Repository navigation
[Feature] Support max_pulling_duration with run configurations #3040
Description
Steps to reproduce
Apply the configuration:
type: dev-environment
ide: vscode
image: nvcr.io/nvidia/tritonserver:25.06-py3
setup:
- pip install uv
resources:
gpu: L4
backends: [aws]
inactivity_duration: 1hActual behaviour
After spending 20 minutes in the pulling status, the run fails because of the pull timeout:
pullImage error: image pull interrupted: downloaded 11320888531 bytes out of 11320888531 (9.00MB/s): context deadline exceeded
Expected behaviour
Some images are expected to take more than 20 minutes to be pulled. It should be possible to use such images with dstack.
Possible solutions (not mutually exclusive):
- Make the timeout configurable in the run configuration.
- Calculate the default timeout based on the image size and instance resources.
- Instead of a timeout, interrupt pulling if there is no progress. Requires [Feature]: Show Docker image pull progress #2187 for observability.
dstack version
0.19.25
Server logs
Additional information
No response
Activity
github-actions commentedon Sep 28, 2025
This issue is stale because it has been open for 30 days with no activity.
github-actions commentedon Oct 15, 2025
This issue was closed because it has been inactive for 14 days since being marked as stale. Please reopen the issue if it is still relevant.
[-]Cannot use images that take more than 20 minutes to be pulled[/-][+][Feature] Support `max_pulling_duration` with run configurations[/+]peterschmidt85 commentedon Dec 2, 2025
Until #3337 is implemented, for container-based backends we can include provisioning time into
max_pulling_duration- this will be accurate in all cases except Kubernetes.