Dagster-cloud

New

- Added a new Users tab on the Cloud Settings page with a number of permissions improvements:

- A way to see all the permissions for each user and deployment in a single place
- A way to set permissions for all branch deployments (there's a separate "Branch deployments" column in the UI that can set its own permissions)
- A new Launcher role that can launch and terminate runs and backfills, but can take no other writes (Only available on Enterprise plans)
- Setting permissions per code location within a deployment (Only available on Enterprise plans)

This new tab replaces the old Permissions tab. See [the docs](https://docs.dagster.io/master/dagster-cloud/account/managing-users) for more information.

- Added Gitlab integration for adding code locations to Serverless deployments

New

- The ECS Agent now allows you to set the `runtimePlatform` for task definitions of the tasks that it creates, allowing you to create tasks using a Windows Docker image.
- Additional information about the code servers and run workers managed by an agent can be viewed in the agent statuses section.

Bugfixes

- Fixed a GraphQL error that occurred when a freshness policy sensor ran with a time-partitioned asset
- Fixed a `dagster-cloud` GitHub workflow bug to gracefully handle the case when the commit author is not a GitHub user.

New

- ECS Hybrid agents can now change the IAM roles used to spin up tasks for different code locations within the same deployment. See [the docs](https://docs.dagster.io/dagster-cloud/deployment/agents/amazon-ecs/configuration-reference#amazon-ecs-agent-configuration-reference) for an example of how to set the IAM roles for a particular code location in a `dagster_cloud.yaml` file.

Breaking Changes

- The `dagster-cloud serverless build-python-executable` command now automatically falls back to using a docker environment to build the Python dependencies, if the current environment cannot build the dependencies. This may happen when some packages do not publish a compatible Linux wheel in PyPI. The `--build-in-linux-docker` flag has been removed and replaced with a `--build-method` flag. This can be be set to `docker-fallback` (default), `local` to use the local environment only, or `docker` to only use the docker environment and skip the local environment.

New

- Alerts can now be configured to notify on code location load error. Just create a new alert policy from the UI, and the option will be available.

Bugfixes

- Fixed an issue where Serverless code location updates sometimes timed out while using the ENABLE_FAST_DEPLOYS flag.
- Fixed an issue where copying permissions from a deployment with only organization admins when creating a new deployment would raise an error message.

New

- Alerts can now be configured for Agent downtime on hybrid deployments. Check out the alerting documentation to learn more: https://docs.dagster.io/dagster-cloud/account/setting-up-alerts#setting-up-alerts-in-dagster-cloud.
- Users with Editor permissions can now edit deployment settings within the Dagster Cloud UI. Previously, they could only edit deployment settings using the dagster-cloud CLI.
- Serverless runs are now isolated by default. Isolated runs receive more memeory and compute, but take longer to start. https://docs.dagster.io/dagster-cloud/deployment/serverless#run-isolation

Bugfixes

- Fixed a bug where schedule / sensor alerts would not fire upon re-enabling an alert policy.

New

- Significant performance improvements for the asset reconciliation sensor when running on large asset graphs.
- [dagit] The Permissions and Alerts sections of Cloud Settings have been moved to the Deployment section of the app.

Bugfixes

- Some issues in `dagster-cloud serverless deploy-python-executable` with handling URLs and comments in `requirements.txt` have been fixed.