Skip to content

Releases: dstackai/dstack

0.19.20

21 Jul 12:48
@r4victor r4victor
0.19.20
d6fc6d6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.20

User interface

Logs

This is a hotfix release addressing three major issues related to the UI:

  • The UI didn’t display newer AWS CloudWatch logs if there was a long gap between old and new logs.
  • Logs received before the 19th appeared as base64-encoded in the UI. The UI now includes a button to decode them automatically.
  • Logs were loaded from start to end, which made viewing very slow for long runs.

Note

The dstack logs CLI command may still be affected by the issues above. However, it’s less critical and will be addressed separately.

What's changed

Full changelog: 0.19.19...0.19.20

Contributors

un-def, olgenn, and 3 other contributors
Loading

0.19.19

17 Jul 05:54
@r4victor r4victor
0.19.19
8cf079a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.19

Fleets

SSH fleets in-place updates

You can now add and remove instances in SSH fleets without recreating the entire fleet.

type: fleet
name: ssh-fleet
ssh_config:
  user: dstack
  identity_file: ~/.ssh/dstack
  hosts:
    - 10.0.0.1
    - 10.0.0.2
$ dstack apply -f fleet.dstack.yml
...
Fleet ssh-fleet does not exist yet.
Create the fleet? [y/n]: y
...
 FLEET      INSTANCE  BACKEND       RESOURCES                PRICE  STATUS  CREATED
 ssh-fleet  0         ssh (remote)  cpu=4 mem=4GB disk=30GB  $0     idle    09:08
            1         ssh (remote)  cpu=2 mem=4GB disk=30GB  $0     idle    09:08

Then, if you update the hosts configuration property to

  hosts:
    #- 10.0.0.1  # removed
    - 10.0.0.2
    - 10.0.0.3  # added

and apply the same configuration again, the fleet will be updated in-place, meaning that you don't need to stop runs on the fleet instances if they are not affected by the changes (in this example, it's okay if the instance 1 is currenty busy, you can still apply the configuration).

$ dstack apply -f fleet.dstack.yml
...
Found fleet ssh-fleet. Configuration changes detected.
Update the fleet in-place? [y/n]: y
...
 FLEET      INSTANCE  BACKEND       RESOURCES                PRICE  STATUS  CREATED
 ssh-fleet  1         ssh (remote)  cpu=2 mem=4GB disk=30GB  $0     idle    09:08
            2         ssh (remote)  cpu=8 mem=4GB disk=30GB  $0     idle    09:12

Note

For in-place updates it's only allowed to add and/or remove instances, the root configuration and configurations of hosts that are not changed must not be changed, otherwise the full fleet recreation is triggered, as before. This restriction may be lifted in the future.

Volumes

Automatic cleanup of unused volumes

The volume configuration gets a new auto_cleanup_duration property:

type: volume
name: my-volume
backend: aws
region: eu-west-1
availability_zone: eu-west-1a
auto_cleanup_duration: 1h

The volume will be automatically deleted after it's not being used for the specified duration.

Logs

Browsable, queryable, and searchable logs

dstack now stores run logs in plaintext, which were previously base64-encoded. This allows you to use the configured log storage, be it AWS CloudWatch or GCP Logging, to browse and query dstack run logs.

Note

Logs generated before this release will be shown as base64-encoded in the UI and CLI after the update.

Server

Faster API response times

The dstack server API has been optimized to serialize json responses faster. The API endpoints are up to 2x faster than before.

Benchmarks

Benchmarking AMD GPUs: bare-metal, containers, partitions

Our new benchmark explores two important areas for optimizing AI workloads on AMD GPUs: First, do containers introduce a performance penalty for network-intensive tasks compared to a bare-metal setup? Second, how does partitioning a powerful GPU like the MI300X affect its real-world performance for different types of AI workloads?

What's Changed

Full Changelog: 0.19.18...0.19.19

Contributors

un-def, r4victor, and 4 other contributors
Loading

0.19.18

09 Jul 09:59
@r4victor r4victor
0.19.18
7435ecf
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.18

Server

Optimized resources processing

This release includes major improvements that allow the dstack server process more resources quickly. It also allows scaling processing rates of one server replica to take advantage of big Postgres instances by setting the DSTACK_SERVER_BACKGROUND_PROCESSING_FACTOR environment variable.

The result is:

  • Faster processing rates: provisioning 100 runs on SQLite with default settings went from ~5m to ~2m.
  • Better scaling: provisioning additional 100 runs is even quicker due to warm cache. Before, it was slower than the first 100 runs.
  • Ability to process more runs per server replica: provisioning 300 runs on Postgres with DSTACK_SERVER_BACKGROUND_PROCESSING_FACTOR=4 is ~4m.

For more details on scaling backgraound processing rates, see the Server deployment guide.

Backends

Private GCP gateways

It's now possible to create GCP gateways without public IPs:

type: gateway
name: example
domain: gateway.example.com
backend: gcp
region: europe-west9
public_ip: false
certificate: null

Note that configuring HTTPS certificates for private GCP gateways is not yet supported, so you need to specify certificate: null.

What's Changed

Full Changelog: 0.19.17...0.19.18

Contributors

un-def, olgenn, and 3 other contributors
Loading

0.19.17

02 Jul 12:14
@un-def un-def
0.19.17
669bf16
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.17

Secrets

dstack gets support for secrets that allow centralized management of sensitive values such as API keys and credentials. They are project-scoped, managed by project admins, and can be referenced in run configurations to pass sensitive values to runs in a secure manner. Example:

$ dstack secret set my_secret some_secret_value
OK
type: task
nodes: 1
name: test-secrets
env:
  - MY_SECRET=${{ secrets.my_secret }}
commands:
  - echo $MY_SECRET
$ dstack apply -f .dstack/confs/task.dstack.yaml

Submit the run test-task? [y/n]: y
 NAME            BACKEND         RESOURCES              PRICE   STATUS   SUBMITTED
 test-task       aws             cpu=2 mem=8GB          $0.107  running  10:48 
                 (eu-west-1)     disk=100GB                                   

test-secrets provisioning completed (running)
some_secret_value
Exited (0)

For more details on secrets, check out the docs.

Files

By default, dstack automatically mounts the repo directory where you ran dstack init to any run configuration.

However, in some cases, you may not want to mount the entire directory (e.g., if it’s too large), or you might want to mount files outside of it. In such cases, you can use the files property.

type: task
name: trl-sft

files:
  - .:examples  # Maps the directory where `.dstack.yml` to `/workflow/examples`
  - ~/.ssh/id_rsa  # Maps `~/.ssh/id_rsa` to `/root/.ssh/id_rsa`

python: 3.12

env:
  - HF_TOKEN
  - HF_HUB_ENABLE_HF_TRANSFER=1
  - MODEL=Qwen/Qwen2.5-0.5B
  - DATASET=stanfordnlp/imdb

commands:
  - uv pip install trl
  - | 
    trl sft \
      --model_name_or_path $MODEL --dataset_name $DATASET
      --num_processes $DSTACK_GPUS_PER_NODE

resources:
  gpu: H100:1

Warning

If you have existing fleets, it's recommended to re-create them after upgrading to version 0.19.17. Otherwise, there is a risk that these instances won't be able to execute jobs if the run uses files.

Services

Rolling deployment

Rolling deployments introduced in 0.19.15 are now supported when deploying new commits or branches from a Git repo, or when changes are made to the repo contents or files listed in the files section.

Additionally, dstack apply now displays a full list of detected changes:

$ dstack apply -f my-service.dstack.yml

Active run my-service already exists. Detected changes that can be updated in-place:
- Repo state (branch, commit, or other)
- File archives
- Configuration properties:
  - env
  - files

Update the run? [y/n]:

Even when a rolling deployment isn't possible, the list of changes is still shown — making it easier to identify which changes are preventing the deployment from proceeding in-place.

What's changed

  • [Bug]: Docker In Docker does not work with AMD by @peterschmidt85 in #2849
  • [Feature] Add files property to run configurations by @un-def in #2848
  • [Feature] Implement project secrets by @r4victor in #2854
  • [Internal] Support fleet configurations for the local backend by @jvstme in #2856
  • [Services] Rolling deployments for repo updates by @jvstme in #2853
  • [Internal] Fix package dependency direction by @jvstme in #2859
  • [Internal] Rolling deployments for files by @jvstme in #2862
  • [Internal] Support the local backend with the in-server proxy by @jvstme in #2858
  • [Docs] Added Files documentation by @peterschmidt85 in #2866
  • [Bug] Fix ~ expansion in files by @un-def in #2865
  • [Feature] Allow in-place update for more run properties by @jvstme in #2867

Full changelog: 0.19.16...0.19.17

Contributors

un-def, r4victor, and 2 other contributors
Loading

0.19.16

26 Jun 11:32
@r4victor r4victor
0.19.16
cd4dc99
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.16

Docker

Docker in Docker

Using Docker in a run configuration is now much easier. Just set docker to true:

type: task
name: docker-nvidia-smi

docker: true

commands:
  - docker run --gpus all nvidia/cuda:12.3.0-base-ubuntu22.04 nvidia-smi

resources:
  gpu: 1

This works with all run configuration types and supports both AMD and NVIDIA GPUs. It’s especially useful if you want to use the docker CLI in your commands—for example, to build Docker images.

The docker property is supported on all backends except vastai, runpod, and kubernetes, and is fully supported on SSH fleets as well.

Backends

CloudRift

The CloudRift team has added support for their GPU cloud, which can now be used with dstack.

To configure it, use a CloudRift API key in the backend configuration:

projects:
  - name: main
    backends:
      - type: cloudrift
        creds:
          type: api_key
          api_key: rift_2prgY1d0laOrf2BblTwx2B2d1zcf1zIp4tZYpj5j88qmNgz38pxNlpX3vAo

CloudRift offers competitive on-demand GPU pricing, with more GPUs and regions coming soon.

dstack apply -f examples/.dstack.yml -b cloudrift

 #  BACKEND                      RESOURCES                                    INSTANCE TYPE   PRICE
 1  cloudrift (us-east-nc-nr-1)  cpu=16 mem=100GB disk=1000GB RTX5090:32GB:1  rtx59-16c-nr.1  $0.65

If you encounter any issues with this backend, please report them.

Server

Public projects

You can now create public projects that any user on the server can join or leave without approval. Previously, all projects were private, and adding new members required manual action by an admin or manager—a step that’s redundant in high-trust environments.

Admins can change a project’s visibility at any time in the project settings.

Metrics

The server exports new Prometheus metrics:

  • dstack_submit_to_provision_duration_seconds: Time from when a run has been submitted and first job provisioning
  • dstack_pending_runs_total: Total number of pending runs

What's changed

New contributors

Full changelog: 0.19.15...0.19.16

Contributors

olgenn, Nadine-H, and 5 other contributors
Loading

0.19.15

19 Jun 20:49
@jvstme jvstme
0.19.15
c10adfb
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.15

Services

Rolling deployments

This update introduces rolling deployments, which help avoid downtime when deploying new versions of your services.

When you apply an updated service configuration, dstack will gradually replace old service replicas with new ones. You can track the progress in the dstack apply output — the deployment number will be lower for old replicas and higher for new ones.

> dstack apply -f my-service.dstack.yml

Active run my-service already exists. Detected configuration changes that can be updated in-place: ['image', 'env', 'commands']
Update the run? [y/n]: y

⠋ Launching my-service...
 NAME                            BACKEND          RESOURCES                        PRICE    STATUS       SUBMITTED
 my-service deployment=1                                                                    running      11 mins ago
   replica=0 job=0 deployment=0  aws (us-west-2)  cpu=2 mem=1GB disk=100GB (spot)  $0.0026  terminating  11 mins ago
   replica=1 job=0 deployment=1  aws (us-west-2)  cpu=2 mem=1GB disk=100GB (spot)  $0.0026  running      1 min ago

Currently, the following service configuration properties can be updated using rolling deployments: resources, volumes, image, user, privileged, entrypoint, python, nvcc, single_branch, env, shell, and commands.

Future releases will allow updating more properties and deploying new git repo commits.

Clusters

Updated default Docker images

If you don't specify a custom image in the run configuration, dstack uses its default images. These images have been improved for cluster environments and now include mpirun and NCCL tests. Additionally, if you are running on AWS EFA-capable instances, dstack will now automatically select an image with the appropriate EFA drivers. See our new AWS EFA guide for more details.

Server

Health metrics

The dstack server now exports some operational Prometheus metrics that allow to monitor its health. If you are running your own production-grade dstack server installation, refer to the metrics docs for details.

What's changed

New Contributors

Full Changelog: 0.19.13...0.19.15

Contributors

olgenn, Nadine-H, and 4 other contributors
Loading

0.19.13

11 Jun 10:33
@r4victor r4victor
0.19.13
0e31236
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.13

Clusters

Built-in InfiniBand support in dstack Docker images

The dstack default Docker images now come with built-in InfiniBand support, which includes the necessary libibverbs library and InfiniBand utilities from rdma-core. This means you can run torch distributed and other workloads utilizing NCCL, and they'll take full advantage of InfiniBand without custom Docker images.

You can try InfiniBand clusters with dstack on Nebius.

Built-in EFA support in dstack VM images

dstack switches to DLAMI as the default AWS GPU VM image from a custom one. DLAMI supports EFA out-of-the-box, so you no longer need to use a custom VM image to take advantage of EFA.

Server

GCS support for code uploads

It's now possible to configure the dstack server to use GCP Cloud Storage for code uploads. Previously, only DB and S3 storages were supported. Learn more in the Server deployment guide.

What's Changed

Full Changelog: 0.19.12...0.19.13

Contributors

olgenn, colinjc, and 2 other contributors
Loading

0.19.12

04 Jun 11:22
@r4victor r4victor
0.19.12
8732138
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.12

Clusters

Simplified use of MPI

startup_order and stop_criteria

New run configuration properties are introduced:

  • startup_order: any/master-first/workers-first specifies the order in which master and workers jobs are started.
  • stop_criteria: all-done/master-done specifies the criteria when a multi-node run should be considered finished.

These properties simplify running certain multi-node workloads. For example, MPI requires that workers are up and running when the master runs mpirun, so you'd use startup_order: workers-first. MPI workload can be considered done when the master is done, so you'd use stop_criteria: master-done and dstack won't wait for workers to exit.

DSTACK_MPI_HOSTFILE

dstack now automatically creates an MPI hostfile and exposes the DSTACK_MPI_HOSTFILE environment variable with the hostfile path. It can be used directly as mpirun --hostfile $DSTACK_MPI_HOSTFILE.

Below is the updated NCCL tests example.

CLI

We've also updated how the CLI displays run and job status. Previously, the CLI displayed the internal status code which was hard to interpret. Now, the the STATUS column in dstack ps and dstack apply displays a status code which is easy to understand why run or job was terminated.

Examples

Distributed training

TRL

The new TRL example walks you through how to run distributed fine-tune using TRL, Accelerate and Deepspeed.

Axolotl

The new Axolotl example walks you through how to run distributed fine-tune using Axolotl with dstack.

What's changed

  • [Feature] Update .gitignore logic to catch more cases by @colinjc in #2695
  • [Bug] Increase upload_code client timeout by @r4victor in #2709
  • [Bug] Fix missing apt-get update by @r4victor in #2710
  • [Internal]: Update git hooks and package.json by @olgenn in #2706
  • [Examples] Add distributed Axolotl and TRL example by @Bihan in #2703
  • [Docs] Update dstack-proxy contributing guide by @jvstme in #2683
  • [Feature] Implement DSTACK_MPI_HOSTFILE by @r4victor in #2718
  • [Feature] Implement startup_order and stop_criteria by @r4victor in #2714
  • [Bug] Fix CLI exiting while master starting by @r4victor in #2720
  • [Examples] Simplify NCCL tests example by @r4victor in #2723
  • [Examples] Update TRL Single Node example to uv by @Bihan in #2715
  • [Bug] Fix backward compatibility when creating fleets by @jvstme in #2727
  • [UX]: Make run status in UI and CLI easier to understand by @peterschmidt85 in #2716
  • [Bug] Fix relative paths in dstack apply --repo by @jvstme in #2733
  • [Internal]: Drop hardcoded regions from the backend template by @jvstme in #2734
  • [Internal]: Update backend template to match ruff formatting by @jvstme in #2735

Full changelog: 0.19.11...0.19.12

Contributors

olgenn, Bihan, and 4 other contributors
Loading

0.19.12rc1

28 May 21:41
@jvstme jvstme
0.19.12rc1
5d57e19
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.12rc1 Pre-release
Pre-release

What's Changed

  • Update gitignore logic to catch more cases by @colinjc in #2695

Full Changelog: 0.19.11...0.19.12rc1

Contributors

colinjc
Loading

0.19.11

28 May 09:52
@r4victor r4victor
0.19.11
25b977e
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
0.19.11

Runs

Replacing conda with uv

dstack's default Docker images now come with uv installed. Installing Python packages with uv can be significantly faster than with pip or conda. Here's for example, uv vs pip times for installing torch on GCP VMs:

# time uv pip install torch
...
real    0m32.771s
user    0m29.070s
sys     0m8.300s

# time pip install torch
...
real    2m26.338s
user    1m37.514s
sys     0m16.711s

To continue supporting pip, dstack now automatically activates a virtual environment with pip available.

conda is no longer included in dstack's default Docker images. If you need to use conda, it should be installed manually:

commands:
  - wget -O miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
  - bash miniconda.sh -b -p /workflow/miniconda
  - eval "$(/workflow/miniconda/bin/conda shell.bash hook)"

Plugins

Built-in rest_plugin

dstack gets support for a built-in rest_plugin that allows writing custom plugins as API servers, so you don't need to install plugins as Python packages.

Plugins implemented as API servers have advantages over plugins implemented as Python packages in some cases:

  • No dependency conflicts with dstack.
  • You can use any programming language.
  • If you run the dstack server via Docker, you don't need to extend the dstack server image with plugins or map them via volumes.

To get started, check out the plugin server example. The rest_plugin server API is documented here.

AWS

New CPU series

dstack now supports most recent AWS CPU VMs based on Intel Xeon Sapphire Rapids: M7i, C7i, and R7i. It also adds support for the burstable T3 family. Previously, only M5, C5 and t2.small CPU instances were supported.

Azure

New CPU series

dstack now supports most recent Azure CPU VMs based on Intel Xeon Sapphire Rapids: general purpose Dsv6 and memory optimized Esv6 series. Previously, only Dsv3, Esv4, and Fsv2 series were supported.

GCP

New CPU series

dstack now supports most recent GCP CPU VMs: C4, M4, H3, N4, N2. Previously, only E2 and M1 were supported.

Note that C4, M4, H3, N4 instances do not currently support Volumes since they require Hyperdisk support.

Examples

Ray+RAGEN

The new Ray+RAGEN example shows how use dstack and RAGEN to fine-tune an agent on multiple nodes.

Breaking changes

  • conda is no longer included in dstack's default Docker images.

Deprecations

  • Azure VM series Dsv3 and Esv4 are deprecated.

What's Changed

New Contributors

Full Changelog: 0.19.10...0.19.11

Contributors

Bihan, colinjc, and 4 other contributors
Loading