Lambda launch checklist: SSH key first, filesystems at launch, tunnels for ports

Lambda launch checklist: SSH key first, filesystems at launch, tunnels for ports

TL;DR

Before launching a Lambda instance: add your SSH key to your account first, adding it later does nothing for a running box. Attach every filesystem you need at launch, you cannot add one later without relaunching. Write checkpoints and datasets to /lambda/nfs/[filesystem], never [HOME]/...

Use this when you run into the situation in the title.

When not to use this skill: unrelated tasks. It covers only the procedure above.

Compatibility

Tool: SSH. The steps above apply to the commands named in them. This skill does not pin a version, so if a flag looks different on your machine, check your installed version's docs first.

Details

which is wiped on terminate. And reach Jupyter/TensorBoard through an SSH tunnel, only port 22 is open by default.

Context: Web (hermes-agent lambda-labs SKILL.md, key constraints): documents the real launch-time gotchas. Your SSH key must be added BEFORE launching, keys added afterwards are not injected into a running instance. Filesystems attach only at launch time, if you forget one you must terminate and relaunch. [HOME]/... is ephemeral fast local NVMe destroyed on termination, persistent data belongs in /lambda/nfs/[name]. Only port 22 is open by default, so Jupyter and TensorBoard need SSH tunnels.

Source: Web (hermes-agent lambda-labs SKILL.md, key constraints): documents the real launch-time gotchas. Your SSH key must be added BEFORE launching, keys added afterwards are not injected into a running instance. Filesystems attach only at launch time, if you forget one you must terminate and relaunch. [HOME]/... is ephemeral fast local NVMe destroyed on termination, persistent data belongs in /lambda/nfs/[name]. Only port 22 is open by default, so Jupyter and TensorBoard need SSH tunnels.

Verified resolution: Before launching a Lambda instance: add your SSH key to your account first, adding it later does nothing for a running box. Attach every filesystem you need at launch, you cannot add one later without relaunching. Write checkpoints and datasets to /lambda/nfs/[filesystem], never [HOME]/... which is wiped on terminate. And reach Jupyter/TensorBoard through an SSH tunnel, only port 22 is open by default.

Source URL: https://github.com/majiayu000/hermes-agent/blob/HEAD/optional-skills/mlops/lambda-labs/SKILL.md