skypilot-org/skypilot
View on GitHub[Core] nvcr dynamo image doesn't work on SkyPilot
Open
#8566 opened on Jan 13, 2026
good first issuegood starter issues
Description
Using base image nvcr.io/nvidia/ai-dynamo/sglang-runtime:0.7.1 stays stuck at initialization:
sky launch -c test3 --infra k8s --image-id docker:nvcr.io/nvidia/ai-dynamo/sglang-runtime:0.7.1 -- echo hi
Command to run: echo hi
Considered resources (1 node):
--------------------------------------------------------------------------------------
INFRA INSTANCE vCPUs Mem(GB) GPUS COST ($) CHOSEN
--------------------------------------------------------------------------------------
Kubernetes (---------) - 2 2 - 0.00 ✔
--------------------------------------------------------------------------------------
Launching a new cluster 'test3'. Proceed? [Y/n]:
⠙ Waiting for 'sky.launch' request to be scheduled: 7812b61a-6f28-41f2-9d70-ee29afda028f
⠼ Waiting for 'sky.launch' request to be scheduled: 7812b61a-6f28-41f2-9d70-ee29afda028f
⚙︎ Launching on Kubernetes.
└── Pod is up.
⠧ Preparing SkyPilot runtime (1/3 - initializing) View logs: sky logs --provision test3
From provision log:
+ '[' -f /etc/apt/sources.list ']'
+ update_apt_sources mirror.umd.edu /etc/apt/sources.list
+ local host=mirror.umd.edu
+ local apt_file=/etc/apt/sources.list
++ prefix_cmd
+++ id -u
++ '[' 1000 -ne 0 ']'
++ echo sudo
+ sudo sed -i -E 's|https?://[a-zA-Z0-9.-]+\.ubuntu\.com/ubuntu|http://mirror.umd.edu/ubuntu|g' /etc/apt/sources.list
environment: line 73: sudo: command not found
+ apt_update_install_with_retries rsync fuse
+ echo 'Install failed with mirror (ubuntu): mirror.umd.edu'
+ restore_source
+ '[' -f /etc/apt/sources.list.backup_skypilot/sources.list ']'
+ set -e
+ '[' 1 = 1 ']'
+ echo 'Error: required package install failed across all mirrors: rsync fuse'
+ return 1
+ echo 'Error: core package installation failed.'
+ exit 1
I suspect it's something to do with sudo not installed/permissions. Adding this works:
# Kubernetes-specific configuration
config:
kubernetes:
pod_config:
spec:
containers:
- securityContext:
# Run as root to allow SkyPilot to install necessary packages
runAsUser: 0
runAsGroup: 0