[bug] ibm-power: job containers run at MTU 1500 on a 1450 network

未关闭
#954 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 4 天内回复

@atishay-kasliwal 已经在做这个了。

开始于 2026年11月10日。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
go, linux

调研方向

Start by finding where the setup reads network_mtu and where it passes --docker-network-mode podman, since the payload says the shared podman network ignores the runner key. Reproduce on a host whose default route is 1450 and compare podman0 against ip link show inside a job container. Done means the job network reports 1450 in both checks, and the provisioning output prints the detected value. Coordinate with #944 first, because both changes must share one source of the MTU.

由索引模型根据 Issue 内容生成。

描述

bug provider/IBMCLOUD
Problem

Job containers on an IBM Power (PowerVS) runner run with a 1500 MTU while the private network carries 1450. Small requests work and large transfers stall: an image pull or a package install hangs for minutes and the job trace shows no error line, because DNS and ping succeed.

The runner registers with no network_mtu, so the key is absent and the executor falls back to 1500. On our fleet the key reads 1450 and the packets are still 1500, because the runner is registered with --docker-network-mode podman (see #944). With a shared network the executor never creates a network for the job, so [runners.docker] network_mtu has no effect. The actual value comes from the podman bridge:

Item Value
underlay MTU 1450
podman0 MTU 1500
network_mtu in config.toml 1450

This issue replaces the earlier request to set network_mtu with the runner key. That key is inert under network_mode = podman.

Proposal
  • Detect the MTU of the default-route interface during setup.
  • Apply it to the network the jobs use:
    • the netavark mtu option on the default podman network, or
    • the bridge interface, if the network is created by the image.
  • When the setup creates the network explicitly, pass the value at creation time. A network created earlier at 1500 keeps 1500 until it is recreated, and recreating it needs no running container.
  • Keep network_mtu in step as well, so a setup without --docker-network-mode podman is correct too.
  • Print the detected value and the resulting bridge MTU in the provisioning output.

Related: #944 pins job containers to the shared podman network and adds a doc step that compares the two MTUs. That step is how this gap was found. The two changes should agree on one source of truth for the value.

Acceptance
  • On a host whose default route runs at 1450, the job network carries 1450 and ip link show inside a job container reports 1450.
  • A large transfer inside a job completes without a stall.
  • The provisioning output names the detected value.
主要语言
Go
星标
25
派生
41
平均合并
5 天 3 小时
30 天内合并 PR
16

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

redhat-developer/mapt 的其他 Issue

查看 redhat-developer/mapt 的全部 Issue

相似的 Issue

更多 Go Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。