feat(orchestrator): harden zero-impact shadow rollout

Gate and bound Portal shadow observations while preserving Native execution. Add fail-closed service boundaries, terminal retention controls, Canary readiness telemetry, ops visibility, and isolated regression coverage.
This commit is contained in:
john
2026-07-25 07:28:37 +08:00
parent 08a48e4849
commit 6df82818c5
33 changed files with 1569 additions and 108 deletions
+18
View File
@@ -61,6 +61,24 @@ H5_PUBLIC_BASE_URL=http://127.0.0.1:5173
# MEMIND_AGENT_RUN_WORKER_ID=<optional-instance-id>
# MEMIND_AGENT_RUN_WORKER_EXPECT_RUNNING=0
# LangGraph Orchestrator Shadow(默认完全不接入 Agent Run 请求路径)。
# 仅在独立 Orchestrator 已健康、service token 已配置后,先显式打开观察 wiring,
# 再通过 memindadm 选择 shadow。关闭 wiring 时不查询编排配置、不排队、不传用户数据。
# MEMIND_ORCHESTRATOR_SHADOW_OBSERVATION_ENABLED=0
# MEMIND_ORCHESTRATOR_URL=http://127.0.0.1:8093
# MEMIND_ORCHESTRATOR_SERVICE_TOKEN=<local-or-deployment-service-token>
# MEMIND_ORCHESTRATOR_SHADOW_MAX_CONCURRENCY=2
# MEMIND_ORCHESTRATOR_SHADOW_MAX_QUEUE=100
# 非 loopback URL 必须精确列入;逗号分隔,只接受 origin。
# MEMIND_ORCHESTRATOR_ALLOWED_ORIGINS=https://orchestrator.internal
# 真实执行交接是另一道独立闸门,默认保持关闭。
# MEMIND_ORCHESTRATOR_EXECUTION_HANDOFF_ENABLED=0
# MEMIND_ORCHESTRATOR_KILL_SWITCH=0
# Orchestrator 终态数据保留清理默认关闭;在独立服务环境配置,建议真实 Canary 前明确周期。
# MEMIND_ORCHESTRATOR_RETENTION_DAYS=0
# MEMIND_ORCHESTRATOR_RETENTION_SWEEP_INTERVAL_MS=86400000
# MEMIND_ORCHESTRATOR_RETENTION_SWEEP_LIMIT=100
# Runtime SLO 日报(默认每日 23:55 写 reports/runtime-slo)。
# MEMIND_RUNTIME_REPORT_DIR=/Users/john/Project/Memind/reports/runtime-slo
# MEMIND_RUNTIME_SLO_HOUR=23