Support daily/weekly automatic quota reset using a rolling period
anchored to first usage. Implements lazy reset via atomic CTE SQL
in IncrementQuotaUsed - no cron job needed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove frontend badge/tooltip and backend counting/query infrastructure
for client affinity. Core client affinity scheduling logic is preserved.
Removed:
- GatewayCache injection in AccountHandler
- GetAccountAffinityCount/Batch/Clients methods and interface
- AffinityClientInfo struct and ClientAffinityTTL function
- /affinity-clients API endpoint and route
- Reverse index (account_affinity:*) Lua scripts
- Frontend blue badge, tooltip, and lazy-load logic
- affinity_client_count field from types and response
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
PR #723 changed all OAuth 401 to use SetTempUnschedulable instead of SetError,
and PR #761 added 401 escalation logic in tryTempUnschedulable. Both caused
non-OpenAI accounts to frequently become temporarily unschedulable.
- HandleUpstreamError case 401: limit temp-unschedulable to OpenAI OAuth only,
other platforms (Anthropic/Gemini/Antigravity) revert to original SetError
- tryTempUnschedulable: skip 401 escalation for Antigravity, let its
applyErrorPolicy temp_unschedulable_rules handle 401 naturally
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Backend: expand IsClientAffinityEnabled() to support all Anthropic
accounts (OAuth/SetupToken/APIKey), not just OAuth/SetupToken
- Frontend: show client affinity toggle for all Anthropic accounts
- Frontend: fix save logic to properly chain extra field updates
- Antigravity accounts explicitly excluded from client affinity
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add reverse Redis index (account_affinity:{accountID}) to track which
clients are bound to each account via affinity scheduling. Display a
blue badge with client count in the admin account list, with a hover
tooltip showing client details (groupID:clientID + TTL).
Backend: reverse index in Lua scripts, batch query methods, new
GET /accounts/:id/affinity-clients endpoint.
Frontend: blue badge + lazy-loaded tooltip in AccountStatusIndicator.
- Fix bulk edit: send 0 instead of null/NaN to clear load_factor
- Fix edit modal: explicit NaN check instead of implicit falsy
- Fix create modal: use ?? instead of || for load_factor
- Add load_factor upper limit validation (max 10000)
- Add //go:build unit tag and self-contained intPtrHelper in test
- Add design intent comments on WaitPlan.MaxConcurrency
- Fix bulk edit: send 0 instead of null/NaN to clear load_factor
- Fix edit modal: explicit NaN check instead of implicit falsy
- Fix create modal: use ?? instead of || for load_factor
- Add load_factor upper limit validation (max 10000)
- Add //go:build unit tag and self-contained intPtrHelper in test
- Add design intent comments on WaitPlan.MaxConcurrency
- Separate load factor from concurrency: concurrency controls actual
slot acquisition, load_factor controls load rate calculation
- Add EffectiveLoadFactor() method: LoadFactor > Concurrency > 1
- Add load_factor field to Create/Edit/BulkEdit account forms
- Fix RPM default value: auto-fill 15 when RPM enabled but not set
- Fix stale test compilation errors in server and handler packages
- Extract postUsageBilling() to consolidate billing logic across
GatewayService.RecordUsage, RecordUsageWithLongContext, and
OpenAIGatewayService.RecordUsage, eliminating ~120 lines of
duplicated code
- Fix account quota to use TotalCost × accountRateMultiplier
(was using raw TotalCost, inconsistent with account cost stats)
- Fix RecordUsageWithLongContext API Key quota only updating in
balance mode (now updates regardless of billing type)
- Fix WebSocket client disconnect detection on Windows by adding
"an established connection was aborted" to known disconnect errors
- Extract postUsageBilling() to consolidate billing logic across
GatewayService.RecordUsage, RecordUsageWithLongContext, and
OpenAIGatewayService.RecordUsage, eliminating ~120 lines of
duplicated code
- Fix account quota to use TotalCost × accountRateMultiplier
(was using raw TotalCost, inconsistent with account cost stats)
- Fix RecordUsageWithLongContext API Key quota only updating in
balance mode (now updates regardless of billing type)
- Fix WebSocket client disconnect detection on Windows by adding
"an established connection was aborted" to known disconnect errors
Antigravity APIKey accounts were incorrectly routed to
testAntigravityAccountConnection which calls AntigravityTokenProvider,
but the token provider only handles OAuth and Upstream types, causing
"not an antigravity oauth account" error.
Extract routeAntigravityTest to route APIKey accounts to native
Claude/Gemini test paths based on model prefix, matching the
gateway_handler routing logic for normal requests.
Antigravity APIKey accounts were incorrectly routed to
testAntigravityAccountConnection which calls AntigravityTokenProvider,
but the token provider only handles OAuth and Upstream types, causing
"not an antigravity oauth account" error.
Extract routeAntigravityTest to route APIKey accounts to native
Claude/Gemini test paths based on model prefix, matching the
gateway_handler routing logic for normal requests.
The existing regex only matched the old format where account_uuid is
empty (account__session_). Real Claude Code clients and newer sub2api
generated user_ids use account_{uuid}_session_ which was silently
skipped, causing the original metadata.user_id to leak to upstream
when User-Agent is rewritten by an intermediate gateway.
Closes#766
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The upstream v0.1.90 changed GetAccountConcurrencyBatch from individual
Lua script calls (which swallowed per-account errors) to a Redis pipeline
approach that propagates errors from rdb.Time() or pipe.Exec(). When the
HTTP request context is cancelled (e.g., browser abort), the entire batch
fails and the handler silently shows all concurrency as 0.
Fix: use context.WithTimeout(context.Background(), 3s) for the Redis call
so HTTP request cancellation doesn't affect the read-only concurrency query.