Egor Sklyarov

[Feature]: Process jobs in parallel

1

### Problem Running a service with multiple replicas starts them sequentially, not parallel. This limitation has no benefits ### Solution Process all available jobs in parallel, similar to `process_runs` ###...

[Feature]: Check instance health via cloud provider API

1

### Problem Interrupted instances (spot or removed from outside of dstack) are shown as `idle` for a long time ### Solution If the shim doesn't respond, ask the backend about...

[Feature]: Use UUID as instance_id in GCP

1

### Problem Resubmitted jobs cause 409 responses from GCP if an interrupted instance is not fully deleted. It leads to provisioning in more expensive regions ### Solution Use truly unique...

[Bug]: `process_submitted_jobs` adds duplicating instance names to a pool

1

### dstack version master ### Python version - ### Host OS - ### Host Arch - ### What happened? Resubmitting interrupted job leads to duplicates in instance names (while the...

The instance cannot be reused by another user

3

Since SSH keys are added to the instance only during creation, other users cannot add their keys later. dstack server must update SSH keys on job submission (docker image submission,...

[Feature]: Always write server debug logs to a file

2

### Problem In case of a problem with the dstack server, it's hard to investigate without debug logs, which are not always turned on. Re-running the server with a debug...