Synthetic / operatormixed
外部探针、追踪查询、告警响应
ARCHITECTURE DOSSIER · 07
日志、可观测与恢复
Nginx、systemd、Docker、Loki、Grafana、备份与故障演练
RISK
OPSCODE + RUNTIME + QA01Client
02Edge
Nginx access/errorrunning
request_id、route、status、latency
03Service
systemd / Docker logsrunning
structured event、domain ids 与 errors
Alloy → Loki → Grafanarunning
采集、索引、查询与 dashboard
Backup / restore jobsrisk
manifest、copy、verify、drill
04State
Single root EBSrisk
当前 DB、repos、packages、logs、backups
CRITICAL SEQUENCE
真实主链与事务边界
每一步都必须能回到调用方、契约、状态 owner 与失败证据；UI 只显示服务端已经确认的真值。
- 01
Edge 接收请求并生成/透传 request_id
→ - 02
服务日志绑定 actor/domain entity/release
→ - 03
Alloy 收集 Nginx、Docker 与 journals
→ - 04
Loki 按 service/release/status 查询
→ - 05
Grafana 告警关联 runbook 与 owner
→ - 06
备份离机、校验并定期在隔离环境恢复
INTERFACE REGISTER
跨项目接口与安全条件
producer、consumer、契约和 guard 同时登记；其中任何一列缺失都不能进入多团队并行开发。
ProducerVersioned contractConsumerSecurity / consistency guard
Nginx
Access event {request_id,host,path,status,duration}LokiPII redaction
Services
Structured log {level,event,entity,release,trace}Alloy/Lokino secrets / bounded labels
Backup job
BackupManifest {source,as_of,hash,target}Restore drilloff-host copy + verify
STATE OWNERSHIP
数据真值与恢复
STORELoki + journal
RETENTION当前约 4 天本机
RECOVERY不作为业务恢复源
STORElocal backup now
RETENTIONtimer policy
RECOVERY需迁离根卷
STORErepos + PostgreSQL
RETENTIONsnapshot timer
RECOVERY需要隔离 restore drill
FAILURE MODEL
故障必须怎样收敛
SYMPTOM业务、日志与备份同时丢失
CONTAIN当前无跨盘隔离
P0 architecture findingSYMPTOMLoki 写入/查询退化
CONTAIN高基数字段放 body，不做 label
ingestion metricsSYMPTOM灾难时无法启动服务
CONTAIN以 restore drill 作为唯一通过标准
drill report + RTOQA RELEASE GATES
不是画完架构图就算完成
这些门禁连接 UI 状态、接口负向路径、运行证据与可恢复性；发布证据必须绑定 exact commit。
request_id 贯穿边缘与服务
synthetic trace
日志不含 token/PII
redaction scan
异地副本与 hash 校验
P0 · 未完成
按服务演练 RPO/RTO
季度 restore drill