Mount GCS via single host gcsfuse and expose in container #2294
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Run gcsfuse once on the host at /tmp/gcsfuse_mount - Expose mount inside container at /opt/gcsfuse_mount via symlink - Avoid Docker bind-mount source-path creation failures - Enable allow_other + permissive modes so container can read FUSE mount - Thread --verbose through to Ray CLI (ray up/down/attach/exec -v)
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refactors the GCS mount strategy to use a single host-level gcsfuse mount and propagates verbose flags through Ray CLI commands.
Key changes:
- Moves gcsfuse mounting from per-container setup to host-level initialization at
/tmp/gcsfuse_mount - Adds FUSE configuration with
allow_otherand permissive modes to enable container access - Implements verbose flag propagation from cluster.py CLI to Ray commands (up/down/attach/exec)
Reviewed changes
Copilot reviewed 12 out of 12 changed files in this pull request and generated 11 comments.
Show a summary per file
| File | Description |
|---|---|
| scripts/ray/cluster.py | Added _maybe_add_ray_verbose helper function and updated all Ray CLI invocations to support verbose flag passthrough; updated _stop_cluster_internal signature to accept Context |
| infra/marin-us-west4.yaml | Added gcsfuse installation and host-level mount in initialization_commands; replaced direct mount with symlink approach in setup_commands |
| infra/marin-us-east5.yaml | Same gcsfuse changes as us-west4 for consistency across regions |
| infra/marin-us-east5-a.yaml | Same gcsfuse changes as us-west4 for consistency across regions |
| infra/marin-us-east1.yaml | Same gcsfuse changes as us-west4 for consistency across regions |
| infra/marin-us-central2.yaml | Same gcsfuse changes as us-west4 for consistency across regions |
| infra/marin-us-central2-staging.yaml | Same gcsfuse changes as us-west4 for staging environment |
| infra/marin-us-central1.yaml | Same gcsfuse changes as us-west4 for consistency across regions |
| infra/marin-eu-west4.yaml | Same gcsfuse changes as us-west4 for EU region |
| infra/marin-eu-west4-a.yaml | Same gcsfuse changes as us-west4 for EU region |
| infra/marin-cluster-template.yaml | Template file updated with gcsfuse changes to propagate to future cluster configurations |
| infra/marin-big-run.yaml | Same gcsfuse changes as us-west4 for big-run cluster |
Comment on lines
+96
to
+97
| - if [ -e /opt/gcsfuse_mount ] && [ ! -L /opt/gcsfuse_mount ]; then sudo rm -rf /opt/gcsfuse_mount; fi | ||
| - sudo ln -sfn /tmp/gcsfuse_mount /opt/gcsfuse_mount |
There was a problem hiding this comment.
The symlink created at /opt/gcsfuse_mount will not be accessible inside the Docker container because /opt is not mounted into the container. Only /tmp is mounted (line 48). For the mount to be accessible at /opt/gcsfuse_mount inside the container, you need to add a volume mount like "-v /opt:/opt" to both head_run_options and worker_run_options.
Member
Author
There was a problem hiding this comment.
lies
lrwxrwxrwx 1 root root 18 Jan 6 22:37 /opt/gcsfuse_mount -> /tmp/gcsfuse_mount
bash: gcsfuse_mount: command not found
a
dedupe
gcsfuse_mount
helmet-data
huggingface-cache
marin-us-central2
medu-models
models
nfliu
nvidia--Llama-Nemotron-Post-Training-Dataset-v1-ed905e6
rjpower
approved these changes
Jan 7, 2026
Collaborator
rjpower
left a comment
rjpower
left a comment
There was a problem hiding this comment.
seems fine, not sure why we use /tmp/ outside but /opt inside but meh
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Part of putting vllm tpu in docker