Files
sub2api/deploy/config.example.yaml

995 lines
42 KiB
YAML
Raw Normal View History

2025-12-18 13:50:39 +08:00
# Sub2API Configuration File
# Sub2API 配置文件
#
2025-12-18 13:50:39 +08:00
# Copy this file to /etc/sub2api/config.yaml and modify as needed
# 复制此文件到 /etc/sub2api/config.yaml 并根据需要修改
#
# Documentation / 文档: https://github.com/Wei-Shaw/sub2api
2025-12-18 13:50:39 +08:00
# =============================================================================
# Server Configuration
# 服务器配置
2025-12-18 13:50:39 +08:00
# =============================================================================
server:
# Bind address (0.0.0.0 for all interfaces)
# 绑定地址0.0.0.0 表示监听所有网络接口)
2025-12-18 13:50:39 +08:00
host: "0.0.0.0"
# Port to listen on
# 监听端口
2025-12-18 13:50:39 +08:00
port: 8080
# Mode: "debug" for development, "release" for production
# 运行模式:"debug" 用于开发,"release" 用于生产环境
2025-12-18 13:50:39 +08:00
mode: "release"
# Frontend base URL used to generate external links in emails (e.g. password reset)
# 用于生成邮件中的外部链接(例如:重置密码链接)的前端基础地址
# Example: "https://example.com"
frontend_url: ""
# Trusted proxies for X-Forwarded-For parsing (CIDR/IP). Empty disables trusted proxies.
# 信任的代理地址CIDR/IP 格式),用于解析 X-Forwarded-For 头。留空则禁用代理信任。
trusted_proxies: []
# Global max request body size in bytes (default: 256MB)
# 全局最大请求体大小(字节,默认 256MB
# Applies to all requests, especially important for h2c first request memory protection
# 适用于所有请求,对 h2c 第一请求的内存保护尤为重要
max_request_body_size: 268435456
# HTTP/2 Cleartext (h2c) configuration
# HTTP/2 Cleartext (h2c) 配置
h2c:
# Enable HTTP/2 Cleartext for client connections
# 启用 HTTP/2 Cleartext 客户端连接
enabled: true
# Max concurrent streams per connection
# 每个连接的最大并发流数量
max_concurrent_streams: 50
# Idle timeout for connections (seconds)
# 连接空闲超时时间(秒)
idle_timeout: 75
# Max frame size in bytes (default: 1MB)
# 最大帧大小(字节,默认 1MB
max_read_frame_size: 1048576
# Max upload buffer per connection in bytes (default: 2MB)
# 每个连接的最大上传缓冲区(字节,默认 2MB
max_upload_buffer_per_connection: 2097152
# Max upload buffer per stream in bytes (default: 512KB)
# 每个流的最大上传缓冲区(字节,默认 512KB
max_upload_buffer_per_stream: 524288
2025-12-18 13:50:39 +08:00
# =============================================================================
# Run Mode Configuration
# 运行模式配置
# =============================================================================
# Run mode: "standard" (default) or "simple" (for internal use)
# 运行模式:"standard"(默认)或 "simple"(内部使用)
# - standard: Full SaaS features with billing/balance checks
# - standard: 完整 SaaS 功能,包含计费和余额校验
# - simple: Hides SaaS features and skips billing/balance checks
# - simple: 隐藏 SaaS 功能,跳过计费和余额校验
run_mode: "standard"
# =============================================================================
# CORS Configuration
# 跨域资源共享 (CORS) 配置
# =============================================================================
cors:
# Allowed origins list. Leave empty to disable cross-origin requests.
# 允许的来源列表。留空则禁用跨域请求。
allowed_origins: []
# Allow credentials (cookies/authorization headers). Cannot be used with "*".
# 允许携带凭证cookies/授权头)。不能与 "*" 通配符同时使用。
allow_credentials: true
# =============================================================================
# Security Configuration
# 安全配置
# =============================================================================
security:
url_allowlist:
# Enable URL allowlist validation (disable to skip all URL checks)
# 启用 URL 白名单验证(禁用则跳过所有 URL 检查)
enabled: false
# Allowed upstream hosts for API proxying
# 允许代理的上游 API 主机列表
upstream_hosts:
- "api.openai.com"
- "api.anthropic.com"
- "api.kimi.com"
- "open.bigmodel.cn"
- "api.minimaxi.com"
- "generativelanguage.googleapis.com"
- "cloudcode-pa.googleapis.com"
- "*.openai.azure.com"
# Allowed hosts for pricing data download
# 允许下载定价数据的主机列表
pricing_hosts:
- "raw.githubusercontent.com"
# Allowed hosts for CRS sync (required when using CRS sync)
# 允许 CRS 同步的主机列表(使用 CRS 同步功能时必须配置)
crs_hosts: []
# Allow localhost/private IPs for upstream/pricing/CRS (use only in trusted networks)
# 允许本地/私有 IP 地址用于上游/定价/CRS仅在可信网络中使用
allow_private_hosts: true
# Allow http:// URLs when allowlist is disabled (default: false, require https)
# 白名单禁用时是否允许 http:// URL默认: false要求 https
allow_insecure_http: true
response_headers:
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
# Enable configurable response header filtering (default: true)
# 启用可配置的响应头过滤(默认启用,过滤上游敏感响应头)
enabled: true
# Extra allowed response headers from upstream
# 额外允许的上游响应头
additional_allowed: []
# Force-remove response headers from upstream
# 强制移除的上游响应头
force_remove: []
csp:
# Enable Content-Security-Policy header
# 启用内容安全策略 (CSP) 响应头
enabled: true
# Default CSP policy (override if you host assets on other domains)
# 默认 CSP 策略(如果静态资源托管在其他域名,请自行覆盖)
# Note: __CSP_NONCE__ will be replaced with 'nonce-xxx' at request time for inline script security
# 注意__CSP_NONCE__ 会在请求时被替换为 'nonce-xxx',用于内联脚本安全
policy: "default-src 'self'; script-src 'self' __CSP_NONCE__ https://challenges.cloudflare.com https://static.cloudflareinsights.com; style-src 'self' 'unsafe-inline' https://fonts.googleapis.com; img-src 'self' data: https:; font-src 'self' data: https://fonts.gstatic.com; connect-src 'self' https:; frame-src https://challenges.cloudflare.com; frame-ancestors 'none'; base-uri 'self'; form-action 'self'"
proxy_probe:
# Allow skipping TLS verification for proxy probe (debug only)
# 允许代理探测时跳过 TLS 证书验证(仅用于调试)
insecure_skip_verify: false
proxy_fallback:
# Allow auxiliary services (update check, pricing data) to fallback to direct
# connection when proxy initialization fails. Does NOT affect AI gateway connections.
# 辅助服务(更新检查、定价数据拉取)代理初始化失败时是否允许回退直连。
# 不影响 AI 账号网关连接。默认 falsefail-fast 防止 IP 泄露。
allow_direct_on_error: false
# =============================================================================
# Gateway Configuration
# 网关配置
# =============================================================================
gateway:
# Timeout for waiting upstream response headers (seconds)
# 等待上游响应头超时时间(秒)
response_header_timeout: 600
# Max request body size in bytes (default: 256MB)
# 请求体最大字节数(默认 256MB
max_body_size: 268435456
# Max bytes to read for non-stream upstream responses (default: 8MB)
# 非流式上游响应体读取上限(默认 8MB
upstream_response_read_max_bytes: 8388608
# Max bytes to read for proxy probe responses (default: 1MB)
# 代理探测响应体读取上限(默认 1MB
proxy_probe_response_read_max_bytes: 1048576
# Enable Gemini upstream response header debug logs (default: false)
# 是否开启 Gemini 上游响应头调试日志(默认 false
gemini_debug_response_headers: false
# Sora max request body size in bytes (0=use max_body_size)
# Sora 请求体最大字节数0=使用 max_body_size
sora_max_body_size: 268435456
# Sora stream timeout (seconds, 0=disable)
# Sora 流式请求总超时0=禁用)
sora_stream_timeout_seconds: 900
# Sora non-stream timeout (seconds, 0=disable)
# Sora 非流式请求超时0=禁用)
sora_request_timeout_seconds: 180
# Sora stream enforcement mode: force/error
# Sora stream 强制策略force/error
sora_stream_mode: "force"
# Sora model filters
# Sora 模型过滤配置
sora_model_filters:
# Hide prompt-enhance models by default
# 默认隐藏 prompt-enhance 模型
hide_prompt_enhance: true
# Require API key for /sora/media proxy (default: false)
# /sora/media 是否强制要求 API Key默认 true
sora_media_require_api_key: true
# Sora media temporary signing key (empty disables signed URL)
# Sora 媒体临时签名密钥(为空则禁用签名)
sora_media_signing_key: ""
# Signed URL TTL seconds (<=0 disables)
# 临时签名 URL 有效期(秒,<=0 表示禁用)
sora_media_signed_url_ttl_seconds: 900
# Connection pool isolation strategy:
# 连接池隔离策略:
# - proxy: Isolate by proxy, same proxy shares connection pool (suitable for few proxies, many accounts)
# - proxy: 按代理隔离,同一代理共享连接池(适合代理少、账户多)
# - account: Isolate by account, same account shares connection pool (suitable for few accounts, strict isolation)
# - account: 按账户隔离,同一账户共享连接池(适合账户少、需严格隔离)
# - account_proxy: Isolate by account+proxy combination (default, finest granularity)
# - account_proxy: 按账户+代理组合隔离(默认,最细粒度)
connection_pool_isolation: "account_proxy"
# Force Codex CLI mode: treat all /openai/v1/responses requests as Codex CLI.
# 强制按 Codex CLI 处理 /openai/v1/responses 请求(用于网关未透传/改写 User-Agent 的兜底)。
#
# 注意:开启后会影响所有客户端的行为(不仅限于 VS Code / Codex CLI请谨慎开启。
force_codex_cli: false
# OpenAI 透传模式是否放行客户端超时头(如 x-stainless-timeout
# 默认 false过滤超时头降低上游提前断流风险。
openai_passthrough_allow_timeout_headers: false
# OpenAI Responses WebSocket 配置(默认开启,可按需回滚到 HTTP
openai_ws:
# 新版 WS mode 路由(默认关闭)。关闭时保持当前 legacy 实现行为。
mode_router_v2_enabled: false
# ingress 默认模式off|shared|dedicated仅 mode_router_v2_enabled=true 生效)
ingress_mode_default: shared
# 全局总开关,默认 true关闭时所有请求保持原有 HTTP/SSE 路由
enabled: true
# 按账号类型细分开关
oauth_enabled: true
apikey_enabled: true
# 全局强制 HTTP紧急回滚开关
force_http: false
# 允许在 WSv2 下按策略恢复 store=true默认 false
allow_store_recovery: false
# ingress 模式收到 previous_response_not_found 时,自动去掉 previous_response_id 重试一次(默认 true
ingress_previous_response_recovery_enabled: true
# store=false 且无可复用会话连接时的策略:
# strict=强制新建连接隔离优先adaptive=仅在高风险失败后强制新建off=尽量复用(性能优先)
store_disabled_conn_mode: strict
# store=false 且无可复用会话连接时,是否强制新建连接(默认 true优先会话隔离
# 兼容旧配置:仅在 store_disabled_conn_mode 未配置时生效
store_disabled_force_new_conn: true
# 是否启用 WSv2 generate=false 预热(默认 false
prewarm_generate_enabled: false
# 协议 feature 开关v2 优先于 v1
responses_websockets: false
responses_websockets_v2: true
# 连接池参数(按账号池化复用)
max_conns_per_account: 128
min_idle_per_account: 4
max_idle_per_account: 12
# 是否按账号并发动态计算连接池上限:
# effective_max_conns = min(max_conns_per_account, ceil(account.concurrency * factor))
dynamic_max_conns_by_account_concurrency_enabled: true
# 按账号类型分别设置系数OAuth / API Key
oauth_max_conns_factor: 1.0
apikey_max_conns_factor: 1.0
dial_timeout_seconds: 10
read_timeout_seconds: 900
write_timeout_seconds: 120
pool_target_utilization: 0.7
queue_limit_per_conn: 64
# 流式写出批量 flush 参数
event_flush_batch_size: 1
event_flush_interval_ms: 10
# 预热触发冷却(毫秒)
prewarm_cooldown_ms: 300
# WS 回退到 HTTP 后的冷却时间(秒),用于避免 WS/HTTP 来回抖动0 表示关闭冷却
fallback_cooldown_seconds: 30
# WS 重试退避参数(毫秒)
retry_backoff_initial_ms: 120
retry_backoff_max_ms: 2000
# 抖动比例0-1
retry_jitter_ratio: 0.2
# 单次请求 WS 重试总预算(毫秒);建议设置为有限值,避免重试拉高 TTFT 长尾
retry_total_budget_ms: 5000
# payload_schema 日志采样率0-1降低热路径日志放大
payload_log_sample_rate: 0.2
# 调度与粘连参数
lb_top_k: 7
sticky_session_ttl_seconds: 3600
# 会话哈希迁移兼容开关:新 key 未命中时回退读取旧 SHA-256 key
session_hash_read_old_fallback: true
# 会话哈希迁移兼容开关:写入时双写旧 SHA-256 key短 TTL
session_hash_dual_write_old: true
# context 元数据迁移兼容开关:保留旧 ctxkey.* 读取/注入桥接
metadata_bridge_enabled: true
sticky_response_id_ttl_seconds: 3600
# 兼容旧键:当 sticky_response_id_ttl_seconds 缺失时回退该值
sticky_previous_response_ttl_seconds: 3600
scheduler_score_weights:
priority: 1.0
load: 1.0
queue: 0.7
error_rate: 0.8
ttft: 0.5
# HTTP upstream connection pool settings (HTTP/2 + multi-proxy scenario defaults)
# HTTP 上游连接池配置HTTP/2 + 多代理场景默认值)
# Max idle connections across all hosts
# 所有主机的最大空闲连接数
max_idle_conns: 2560
# Max idle connections per host
# 每个主机的最大空闲连接数
max_idle_conns_per_host: 120
# Max connections per host
# 每个主机的最大连接数
max_conns_per_host: 1024
# Idle connection timeout (seconds)
# 空闲连接超时时间(秒)
idle_conn_timeout_seconds: 90
# Upstream client cache settings
# 上游连接池客户端缓存配置
# max_upstream_clients: Max cached clients, evicts least recently used when exceeded
# max_upstream_clients: 最大缓存客户端数量,超出后淘汰最久未使用的
max_upstream_clients: 5000
# client_idle_ttl_seconds: Client idle reclaim threshold (seconds), reclaimed when idle and no active requests
# client_idle_ttl_seconds: 客户端空闲回收阈值(秒),超时且无活跃请求时回收
client_idle_ttl_seconds: 900
# Concurrency slot expiration time (minutes)
# 并发槽位过期时间(分钟)
concurrency_slot_ttl_minutes: 30
# Stream data interval timeout (seconds), 0=disable
# 流数据间隔超时0=禁用
stream_data_interval_timeout: 180
# Stream keepalive interval (seconds), 0=disable
# 流式 keepalive 间隔0=禁用
stream_keepalive_interval: 10
# SSE max line size in bytes (default: 40MB)
# SSE 单行最大字节数(默认 40MB
max_line_size: 41943040
# Log upstream error response body summary (safe/truncated; does not log request content)
# 记录上游错误响应体摘要(安全/截断;不记录请求内容)
log_upstream_error_body: true
# Max bytes to log from upstream error body
# 记录上游错误响应体的最大字节数
log_upstream_error_body_max_bytes: 2048
# Auto inject anthropic-beta header for API-key accounts when needed (default: off)
# 需要时自动为 API-key 账户注入 anthropic-beta 头(默认:关闭)
inject_beta_for_apikey: false
# Allow failover on selected 400 errors (default: off)
# 允许在特定 400 错误时进行故障转移(默认:关闭)
failover_on_400: false
# Scheduling configuration
# 调度配置
scheduling:
# Sticky session max waiting queue size
# 粘性会话最大排队长度
sticky_session_max_waiting: 3
# Sticky session wait timeout (duration)
# 粘性会话等待超时(时间段)
sticky_session_wait_timeout: 120s
# Fallback wait timeout (duration)
# 兜底排队等待超时(时间段)
fallback_wait_timeout: 30s
# Fallback max waiting queue size
# 兜底最大排队长度
fallback_max_waiting: 100
# Enable batch load calculation for scheduling
# 启用调度批量负载计算
load_batch_enabled: true
# Slot cleanup interval (duration)
# 并发槽位清理周期(时间段)
slot_cleanup_interval: 30s
# 是否允许受控回源到 DB默认 true保持现有行为
db_fallback_enabled: true
# 受控回源超时0 表示不额外收紧超时
db_fallback_timeout_seconds: 0
# 受控回源限流(实例级 QPS0 表示不限制
db_fallback_max_qps: 0
# outbox 轮询周期(秒)
outbox_poll_interval_seconds: 1
# outbox 滞后告警阈值(秒)
outbox_lag_warn_seconds: 5
# outbox 触发强制重建阈值(秒)
outbox_lag_rebuild_seconds: 10
# outbox 连续滞后触发次数
outbox_lag_rebuild_failures: 3
# outbox 积压触发重建阈值(行数)
outbox_backlog_rebuild_rows: 10000
# 全量重建周期0 表示禁用
full_rebuild_interval_seconds: 300
# TLS fingerprint simulation / TLS 指纹伪装
# Default profile "claude_cli_v2" simulates Node.js 20.x
# 默认模板 "claude_cli_v2" 模拟 Node.js 20.x 指纹
tls_fingerprint:
enabled: true
# profiles:
# profile_1:
# name: "Custom Profile 1"
# profile_2:
# name: "Custom Profile 2"
# =============================================================================
# Logging Configuration
# 日志配置
# =============================================================================
log:
# Log level: debug/info/warn/error
# 日志级别debug/info/warn/error
level: "info"
# Log format: json/console
# 日志格式json/console
format: "console"
# Service name field written into each log line
# 每条日志都会附带 service 字段
service_name: "sub2api"
# Environment field written into each log line
# 每条日志都会附带 env 字段
env: "production"
# Include caller information
# 是否输出调用方位置信息
caller: true
# Stacktrace threshold: none/error/fatal
# 堆栈输出阈值none/error/fatal
stacktrace_level: "error"
output:
# Keep stdout/stderr output for container log collection
# 保持标准输出用于容器日志采集
to_stdout: true
# Enable file output (default path auto-derived)
# 启用文件输出(默认路径自动推导)
to_file: true
# Empty means:
# - DATA_DIR set: {{DATA_DIR}}/logs/sub2api.log
# - otherwise: /app/data/logs/sub2api.log
# 留空时:
# - 设置 DATA_DIR{{DATA_DIR}}/logs/sub2api.log
# - 否则:/app/data/logs/sub2api.log
file_path: ""
rotation:
# Max file size before rotation (MB)
# 单文件滚动阈值MB
max_size_mb: 100
# Number of rotated files to keep (0 means unlimited)
# 保留历史文件数量0 表示不限制)
max_backups: 10
# Number of days to keep old log files (0 means unlimited)
# 历史日志保留天数0 表示不限制)
max_age_days: 7
# Compress rotated files
# 是否压缩历史日志
compress: true
# Use local time for timestamp in rotated filename
# 滚动文件名时间戳使用本地时区
local_time: true
sampling:
# Enable zap sampler (reduce high-frequency repetitive logs)
# 启用 zap 采样(减少高频重复日志)
enabled: false
# Number of first entries per second to always log
# 每秒无采样保留的前 N 条日志
initial: 100
# Thereafter keep 1 out of N entries per second
# 之后每 N 条保留 1 条
thereafter: 100
# =============================================================================
# Sora Direct Client Configuration
# Sora 直连配置
# =============================================================================
sora:
client:
# Sora backend base URL
# Sora 上游 Base URL
base_url: "https://sora.chatgpt.com/backend"
# Request timeout (seconds)
# 请求超时(秒)
timeout_seconds: 120
# Max retries for upstream requests
# 上游请求最大重试次数
max_retries: 3
# Account+proxy cooldown window after Cloudflare challenge (seconds, 0 to disable)
# Cloudflare challenge 后按账号+代理冷却窗口0 表示关闭)
cloudflare_challenge_cooldown_seconds: 900
# Poll interval (seconds)
# 轮询间隔(秒)
poll_interval_seconds: 2
# Max poll attempts
# 最大轮询次数
max_poll_attempts: 600
# Recent task query limit (image)
# 最近任务查询数量(图片轮询)
recent_task_limit: 50
# Recent task query max limit (fallback)
# 最近任务查询最大数量(回退)
recent_task_limit_max: 200
# Enable debug logs for Sora upstream requests
# 启用 Sora 直连调试日志
# 调试日志会输出上游请求尝试、重试、响应摘要Authorization/openai-sentinel-token 等敏感头会自动脱敏
debug: false
# Allow Sora client to fetch token via OpenAI token provider
# 是否允许 Sora 客户端通过 OpenAI token provider 取 token默认 false避免误走 OpenAI 刷新链路)
use_openai_token_provider: false
# Optional custom headers (key-value)
# 额外请求头(键值对)
headers: {}
# Default User-Agent for Sora requests
# Sora 默认 User-Agent
user_agent: "Sora/1.2026.007 (Android 15; 24122RKC7C; build 2600700)"
# Disable TLS fingerprint for Sora upstream
# 关闭 Sora 上游 TLS 指纹伪装
disable_tls_fingerprint: false
# curl_cffi sidecar for Sora only (required)
# 仅 Sora 链路使用的 curl_cffi sidecar必需
curl_cffi_sidecar:
# Sora 强制通过 sidecar 请求,必须启用
# Sora is forced to use sidecar only; keep enabled=true
enabled: true
# Sidecar base URL (default endpoint: /request)
# sidecar 基础地址(默认请求端点:/request
base_url: "http://sora-curl-cffi-sidecar:8080"
# curl_cffi impersonate profile, e.g. chrome131/chrome124/safari18_0
# curl_cffi 指纹伪装 profile例如 chrome131/chrome124/safari18_0
impersonate: "chrome131"
# Sidecar request timeout (seconds)
# sidecar 请求超时(秒)
timeout_seconds: 60
# Reuse session key per account+proxy to let sidecar persist cookies/session
# 按账号+代理复用 session key让 sidecar 持久化 cookies/session
session_reuse_enabled: true
# Session TTL in sidecar (seconds)
# sidecar 会话 TTL
session_ttl_seconds: 3600
storage:
# Storage type (local only for now)
# 存储类型(首发仅支持 local
type: "local"
# Local base path; empty uses /app/data/sora
# 本地存储基础路径;为空使用 /app/data/sora
local_path: ""
# Fallback to upstream URL when download fails
# 下载失败时回退到上游 URL
fallback_to_upstream: true
# Max concurrent downloads
# 并发下载上限
max_concurrent_downloads: 4
# Download timeout (seconds)
# 下载超时(秒)
download_timeout_seconds: 120
# Max download bytes
# 最大下载字节数
max_download_bytes: 209715200
# Enable debug logs for media storage
# 启用媒体存储调试日志
debug: false
cleanup:
# Enable cleanup task
# 启用清理任务
enabled: true
# Retention days
# 保留天数
retention_days: 7
# Cron schedule
# Cron 调度表达式
schedule: "0 3 * * *"
# Token refresh behavior
# token 刷新行为控制
token_refresh:
# Whether OpenAI refresh flow is allowed to sync linked Sora accounts
# 是否允许 OpenAI 刷新流程同步覆盖 linked_openai_account_id 关联的 Sora 账号 token
sync_linked_sora_accounts: false
# =============================================================================
# API Key Auth Cache Configuration
# API Key 认证缓存配置
# =============================================================================
api_key_auth_cache:
# L1 cache size (entries), in-process LRU/TTL cache
# L1 缓存容量(条目数),进程内 LRU/TTL 缓存
l1_size: 65535
# L1 cache TTL (seconds)
# L1 缓存 TTL
l1_ttl_seconds: 15
# L2 cache TTL (seconds), stored in Redis
# L2 缓存 TTLRedis 中存储
l2_ttl_seconds: 300
# Negative cache TTL (seconds)
# 负缓存 TTL
negative_ttl_seconds: 30
# TTL jitter percent (0-100)
# TTL 抖动百分比0-100
jitter_percent: 10
# Enable singleflight for cache misses
# 缓存未命中时启用 singleflight 合并回源
singleflight: true
# =============================================================================
# Dashboard Cache Configuration
# 仪表盘缓存配置
# =============================================================================
dashboard_cache:
# Enable dashboard cache
# 启用仪表盘缓存
enabled: true
# Redis key prefix for multi-environment isolation
# Redis key 前缀,用于多环境隔离
key_prefix: "sub2api:"
# Fresh TTL (seconds); within this window cached stats are considered fresh
# 新鲜阈值(秒);命中后处于该窗口视为新鲜数据
stats_fresh_ttl_seconds: 15
# Cache TTL (seconds) stored in Redis
# Redis 缓存 TTL
stats_ttl_seconds: 30
# Async refresh timeout (seconds)
# 异步刷新超时(秒)
stats_refresh_timeout_seconds: 30
# =============================================================================
# Dashboard Aggregation Configuration
# 仪表盘预聚合配置(重启生效)
# =============================================================================
dashboard_aggregation:
# Enable aggregation job
# 启用聚合作业
enabled: true
# Refresh interval (seconds)
# 刷新间隔(秒)
interval_seconds: 60
# Lookback window (seconds) for late-arriving data
# 回看窗口(秒),处理迟到数据
lookback_seconds: 120
# Allow manual backfill
# 允许手动回填
backfill_enabled: false
# Backfill max range (days)
# 回填最大跨度(天)
backfill_max_days: 31
# Recompute recent N days on startup
# 启动时重算最近 N 天
recompute_days: 2
# Retention windows (days)
# 保留窗口(天)
retention:
# Raw usage_logs retention
# 原始 usage_logs 保留天数
usage_logs_days: 90
# Hourly aggregation retention
# 小时聚合保留天数
hourly_days: 180
# Daily aggregation retention
# 日聚合保留天数
daily_days: 730
# =============================================================================
# Usage Cleanup Task Configuration
# 使用记录清理任务配置(重启生效)
# =============================================================================
usage_cleanup:
# Enable cleanup task worker
# 启用清理任务执行器
enabled: true
# Max date range (days) per task
# 单次任务最大时间跨度(天)
max_range_days: 31
# Batch delete size
# 单批删除数量
batch_size: 5000
# Worker interval (seconds)
# 执行器轮询间隔(秒)
worker_interval_seconds: 10
# Task execution timeout (seconds)
# 单次任务最大执行时长(秒)
task_timeout_seconds: 1800
# =============================================================================
# HTTP 写接口幂等配置
# Idempotency Configuration
# =============================================================================
idempotency:
# Observe-only 模式:
# true: 观察期,不带 Idempotency-Key 仍放行(但会记录)
# false: 强制期,不带 Idempotency-Key 直接拒绝(仅对接入幂等保护的接口生效)
observe_only: true
# 关键写接口幂等记录 TTL
default_ttl_seconds: 86400
# 系统操作接口update/rollback/restart幂等记录 TTL
system_operation_ttl_seconds: 3600
# processing 锁超时(秒)
processing_timeout_seconds: 30
# 可重试失败退避窗口(秒)
failed_retry_backoff_seconds: 5
# 持久化响应体最大长度(字节)
max_stored_response_len: 65536
# 过期幂等记录清理周期(秒)
cleanup_interval_seconds: 60
# 每轮清理最大删除条数
cleanup_batch_size: 500
# =============================================================================
# Concurrency Wait Configuration
# 并发等待配置
# =============================================================================
concurrency:
# SSE ping interval during concurrency wait (seconds)
# 并发等待期间的 SSE ping 间隔(秒)
ping_interval: 10
2025-12-18 13:50:39 +08:00
# =============================================================================
# Database Configuration (PostgreSQL)
# 数据库配置 (PostgreSQL)
2025-12-18 13:50:39 +08:00
# =============================================================================
database:
# Database host address
# 数据库主机地址
2025-12-18 13:50:39 +08:00
host: "localhost"
# Database port
# 数据库端口
2025-12-18 13:50:39 +08:00
port: 5432
# Database username
# 数据库用户名
2025-12-18 13:50:39 +08:00
user: "postgres"
# Database password
# 数据库密码
2025-12-18 13:50:39 +08:00
password: "your_secure_password_here"
# Database name
# 数据库名称
2025-12-18 13:50:39 +08:00
dbname: "sub2api"
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
# SSL mode: disable, prefer, require, verify-ca, verify-full
# SSL 模式disable禁用, prefer优先加密默认, require要求, verify-ca验证CA, verify-full完全验证
# 默认值为 "prefer",数据库支持 SSL 时自动使用加密连接,不支持时回退明文
sslmode: "prefer"
# Max open connections (高并发场景建议 256+,需配合 PostgreSQL max_connections 调整)
# 最大打开连接数
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
max_open_conns: 256
# Max idle connections (建议为 max_open_conns 的 50%,减少频繁建连开销)
# 最大空闲连接数
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
max_idle_conns: 128
# Connection max lifetime (minutes)
# 连接最大存活时间(分钟)
conn_max_lifetime_minutes: 30
# Connection max idle time (minutes)
# 空闲连接最大存活时间(分钟)
conn_max_idle_time_minutes: 5
2025-12-18 13:50:39 +08:00
# =============================================================================
# Redis Configuration
# Redis 配置
2025-12-18 13:50:39 +08:00
# =============================================================================
redis:
# Redis host address
# Redis 主机地址
2025-12-18 13:50:39 +08:00
host: "localhost"
# Redis port
# Redis 端口
2025-12-18 13:50:39 +08:00
port: 6379
# Redis password (leave empty if no password is set)
# Redis 密码(如果未设置密码则留空)
2025-12-18 13:50:39 +08:00
password: ""
# Database number (0-15)
# 数据库编号0-15
2025-12-18 13:50:39 +08:00
db: 0
# Connection pool size (max concurrent connections)
# 连接池大小(最大并发连接数)
pool_size: 1024
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
# Minimum number of idle connections (高并发场景建议 128+,保持足够热连接)
# 最小空闲连接数
fix(audit): 第二批审计修复 — P0 生产 Bug、安全加固、性能优化、缓存一致性、代码质量 基于 backend-code-audit 审计报告,修复剩余 P0/P1/P2 共 34 项问题: P0 生产 Bug: - 修复 time.Since(time.Now()) 计时逻辑错误 (P0-03) - generateRandomID 改用 crypto/rand 替代固定索引 (P0-04) - IncrementQuotaUsed 重写为 Ent 原子操作消除 TOCTOU 竞态 (P0-05) 安全加固: - gateway/openai handler 错误响应替换为泛化消息,防止内部信息泄露 (P1-14) - usage_log_repo dateFormat 参数改用白名单映射,防止 SQL 注入 (P1-16) - 默认配置安全加固:sslmode=prefer、response_headers=true、mode=release (P1-18/19, P2-15) 性能优化: - gateway handler 循环内 defer 替换为显式 releaseWait 闭包 (P1-02) - group_repo/promo_code_repo Count 前 Clone 查询避免状态污染 (P1-03) - usage_log_repo 四个查询添加 LIMIT 10000 防止 OOM (P1-07) - GetBatchUsageStats 添加时间范围参数,默认最近 30 天 (P1-10) - ip.go CIDR 预编译为包级变量 (P1-11) - BatchUpdateCredentials 重构为先验证后更新 (P1-13) 缓存一致性: - billing_cache 添加 jitteredTTL 防止缓存雪崩 (P2-10) - DeductUserBalance/UpdateSubscriptionUsage 错误传播修复 (P2-12) - UserService.UpdateBalance 成功后异步失效 billingCache (P2-13) 代码质量: - search 截断改为按 rune 处理,支持多字节字符 (P2-01) - TLS Handshake 改为 HandshakeContext 支持 context 取消 (P2-07) - CORS 预检添加 Access-Control-Max-Age: 86400 (P2-16) 测试覆盖: - 新增 user_service_test.go(UpdateBalance 缓存失效 6 个用例) - 新增 batch_update_credentials_test.go(fail-fast + 类型验证 7 个用例) - 新增 response_transformer_test.go、ip_test.go、usage_log_repo_unit_test.go、search_truncate_test.go - 集成测试:IncrementQuotaUsed 并发测试、billing_cache 错误传播测试 - config_test.go 补充 server.mode/sslmode 默认值断言 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 19:46:42 +08:00
min_idle_conns: 128
# Enable TLS/SSL connection
# 是否启用 TLS/SSL 连接
enable_tls: false
2025-12-18 13:50:39 +08:00
# =============================================================================
# Ops Monitoring (Optional)
# 运维监控 (可选)
# =============================================================================
ops:
# Enable ops monitoring features (background jobs and APIs)
# 是否启用运维监控功能(后台任务和接口)
# Set to false to hide ops menu in sidebar and disable all ops features
# 设置为 false 可在左侧栏隐藏运维监控菜单并禁用所有运维监控功能
# Other detailed settings (cleanup, aggregation, etc.) are configured in ops settings dialog
# 其他详细设置(数据清理、预聚合等)在运维监控设置对话框中配置
enabled: true
2025-12-18 13:50:39 +08:00
# =============================================================================
# JWT Configuration
# JWT 配置
2025-12-18 13:50:39 +08:00
# =============================================================================
jwt:
# IMPORTANT: Change this to a random string in production!
# 重要:生产环境中请更改为随机字符串!
# Generate with / 生成命令: openssl rand -hex 32
2025-12-18 13:50:39 +08:00
secret: "change-this-to-a-secure-random-string"
# Token expiration time in hours (max 168)
# 令牌过期时间(小时,最大 168
2025-12-18 13:50:39 +08:00
expire_hour: 24
# Access Token 过期时间(分钟)
# 优先级说明:
# - >0: 按分钟生效(优先于 expire_hour
# - =0: 回退使用 expire_hour
access_token_expire_minutes: 0
2025-12-18 13:50:39 +08:00
# =============================================================================
# TOTP (2FA) Configuration
# TOTP 双因素认证配置
# =============================================================================
totp:
# IMPORTANT: Set a fixed encryption key for TOTP secrets.
# 重要:设置固定的 TOTP 加密密钥。
# If left empty, a random key will be generated on each startup, causing all
# existing TOTP configurations to become invalid (users won't be able to
# login with 2FA).
# 如果留空,每次启动将生成随机密钥,导致现有的 TOTP 配置失效(用户无法使用
# 双因素认证登录)。
# Generate with / 生成命令: openssl rand -hex 32
encryption_key: ""
# =============================================================================
# LinuxDo Connect OAuth Login (SSO)
# LinuxDo Connect OAuth 登录(用于 Sub2API 用户登录)
# =============================================================================
linuxdo_connect:
enabled: false
client_id: ""
client_secret: ""
authorize_url: "https://connect.linux.do/oauth2/authorize"
token_url: "https://connect.linux.do/oauth2/token"
userinfo_url: "https://connect.linux.do/api/user"
scopes: "user"
# 示例: "https://your-domain.com/api/v1/auth/oauth/linuxdo/callback"
redirect_url: ""
# 安全提示:
# - 建议使用同源相对路径(以 / 开头),避免把 token 重定向到意外的第三方域名
# - 该地址不应包含 #fragment本实现使用 URL fragment 传递 access_token
frontend_redirect_url: "/auth/linuxdo/callback"
token_auth_method: "client_secret_post" # client_secret_post | client_secret_basic | none
# 注意:当 token_auth_method=nonepublic client必须启用 PKCE
use_pkce: false
userinfo_email_path: ""
userinfo_id_path: ""
userinfo_username_path: ""
2025-12-18 13:50:39 +08:00
# =============================================================================
# Default Settings
# 默认设置
2025-12-18 13:50:39 +08:00
# =============================================================================
default:
# Initial admin account (created on first run)
# 初始管理员账户(首次运行时创建)
2025-12-18 13:50:39 +08:00
admin_email: "admin@example.com"
admin_password: "admin123"
# Default settings for new users
# 新用户默认设置
# Max concurrent requests per user
# 每用户最大并发请求数
user_concurrency: 5
# Initial balance for new users
# 新用户初始余额
user_balance: 0
2025-12-18 13:50:39 +08:00
# API key settings
# API 密钥设置
# Prefix for generated API keys
# 生成的 API 密钥前缀
api_key_prefix: "sk-"
2025-12-18 13:50:39 +08:00
# Rate multiplier (affects billing calculation)
# 费率倍数(影响计费计算)
2025-12-18 13:50:39 +08:00
rate_multiplier: 1.0
# =============================================================================
# Rate Limiting
# 速率限制
2025-12-18 13:50:39 +08:00
# =============================================================================
rate_limit:
# Cooldown time (in minutes) when upstream returns 529 (overloaded)
# 上游返回 529过载时的冷却时间分钟
2025-12-18 13:50:39 +08:00
overload_cooldown_minutes: 10
# =============================================================================
# Pricing Data Source (Optional)
# 定价数据源(可选)
2025-12-18 13:50:39 +08:00
# =============================================================================
pricing:
# URL to fetch model pricing data (default: pinned model-price-repo commit)
# 获取模型定价数据的 URL默认固定 commit 的 model-price-repo
remote_url: "https://raw.githubusercontent.com/Wei-Shaw/model-price-repo/c7947e9871687e664180bc971d4837f1fc2784a9/model_prices_and_context_window.json"
2025-12-18 13:50:39 +08:00
# Hash verification URL (optional)
# 哈希校验 URL可选
hash_url: "https://raw.githubusercontent.com/Wei-Shaw/model-price-repo/c7947e9871687e664180bc971d4837f1fc2784a9/model_prices_and_context_window.sha256"
2025-12-18 13:50:39 +08:00
# Local data directory for caching
# 本地数据缓存目录
2025-12-18 13:50:39 +08:00
data_dir: "./data"
# Fallback pricing file
# 备用定价文件
2025-12-18 13:50:39 +08:00
fallback_file: "./resources/model-pricing/model_prices_and_context_window.json"
# Update interval in hours
# 更新间隔(小时)
2025-12-18 13:50:39 +08:00
update_interval_hours: 24
# Hash check interval in minutes
# 哈希检查间隔(分钟)
2025-12-18 13:50:39 +08:00
hash_check_interval_minutes: 10
# =============================================================================
# Billing Configuration
# 计费配置
# =============================================================================
billing:
circuit_breaker:
# Enable circuit breaker for billing service
# 启用计费服务熔断器
enabled: true
# Number of failures before opening circuit
# 触发熔断的失败次数阈值
failure_threshold: 5
# Time to wait before attempting reset (seconds)
# 熔断后重试等待时间(秒)
reset_timeout_seconds: 30
# Number of requests to allow in half-open state
# 半开状态允许通过的请求数
half_open_requests: 3
# =============================================================================
# Turnstile Configuration
# Turnstile 人机验证配置
# =============================================================================
turnstile:
# Require Turnstile in release mode (when enabled, login/register will fail if not configured)
# 在 release 模式下要求 Turnstile 验证(启用后,若未配置则登录/注册会失败)
required: false
# =============================================================================
# Gemini OAuth (Required for Gemini accounts)
# Gemini OAuth 配置Gemini 账户必需)
# =============================================================================
# Sub2API supports TWO Gemini OAuth modes:
# Sub2API 支持两种 Gemini OAuth 模式:
#
# 1. Code Assist OAuth (requires GCP project_id)
# 1. Code Assist OAuth需要 GCP project_id
# - Uses: cloudcode-pa.googleapis.com (Code Assist API)
# - 使用cloudcode-pa.googleapis.comCode Assist API
#
# 2. AI Studio OAuth (no project_id needed)
# 2. AI Studio OAuth不需要 project_id
# - Uses: generativelanguage.googleapis.com (AI Studio API)
# - 使用generativelanguage.googleapis.comAI Studio API
#
# Default: Uses Gemini CLI's public OAuth credentials (same as Google's official CLI tool)
# 默认:使用 Gemini CLI 的公开 OAuth 凭证(与 Google 官方 CLI 工具相同)
gemini:
oauth:
# OAuth 客户端配置说明:
# 1) 留空 client_id/client_secret使用 Gemini CLI 内置 OAuth Client其 client_secret 需通过环境变量注入)
# - GEMINI_CLI_OAUTH_CLIENT_SECRET
# 2) 同时设置 client_id/client_secret使用你自建的 OAuth Client推荐权限更完整
#
# 注意client_id 与 client_secret 必须同时为空或同时非空。
client_id: ""
client_secret: ""
# Optional scopes (space-separated). Leave empty to auto-select based on oauth_type.
# 可选的权限范围(空格分隔)。留空则根据 oauth_type 自动选择。
scopes: ""
quota:
# Optional: local quota simulation for Gemini Code Assist (local billing).
# 可选Gemini Code Assist 本地配额模拟(本地计费)。
# These values are used for UI progress + precheck scheduling, not official Google quotas.
# 这些值用于 UI 进度显示和预检调度,并非 Google 官方配额。
tiers:
LEGACY:
# Pro model requests per day
# Pro 模型每日请求数
pro_rpd: 50
# Flash model requests per day
# Flash 模型每日请求数
flash_rpd: 1500
# Cooldown time (minutes) after hitting quota
# 达到配额后的冷却时间(分钟)
cooldown_minutes: 30
PRO:
# Pro model requests per day
# Pro 模型每日请求数
pro_rpd: 1500
# Flash model requests per day
# Flash 模型每日请求数
flash_rpd: 4000
# Cooldown time (minutes) after hitting quota
# 达到配额后的冷却时间(分钟)
cooldown_minutes: 5
ULTRA:
# Pro model requests per day
# Pro 模型每日请求数
pro_rpd: 2000
# Flash model requests per day (0 = unlimited)
# Flash 模型每日请求数0 = 无限制)
flash_rpd: 0
# Cooldown time (minutes) after hitting quota
# 达到配额后的冷却时间(分钟)
cooldown_minutes: 5
# =============================================================================
# Update Configuration (在线更新配置)
# =============================================================================
update:
# Proxy URL for accessing GitHub (used for online updates and pricing data)
# 用于访问 GitHub 的代理地址(用于在线更新和定价数据获取)
# Supports: http, https, socks5, socks5h
# Examples:
# - HTTP proxy: "http://127.0.0.1:7890"
# - SOCKS5 proxy: "socks5://127.0.0.1:1080"
# - With authentication: "http://user:pass@proxy.example.com:8080"
# Leave empty for direct connection (recommended for overseas servers)
# 留空表示直连(适用于海外服务器)
proxy_url: ""