hectorzhao/lislgosms
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf: batch CMPP inbound workflow processing

Browse Source
This commit is contained in:
hectorzhao
2026-08-21 09:44:51 +08:00
parent 57192b7586
commit 52028b9bbd
15 changed files with 662 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/production-deployment.md
+2 -1
View File
@@ -154,7 +154,7 @@ curl http://127.0.0.1:12026/
redis-cli -h 127.0.0.1 -p 6379 ping
pg_isready -d "$(grep '^DATABASE_URL=' /etc/cmpp-platform/cmpp-platform.env | cut -d= -f2-)"
grep -E '^(API_ENABLE_SEND_WORKER|API_SEND_WORKER_CONCURRENCY|GATEWAY_SUBMIT_WORKER_CONCURRENCY|GATEWAY_SUBMIT_RESULT_WORKER_CONCURRENCY|GATEWAY_CMPP_INBOUND_MAX_CONCURRENCY)=' /etc/cmpp-platform/cmpp-platform.env
grep -E '^(CMPP_INBOUND_FAST_PATH_ENABLED|CMPP_INBOUND_WORKFLOW_WORKER_ENABLED|API_INBOUND_WORKFLOW_CONCURRENCY|API_INBOUND_WORKFLOW_POLL_INTERVAL_MS|API_INBOUND_WORKFLOW_STALE_SECONDS|API_DB_POOL_MAX|API_WORKER_DB_POOL_MAX|API_WORKER_METRICS_PORT)=' /etc/cmpp-platform/cmpp-platform.env
grep -E '^(CMPP_INBOUND_FAST_PATH_ENABLED|CMPP_INBOUND_WORKFLOW_WORKER_ENABLED|API_INBOUND_WORKFLOW_CONCURRENCY|API_INBOUND_WORKFLOW_BATCH_ENABLED|API_INBOUND_WORKFLOW_BATCH_SIZE|API_INBOUND_WORKFLOW_POLL_INTERVAL_MS|API_INBOUND_WORKFLOW_STALE_SECONDS|API_DB_POOL_MAX|API_WORKER_DB_POOL_MAX|API_WORKER_METRICS_PORT)=' /etc/cmpp-platform/cmpp-platform.env
systemctl is-active cmpp-api cmpp-send-worker cmpp-gateway
curl -fsS http://127.0.0.1:9465/metrics | grep '^cmpp_worker_inbound_workflow_'
redis-cli --scan --pattern 'rate:gateway:channel:*'
@@ -186,3 +186,4 @@ bash tools/deploy/production-deploy.sh
- 环境必须显式启用`CMPP_INBOUND_FAST_PATH_ENABLED=true`、`CMPP_INBOUND_WORKFLOW_WORKER_ENABLED=true`，并给出正整数`API_INBOUND_WORKFLOW_CONCURRENCY`、`API_DB_POOL_MAX`和`API_WORKER_DB_POOL_MAX`；推荐初始值分别为32槽、API池32、Worker池8、轮询100ms、租约300秒。API systemd角色必须是`api`，Worker角色必须是`worker`；Worker可通过`API_WORKER_DATABASE_URL`使用独立地址，未配置时仍使用同一数据库地址但保持独立进程和有界连接池。发布前必须核对两池之和、其他服务连接与运维余量不超过PostgreSQL`max_connections`。
- Worker日志目录归`cmpp-api:cmpp-security`且仅服务可写，9465只监听回环并加入Prometheus `cmpp-send-worker` target。发布后必须验证两进程均为非root、API/Gateway health、Worker metrics、PostgreSQL/Redis，以及Inbox pending/processing/最老等待可观测。
- 回滚前先停止Gateway、API和Worker，保留故障现场Inbox及日志；如恢复旧数据库备份，必须同时恢复对应源码和环境/systemd资产。不得在回滚时删除pending Inbox或重投真实短信。
- 第三阶段要求环境显式设置`API_INBOUND_WORKFLOW_BATCH_ENABLED=true`和正整数`API_INBOUND_WORKFLOW_BATCH_SIZE`，初始建议64且不得大于Worker业务槽的可解释倍数。批次增大前必须核对PostgreSQL参数数量、单事务持续时间、Worker RSS和租约时长；付费短信仍走逐条账务锁，不能用零计费批量结果替代付费链路验收。