122 |
-
const sessionPremiumBilled = premiumBilled
|
| 123 |
const sessionPremiumQuotaCounted =
|
| 124 |
-
premiumQuotaCounted
|
| 125 |
const premiumBillingNotice =
|
| 126 |
sessionPremiumBilled ||
|
| 127 |
(isActive &&
|
|
|
|
| 94 |
// account (past the subsidized daily allowance). Re-read after each turn,
|
| 95 |
// since the backend flips it at submit time. Only premium-model sessions can
|
| 96 |
// ever be user-billed, so skip the fetch for free models.
|
| 97 |
+
const [premiumBilled, setPremiumBilled] = useState<boolean | null>(null);
|
| 98 |
+
const [premiumQuotaCounted, setPremiumQuotaCounted] = useState<boolean | null>(null);
|
| 99 |
const onPremiumModel = isPremiumPath(sessionMeta?.model ?? undefined);
|
| 100 |
useEffect(() => {
|
| 101 |
if (!isActive || !onPremiumModel) {
|
| 102 |
+
setPremiumBilled(null);
|
| 103 |
+
setPremiumQuotaCounted(null);
|
| 104 |
return;
|
| 105 |
}
|
| 106 |
if (busy) return;
|
|
|
|
| 119 |
};
|
| 120 |
}, [busy, isActive, onPremiumModel, sessionId]);
|
| 121 |
|
| 122 |
+
const sessionPremiumBilled = premiumBilled ?? Boolean(sessionMeta?.premiumUserBilled);
|
| 123 |
const sessionPremiumQuotaCounted =
|
| 124 |
+
premiumQuotaCounted ?? Boolean(sessionMeta?.premiumQuotaCounted);
|
| 125 |
const premiumBillingNotice =
|
| 126 |
sessionPremiumBilled ||
|
| 127 |
(isActive &&
|
frontend/src/store/sessionStore.ts
CHANGED
|
@@ -252,10 +252,14 @@ export const useSessionStore = create<SessionStore>()(
|
|
| 252 |
{
|
| 253 |
name: 'hf-agent-sessions',
|
| 254 |
partialize: (state) => ({
|
| 255 |
-
// Reset
|
| 256 |
-
//
|
| 257 |
-
|
| 258 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 259 |
activeSessionId: state.activeSessionId,
|
| 260 |
}),
|
| 261 |
}
|
|
|
|
| 252 |
{
|
| 253 |
name: 'hf-agent-sessions',
|
| 254 |
partialize: (state) => ({
|
| 255 |
+
// Reset transient session flags so cold-load state is re-derived from
|
| 256 |
+
// the live GET /sessions list and current daily quota window.
|
| 257 |
+
sessions: state.sessions.map((s) => ({
|
| 258 |
+
...s,
|
| 259 |
+
isProcessing: false,
|
| 260 |
+
premiumUserBilled: false,
|
| 261 |
+
premiumQuotaCounted: false,
|
| 262 |
+
})),
|
| 263 |
activeSessionId: state.activeSessionId,
|
| 264 |
}),
|
| 265 |
}
|
tests/unit/test_agent_model_gating.py
CHANGED
|
@@ -24,6 +24,7 @@ def _reset_quota_store():
|
|
| 24 |
def _premium_session(model: str = agent.DEFAULT_PREMIUM_MODEL_ID):
|
| 25 |
return SimpleNamespace(
|
| 26 |
claude_counted=False,
|
|
|
|
| 27 |
session=SimpleNamespace(
|
| 28 |
config=SimpleNamespace(model_name=model),
|
| 29 |
premium_user_billed=False,
|
|
@@ -198,11 +199,66 @@ async def test_premium_quota_charges_without_user_billing_inside_allowance(monke
|
|
| 198 |
)
|
| 199 |
|
| 200 |
assert agent_session.claude_counted is True
|
|
|
|
| 201 |
assert agent_session.session.premium_user_billed is False
|
| 202 |
assert persisted == [agent_session]
|
| 203 |
assert await agent.user_quotas.get_claude_used_today("u1") == 1
|
| 204 |
|
| 205 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
@pytest.mark.asyncio
|
| 207 |
async def test_free_user_gets_two_subsidized_premium_sessions_then_user_billing(
|
| 208 |
monkeypatch,
|
|
@@ -227,6 +283,7 @@ async def test_free_user_gets_two_subsidized_premium_sessions_then_user_billing(
|
|
| 227 |
third = _premium_session()
|
| 228 |
await agent._enforce_premium_model_quota({"user_id": "g1", "plan": "free"}, third)
|
| 229 |
assert third.session.premium_user_billed is True
|
|
|
|
| 230 |
assert await agent.user_quotas.get_claude_used_today("g1") == 2
|
| 231 |
|
| 232 |
|
|
@@ -249,6 +306,7 @@ async def test_free_model_does_not_consume_premium_quota(monkeypatch):
|
|
| 249 |
)
|
| 250 |
|
| 251 |
assert agent_session.claude_counted is False
|
|
|
|
| 252 |
assert await agent.user_quotas.get_claude_used_today("u1") == 0
|
| 253 |
|
| 254 |
|
|
@@ -274,6 +332,7 @@ async def test_free_user_cannot_spend_quota_on_pro_only_premium_model(monkeypatc
|
|
| 274 |
assert exc_info.value.status_code == 403
|
| 275 |
assert exc_info.value.detail["error"] == "model_requires_pro"
|
| 276 |
assert agent_session.claude_counted is False
|
|
|
|
| 277 |
assert await agent.user_quotas.get_claude_used_today("u1") == 0
|
| 278 |
|
| 279 |
|
|
@@ -321,6 +380,7 @@ async def test_pro_user_uses_pro_premium_quota(monkeypatch):
|
|
| 321 |
agent_session,
|
| 322 |
)
|
| 323 |
assert agent_session.claude_counted is True
|
|
|
|
| 324 |
assert agent_session.session.premium_user_billed is False
|
| 325 |
assert await agent.user_quotas.get_claude_used_today("pro-user") == index + 1
|
| 326 |
|
|
|
|
| 24 |
def _premium_session(model: str = agent.DEFAULT_PREMIUM_MODEL_ID):
|
| 25 |
return SimpleNamespace(
|
| 26 |
claude_counted=False,
|
| 27 |
+
claude_counted_day=None,
|
| 28 |
session=SimpleNamespace(
|
| 29 |
config=SimpleNamespace(model_name=model),
|
| 30 |
premium_user_billed=False,
|
|
|
|
| 199 |
)
|
| 200 |
|
| 201 |
assert agent_session.claude_counted is True
|
| 202 |
+
assert agent_session.claude_counted_day == agent.user_quotas.current_quota_day()
|
| 203 |
assert agent_session.session.premium_user_billed is False
|
| 204 |
assert persisted == [agent_session]
|
| 205 |
assert await agent.user_quotas.get_claude_used_today("u1") == 1
|
| 206 |
|
| 207 |
|
| 208 |
+
@pytest.mark.asyncio
|
| 209 |
+
async def test_premium_quota_counts_same_session_once_per_day(monkeypatch):
|
| 210 |
+
async def fake_persist_session_snapshot(_agent_session):
|
| 211 |
+
return None
|
| 212 |
+
|
| 213 |
+
monkeypatch.setattr(
|
| 214 |
+
agent.session_manager,
|
| 215 |
+
"persist_session_snapshot",
|
| 216 |
+
fake_persist_session_snapshot,
|
| 217 |
+
)
|
| 218 |
+
|
| 219 |
+
agent_session = _premium_session()
|
| 220 |
+
|
| 221 |
+
await agent._enforce_premium_model_quota(
|
| 222 |
+
{"user_id": "u1", "plan": "free"},
|
| 223 |
+
agent_session,
|
| 224 |
+
)
|
| 225 |
+
await agent._enforce_premium_model_quota(
|
| 226 |
+
{"user_id": "u1", "plan": "free"},
|
| 227 |
+
agent_session,
|
| 228 |
+
)
|
| 229 |
+
|
| 230 |
+
assert agent_session.claude_counted is True
|
| 231 |
+
assert agent_session.claude_counted_day == agent.user_quotas.current_quota_day()
|
| 232 |
+
assert await agent.user_quotas.get_claude_used_today("u1") == 1
|
| 233 |
+
|
| 234 |
+
|
| 235 |
+
@pytest.mark.asyncio
|
| 236 |
+
async def test_premium_quota_counts_stale_session_again_today(monkeypatch):
|
| 237 |
+
async def fake_persist_session_snapshot(_agent_session):
|
| 238 |
+
return None
|
| 239 |
+
|
| 240 |
+
monkeypatch.setattr(
|
| 241 |
+
agent.session_manager,
|
| 242 |
+
"persist_session_snapshot",
|
| 243 |
+
fake_persist_session_snapshot,
|
| 244 |
+
)
|
| 245 |
+
|
| 246 |
+
agent_session = _premium_session()
|
| 247 |
+
agent_session.claude_counted = True
|
| 248 |
+
agent_session.claude_counted_day = "2000-01-01"
|
| 249 |
+
agent_session.session.premium_user_billed = True
|
| 250 |
+
|
| 251 |
+
await agent._enforce_premium_model_quota(
|
| 252 |
+
{"user_id": "u1", "plan": "free"},
|
| 253 |
+
agent_session,
|
| 254 |
+
)
|
| 255 |
+
|
| 256 |
+
assert agent_session.claude_counted is True
|
| 257 |
+
assert agent_session.claude_counted_day == agent.user_quotas.current_quota_day()
|
| 258 |
+
assert agent_session.session.premium_user_billed is False
|
| 259 |
+
assert await agent.user_quotas.get_claude_used_today("u1") == 1
|
| 260 |
+
|
| 261 |
+
|
| 262 |
@pytest.mark.asyncio
|
| 263 |
async def test_free_user_gets_two_subsidized_premium_sessions_then_user_billing(
|
| 264 |
monkeypatch,
|
|
|
|
| 283 |
third = _premium_session()
|
| 284 |
await agent._enforce_premium_model_quota({"user_id": "g1", "plan": "free"}, third)
|
| 285 |
assert third.session.premium_user_billed is True
|
| 286 |
+
assert third.claude_counted_day == agent.user_quotas.current_quota_day()
|
| 287 |
assert await agent.user_quotas.get_claude_used_today("g1") == 2
|
| 288 |
|
| 289 |
|
|
|
|
| 306 |
)
|
| 307 |
|
| 308 |
assert agent_session.claude_counted is False
|
| 309 |
+
assert agent_session.claude_counted_day is None
|
| 310 |
assert await agent.user_quotas.get_claude_used_today("u1") == 0
|
| 311 |
|
| 312 |
|
|
|
|
| 332 |
assert exc_info.value.status_code == 403
|
| 333 |
assert exc_info.value.detail["error"] == "model_requires_pro"
|
| 334 |
assert agent_session.claude_counted is False
|
| 335 |
+
assert agent_session.claude_counted_day is None
|
| 336 |
assert await agent.user_quotas.get_claude_used_today("u1") == 0
|
| 337 |
|
| 338 |
|
|
|
|
| 380 |
agent_session,
|
| 381 |
)
|
| 382 |
assert agent_session.claude_counted is True
|
| 383 |
+
assert agent_session.claude_counted_day == agent.user_quotas.current_quota_day()
|
| 384 |
assert agent_session.session.premium_user_billed is False
|
| 385 |
assert await agent.user_quotas.get_claude_used_today("pro-user") == index + 1
|
| 386 |
|
tests/unit/test_session_manager_persistence.py
CHANGED
|
@@ -757,6 +757,7 @@ async def test_list_sessions_dev_uses_store_dev_visibility():
|
|
| 757 |
"created_at": datetime.now(UTC),
|
| 758 |
"premium_user_billed": True,
|
| 759 |
"claude_counted": True,
|
|
|
|
| 760 |
"auto_approval_enabled": True,
|
| 761 |
"auto_approval_cost_cap_usd": 5.0,
|
| 762 |
"auto_approval_estimated_spend_usd": 2.0,
|
|
@@ -786,3 +787,44 @@ async def test_list_sessions_dev_uses_store_dev_visibility():
|
|
| 786 |
"estimated_spend_usd": 2.0,
|
| 787 |
"remaining_usd": 3.0,
|
| 788 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 757 |
"created_at": datetime.now(UTC),
|
| 758 |
"premium_user_billed": True,
|
| 759 |
"claude_counted": True,
|
| 760 |
+
"claude_counted_day": datetime.now(UTC).date().isoformat(),
|
| 761 |
"auto_approval_enabled": True,
|
| 762 |
"auto_approval_cost_cap_usd": 5.0,
|
| 763 |
"auto_approval_estimated_spend_usd": 2.0,
|
|
|
|
| 787 |
"estimated_spend_usd": 2.0,
|
| 788 |
"remaining_usd": 3.0,
|
| 789 |
}
|
| 790 |
+
|
| 791 |
+
|
| 792 |
+
def test_get_session_info_marks_stale_premium_quota_as_unused_today():
|
| 793 |
+
manager = _manager_with_store(NoopSessionStore())
|
| 794 |
+
agent_session = _runtime_agent_session("s1", user_id="alice")
|
| 795 |
+
agent_session.claude_counted = True
|
| 796 |
+
agent_session.claude_counted_day = "2000-01-01"
|
| 797 |
+
agent_session.session.premium_user_billed = True
|
| 798 |
+
manager.sessions["s1"] = agent_session
|
| 799 |
+
|
| 800 |
+
info = manager.get_session_info("s1")
|
| 801 |
+
|
| 802 |
+
assert info is not None
|
| 803 |
+
assert info["premium_user_billed"] is False
|
| 804 |
+
assert info["premium_quota_counted"] is False
|
| 805 |
+
|
| 806 |
+
|
| 807 |
+
@pytest.mark.asyncio
|
| 808 |
+
async def test_list_sessions_marks_stale_premium_quota_as_unused_today():
|
| 809 |
+
class ListStore(NoopSessionStore):
|
| 810 |
+
enabled = True
|
| 811 |
+
|
| 812 |
+
async def list_sessions(self, user_id: str, **_: Any) -> list[dict[str, Any]]:
|
| 813 |
+
return [
|
| 814 |
+
{
|
| 815 |
+
"session_id": "s1",
|
| 816 |
+
"user_id": user_id,
|
| 817 |
+
"model": "m",
|
| 818 |
+
"created_at": datetime.now(UTC),
|
| 819 |
+
"premium_user_billed": True,
|
| 820 |
+
"claude_counted": True,
|
| 821 |
+
"claude_counted_day": "2000-01-01",
|
| 822 |
+
}
|
| 823 |
+
]
|
| 824 |
+
|
| 825 |
+
manager = _manager_with_store(ListStore())
|
| 826 |
+
|
| 827 |
+
sessions = await manager.list_sessions(user_id="alice")
|
| 828 |
+
|
| 829 |
+
assert sessions[0]["premium_user_billed"] is False
|
| 830 |
+
assert sessions[0]["premium_quota_counted"] is False
|