tao.chen
cdcfcb2e43
健康检查与根路径（/health/live、/health/ready、/api/v1/health、 /、/health/storage）没有用户、没业务动作，每秒被 K8s/LB 探针刷一次只会灌进无意义噪音。命中排除集即跳过审计行； 诊断日志（method/path/status/ms 走 stderr）照常打，对容器 运维排错仍有用。 * settings.audit_excluded_paths: list[str] 默认覆盖 5 条 基础设施路径，env AUDIT_EXCLUDED_PATHS 用逗号分隔 （pydantic NoDecode + field_validator 兼容 str/list） * main.py 模块级 _AUDIT_EXCLUDED = frozenset(...)， access_log 的 success/exception 两条审计行各加守卫 诊断无条件打 * 测试用 _AccessLogReplica 复刻 access_log 契约（不 import 真实 main.py），新增 4 个 case：排除根路径、排除 /health/live、 不排除路径照写审计、自定义排除集 顺带 schedule 模块：ExecutionResult 与 context 已迁到 schedule.domain.*（execution.py / orchestrator.py / scheduler.py / worker.py），调用点跟进；schedule 自身 18 个测试在改前改后均通过。
98 lines
4.6 KiB
Bash
98 lines
4.6 KiB
Bash
COMPOSE_PROJECT_NAME=model-platform-develop
|
|
SCHEDULE_EVENT_NAMESPACE=model-platform-develop
|
|
|
|
# External port of the Nginx gateway. Only Nginx is exposed to the host
|
|
# (architecture §2.2); backend/runtime/schedule stay on the Docker internal
|
|
# network. Override here to expose Nginx on a different host port.
|
|
GATEWAY_PORT=8890
|
|
|
|
# External MySQL. URL-encode reserved characters in DATABASE_URL.
|
|
MYSQL_HOST=127.0.0.1
|
|
MYSQL_PORT=3306
|
|
MYSQL_USER=root
|
|
MYSQL_PASSWORD=change-me
|
|
MYSQL_DATABASE=model_platform
|
|
DATABASE_URL=mysql+asyncmy://root:change-me@127.0.0.1:3306/model_platform?charset=utf8mb4
|
|
|
|
# Demo login is only intended for this self-hosted development UI.
|
|
DEMO_AUTH_ENABLED=true
|
|
JWT_SECRET=change-this-development-secret
|
|
|
|
# ============================================================================
|
|
# CRITICAL: must set BEFORE first run. The initial admin user is seeded by
|
|
# the deployment bootstrap. Never keep the development default in production.
|
|
# ============================================================================
|
|
INITIAL_ADMIN_PASSWORD=admin12345
|
|
|
|
# Backend loguru stderr sink level. One of DEBUG / INFO / WARNING / ERROR
|
|
# / CRITICAL. Anything else (e.g. lowercase) falls back to INFO inside
|
|
# configure_logging(). Change to DEBUG to see request bodies in
|
|
# runtime_client._jupyter_request.
|
|
LOG_LEVEL=INFO
|
|
|
|
# Audit log (backend HTTP interface compliance log). One line per HTTP
|
|
# request, written to data/logs/audit/audit-YYYY-MM-DD.log (one file per
|
|
# day). AUDIT_LOG_DIR is relative to the backend process cwd (/app in the
|
|
# container). AUDIT_LOG_RETENTION_DAYS=0 disables cleanup of old files.
|
|
AUDIT_LOG_DIR=
|
|
AUDIT_LOG_RETENTION_DAYS=30
|
|
# Exact paths excluded from the audit line (health/root probes carry no
|
|
# business value but fire every second from K8s/LB). The default already
|
|
# covers /health/live /health/ready /api/v1/health / /health/storage;
|
|
# leave empty to keep the default. Comma-separated, e.g. /health/live,/api/v1/health.
|
|
AUDIT_EXCLUDED_PATHS=
|
|
|
|
# Object storage. Two modes are supported:
|
|
# STORAGE_BACKEND=s3 — connects to an S3-compatible service (MinIO,
|
|
# RustFS, SeaweedFS, AWS S3, …). Requires the
|
|
# S3_* block below.
|
|
# STORAGE_BACKEND=local — stores objects on the local filesystem under
|
|
# LOCAL_STORAGE_BASE_DIR. Backend and runtime
|
|
# share this directory via a Docker volume
|
|
# (docker-compose.yml mounts `local-storage`).
|
|
# Useful for dev, single-node, air-gapped.
|
|
STORAGE_BACKEND=s3
|
|
LOCAL_STORAGE_BASE_DIR=/data
|
|
|
|
# Object storage (S3-compatible). Only used when STORAGE_BACKEND=s3.
|
|
# S3_ENDPOINT is the single upstream URL consumed by all 4 services:
|
|
# - nginx (via scripts/nginx-entrypoint.sh, which parses host + port)
|
|
# - backend / runtime / schedule (passed through to boto3 / rclone)
|
|
# S3_ACCESS_KEY / S3_SECRET_KEY are read by Python code in
|
|
# backend/ and schedule/ (boto3 credentials).
|
|
#
|
|
# S3 buckets are purpose-named:
|
|
# S3_WORKSPACE_BUCKET — workspace files (notebooks, scripts, working
|
|
# copies); layout is ``s3://<bucket>/<workspace_id>/...``.
|
|
# S3_VERSION_BUCKET — immutable script-version artifacts.
|
|
# S3_RUN_LOG_BUCKET — schedule run logs and execution results.
|
|
# S3_TRASH_BUCKET — soft-deleted objects; source bucket key is preserved
|
|
# as a prefix so restore is a same-key move.
|
|
S3_HOST=127.0.0.1
|
|
S3_PORT=9000
|
|
S3_ENDPOINT=http://127.0.0.1:9000
|
|
S3_ACCESS_KEY=change-me
|
|
S3_SECRET_KEY=change-me
|
|
S3_WORKSPACE_BUCKET=workspace
|
|
S3_VERSION_BUCKET=version
|
|
S3_RUN_LOG_BUCKET=run-log
|
|
S3_TRASH_BUCKET=trash
|
|
S3_TRASH_RETENTION_DAYS=30
|
|
|
|
# rclone RC (HTTP control API). The runtime container starts rclone with
|
|
# `--rc --rc-addr 0.0.0.0:5572 --rc-no-auth` (see runtime/src/runtime/mount.py),
|
|
# so the backend can POST /vfs/refresh here to invalidate the FUSE dir-cache
|
|
# after writing new workspace files. Default points at the runtime service
|
|
# over the compose network.
|
|
RCLONE_RC_URL=http://runtime:5572
|
|
|
|
# ============================================================================
|
|
# Service-to-service auth (P0-1 fix).
|
|
# Backend's /internal/v1/* storage control plane requires this shared secret.
|
|
# The schedule worker reads the same value and sends it as the
|
|
# ``X-Internal-Service-Token`` header. Value MUST match between backend and
|
|
# schedule. Generate a random 64-char string for any non-dev deployment:
|
|
# python -c "import secrets; print(secrets.token_urlsafe(48))"
|
|
# ============================================================================
|
|
INTERNAL_SERVICE_TOKEN=change-me-internal-service-token
|