fix(p0): GPU mezzanine传输切回OSS绕开Tailscale反向卡顿 #2084

Merged
xiaoxia merged 1 commits from fix/p0-gpu-mezzanine-oss into develop 2026-09-28 08:51:38 +08:00
Owner

背景

Tailscale 反向链路(P4000 从 116 relay 下载 mezzanine)间歇性 TCP stall,导致 GPU 编码首字节 20s 超时触发 fallback 到 CPU 软编,渲染退化到 ~3 分钟。

正向链路(116→P4000 POST 编码请求)正常,staging 业务机 116 上 tailscaled 是 systemd 服务稳定运行;问题集中在 P4000 → 116 nginx:8092 这个反向下载方向(DERP relay 路径抖动/打洞不稳定)。

修改

单文件:deploy/configs/.env.staging

- GPU_ENCODE_MEZZANINE_TRANSPORT=relay
+ GPU_ENCODE_MEZZANINE_TRANSPORT=oss

效果

  • P4000 直接从阿里云 OSS 公网 URL 下载 mezzanine(GPU_ENCODE_OSS_TMP_PREFIX=tmp/gpu-mezzanine/),绕开 Tailscale 反向链路
  • 编码结果回传仍走 relay(116→P4000 正向 POST 正常)
  • GPU_ENCODE_FALLBACK_CPU=true 保留,万一 OSS 下载也失败仍可 CPU 兜底

影响面

  • 仅 staging 环境变量(.env.staging);生产 .env 不受影响
  • 部署后 watchtower 自动拉起 worker 新容器,env 生效;API 需同步重启读取新 .env

验证

部署后观察:

  1. 新提交视频渲染耗时回到 GPU 加速水平(正常大素材 16-39s)
  2. worker 日志不再出现 [gpu-encoder] P4000 首字节超时后 fallback
  3. 无需回滚代码,若 OSS 下载出现新问题可把 .env 改回 relay 重启即可
## 背景 Tailscale **反向链路**(P4000 从 116 relay 下载 mezzanine)间歇性 TCP stall,导致 GPU 编码首字节 20s 超时触发 fallback 到 CPU 软编,渲染退化到 ~3 分钟。 正向链路(116→P4000 POST 编码请求)正常,staging 业务机 116 上 tailscaled 是 systemd 服务稳定运行;问题集中在 P4000 → 116 nginx:8092 这个反向下载方向(DERP relay 路径抖动/打洞不稳定)。 ## 修改 **单文件:`deploy/configs/.env.staging`** ```diff - GPU_ENCODE_MEZZANINE_TRANSPORT=relay + GPU_ENCODE_MEZZANINE_TRANSPORT=oss ``` ## 效果 - P4000 直接从阿里云 OSS 公网 URL 下载 mezzanine(GPU_ENCODE_OSS_TMP_PREFIX=tmp/gpu-mezzanine/),绕开 Tailscale 反向链路 - 编码结果回传仍走 relay(116→P4000 正向 POST 正常) - GPU_ENCODE_FALLBACK_CPU=true 保留,万一 OSS 下载也失败仍可 CPU 兜底 ## 影响面 - 仅 staging 环境变量(.env.staging);生产 .env 不受影响 - 部署后 watchtower 自动拉起 worker 新容器,env 生效;API 需同步重启读取新 .env ## 验证 部署后观察: 1. 新提交视频渲染耗时回到 GPU 加速水平(正常大素材 16-39s) 2. worker 日志不再出现 `[gpu-encoder] P4000` 首字节超时后 fallback 3. 无需回滚代码,若 OSS 下载出现新问题可把 .env 改回 relay 重启即可
xiaoxia added 1 commit 2026-09-28 08:31:59 +08:00
fix(p0): GPU mezzanine传输切回OSS绕开Tailscale反向卡顿
CI/CD Pipeline / Check push changed paths (pull_request) Has been skipped
CI/CD Pipeline / Dedup Check - skip PR tests when covered by push pipeline (pull_request) Successful in 1s
CI/CD Pipeline / Check if frontend-only change (pull_request) Successful in 2s
CI/CD Pipeline / Frontend Lint (pull_request) Has been skipped
CI/CD Pipeline / Frontend Unit Tests (pull_request) Has been skipped
CI/CD Pipeline / PR Build Web Image (pull_request) Has been skipped
CI/CD Pipeline / Build Staging API Image (pull_request) Has been skipped
CI/CD Pipeline / Build Staging Web Image (pull_request) Has been skipped
CI/CD Pipeline / Build Staging Worker Image (pull_request) Has been skipped
CI/CD Pipeline / Retag skipped Staging API Image (pull_request) Has been skipped
CI/CD Pipeline / Retag skipped Staging Web Image (pull_request) Has been skipped
CI/CD Pipeline / Retag skipped Staging Worker Image (pull_request) Has been skipped
CI/CD Pipeline / Deploy Staging (Watchtower auto-deploy) (pull_request) Has been skipped
CI/CD Pipeline / Staging E2E Tests (pull_request) Has been skipped
CI/CD Pipeline / Staging API Integration Tests (pull_request) Has been skipped
CI/CD Pipeline / ACR Image Cleanup (pull_request) Has been skipped
CI/CD Pipeline / PR Build API Image (pull_request) Successful in 52s
Preview Deploy / Deploy Preview Environment (pull_request) Successful in 1m28s
CI/CD Pipeline / Integration Tests (pull_request) Successful in 2m3s
CI/CD Pipeline / Validate - Python (mypy + alembic) (pull_request) Successful in 2m17s
CI/CD Pipeline / Validate - Style (pull_request) Successful in 2m52s
PR Automation / Auto Approve on CI Green (pull_request) Successful in 3m6s
CI/CD Pipeline / Validate - Security (pull_request) Successful in 4m28s
AI Code Review / AI Code Review (pull_request) Successful in 6m37s
CI/CD Pipeline / PR Build Worker Image (pull_request) Failing after 7m34s
CI/CD Pipeline / Unit Tests (pull_request) Successful in 9m11s
CI/CD Pipeline / Build Production API Image (pull_request) Has been skipped
CI/CD Pipeline / Build Production Web Image (pull_request) Has been skipped
CI/CD Pipeline / Build Production Worker Image (pull_request) Has been skipped
CI/CD Pipeline / CI Gate (pull_request) Failing after 1s
CI/CD Pipeline / Deploy Production (pull_request) Has been skipped
CI/CD Pipeline / Canary Release to Production (pull_request) Has been skipped
CI/CD Pipeline / Production Browser E2E (pull_request) Has been skipped
PR Automation / Auto Merge on CI Green + Approved (pull_request) Successful in 6m17s
Preview Cleanup / Cleanup Preview Environment (pull_request) Successful in 27s
ACR Cleanup / ACR Image Cleanup (pull_request_target) Successful in 4m31s
6d9d2e8179
Tailscale 反向链路(P4000 从 116 relay 下载 mezzanine)间歇性 TCP stall,
导致 GPU 编码 20s 首字节超时 fallback 到 CPU 软编,渲染变慢到 ~3 分钟。

改 transport=oss:P4000 直接从阿里云 OSS 公网下载 mezzanine,绕开不稳定的
Tailscale 反向链路;编码结果回传仍走 relay(116→P4000 正向链路正常)。

🚀 预览环境已部署

项目 详情
PR号 #2084
预览链接 https://pr-2084.preview.xiaoxiajianji.com
API环境 staging

💡 预览环境使用 staging API 数据,请勿在预览环境中操作重要数据。

🔄 每次提交新代码后预览环境会自动更新。

🗑️ PR 关闭或合并后,预览环境会自动清理。

🚀 **预览环境已部署** | 项目 | 详情 | |------|------| | PR号 | #2084 | | 预览链接 | [https://pr-2084.preview.xiaoxiajianji.com](https://pr-2084.preview.xiaoxiajianji.com) | | API环境 | staging | > 💡 预览环境使用 staging API 数据,请勿在预览环境中操作重要数据。 > > 🔄 每次提交新代码后预览环境会自动更新。 > > 🗑️ PR 关闭或合并后,预览环境会自动清理。
xiaoxia merged commit 423be1446f into develop 2026-09-28 08:51:38 +08:00

🗑️ 预览环境已清理

PR #2084 已关闭或合并,对应的预览环境已被清理。

如有需要,可以重新打开 PR 来重新生成预览环境。

🗑️ **预览环境已清理** PR #2084 已关闭或合并,对应的预览环境已被清理。 > 如有需要,可以重新打开 PR 来重新生成预览环境。
Sign in to join this conversation.