wangbo
92e328a575
将协议模拟验收的供应商并发限额与 Worker 容量门禁分离，真实金丝雀继续保留生产限流。readyz 改用关键连接池并预热关键及 River 池，延长生产连接空闲周期，降低跨地域连接抖动。失败 Run 现在可以原子替换且失败报告记录任务数量，避免 validation 之间短暂放开正式流量。\n\n验证：Go 全量测试、go vet、PostgreSQL 集成测试、ShellCheck、迁移安全、发布脚本、pnpm lint/test/build、OpenAPI 无漂移均通过。
40 lines
1.8 KiB
Plaintext
40 lines
1.8 KiB
Plaintext
RELEASES_DIR=/var/lib/easyai-ai-gateway-cluster-release/releases
|
||
NAMESPACE=easyai
|
||
PUBLIC_BASE_URL=https://ai.51easyai.com
|
||
K3S_BIN=/usr/local/bin/k3s
|
||
DESIRED_STATE_DIR=/usr/local/share/easyai-ai-gateway-release/production
|
||
|
||
# Worker 容量只需修改本发布配置并执行获授权的 `capacity`，不再修改 Go 代码。
|
||
# 副本数属于 Deployment 配置，不能由容器内部环境变量改变；发布工具读取这两个变量并写入 replicas。
|
||
AI_GATEWAY_WORKER_REPLICAS_NINGBO=0
|
||
AI_GATEWAY_WORKER_REPLICAS_HONGKONG=2
|
||
AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=24
|
||
AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
|
||
AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT=48
|
||
AI_GATEWAY_WORKER_MEMORY_REQUEST_MIB=1536
|
||
AI_GATEWAY_WORKER_CPU_REQUEST_MILLICORES=500
|
||
AI_GATEWAY_DATABASE_MAX_CONNS=32
|
||
AI_GATEWAY_API_DATABASE_MAX_CONNS=32
|
||
AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS=4
|
||
AI_GATEWAY_DATABASE_RIVER_MAX_CONNS=8
|
||
AI_GATEWAY_API_DATABASE_RIVER_MAX_CONNS=4
|
||
AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=4
|
||
AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=300
|
||
AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
|
||
AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24
|
||
AI_GATEWAY_WORKER_AUTOSCALING_ENABLED=false
|
||
AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=0
|
||
AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG=1
|
||
AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=0
|
||
AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG=2
|
||
AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA=48
|
||
AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS=20
|
||
AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS=600
|
||
AI_GATEWAY_WORKER_DRAIN_TIMEOUT_SECONDS=600
|
||
AI_GATEWAY_NODE_MEMORY_TARGET_PERCENT=75
|
||
AI_GATEWAY_NODE_MEMORY_HARD_PERCENT=85
|
||
AI_GATEWAY_NODE_CPU_TARGET_PERCENT=70
|
||
AI_GATEWAY_POSTGRES_CONNECTION_BUDGET=150
|
||
# 2 个 API 池（2×32）加 2 个 capacity-controller 池（2×4）。
|
||
AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET=72
|