dstack

dstack copied to clipboard

[Feature]: Volume lifecycle management

4

**1.1. [Long-term] Registering an existing network volume** They want to be able to register an existing volume (just like we already support for AWS). Example for AWS: ```yaml type: volume...

peterschmidt85

[Bug] Cannot create a new gateway while the service is still running (after deleting the previous gateway)

1

Steps to reproduce 1. Create a gateway 2. Run a service 3. Delete the gateway 4. Create a gateway Actual behavour: 1. Coudn't create a gateway second time (while the...

peterschmidt85

[UI]: Instance details page + Link from run/job to instance

1

### Problem Users cannot find out which instance a particular job is running on. Knowing the exact instance can be useful in many cases, such as terminating an underperforming instance....

jvstme

[Feature]: Select the best available offer in terms of compute/price

1

### Problem dstack apparently does not select the best offer available in terms of compute resources and price. Consider the partial config below: ```yaml ... resources: gpu: A5000:24GB:1 memory: 30GB.....

hamzashezad

[Feature]: Allow one instance to run multiple tasks/services

1

### Problem First, thank you for the great work on this project! I have an on-prem server with one A100 GPU, and I’m using ssh-fleet with auto block enabled. My...

joesu-angible

[Bug]: Dangling services and replicas on the gateway

1

### Steps to reproduce 1. Apply a service configuration with `name` set. For simpler reproduction, use `replicas: 0..1`. 2. Simulate a server-to-gateway communication issue, e.g., turn off the network on...

jvstme

Do not bind `dstack-runner` to the public IP address

1

### Problem Currently, `dstack-runner` always [listens](https://github.com/dstackai/dstack/blob/b6aa7e89781d59ac2caebf3590afbf42a3a43d4a/runner/cmd/runner/main.go#L41) on all network interfaces, including the instance's public IP address if the container runs in [host network mode](https://github.com/dstackai/dstack/blob/b6aa7e89781d59ac2caebf3590afbf42a3a43d4a/src/dstack/_internal/core/models/common.py#L118). In general, this is not an...

jvstme

[UX] Change the default `retry` policy to infinitely retry on `no-capacity`

1

By default, runs have no retry policy. If a run can’t find capacity at submission time, it fails with no offers. In most cases, this isn’t the desired default. A...

peterschmidt85

[Bug]: `dstack attach` doesn't exit when connection is dropped

3

### Steps to reproduce When attached via `dstack apply`, if the connection drops (possibly due to a laptop going to sleep and coming back), dstack will notify the user and...

jsadusk

[Bug]: A run fails if a private Git has (using SSH auth) has a Git submodule using HTTPS auth

2

### Steps to reproduce 1. Clone a Git repo using SSH auth Example: ```shell git clone [email protected]:peterschmidt85/git-submodule-parent.git cd git-submodule-parent ``` 2. Add a submodule using HTTPS auth ```shell git submodule...

peterschmidt85