In October of this year we released alchemiscale v0.7.0, and followed this up with v0.7.1 and v0.7.2. We wanted to take a moment to highlight these releases, and the improvements they bring for users!
v0.7.0#
As a major release, v0.7.0 introduced two significant features that advance alchemiscale’s capabilities for automated, large-scale execution.
First, this release introduced the ComputeManager base class, which lays the foundation for upcoming compute autoscaling implementations.
This abstraction enables alchemiscale to dynamically scale compute resources across different platforms, including HPC clusters and Kubernetes environments.
With this foundation in place, compute services can now be managed more intelligently, responding to workload demands without manual intervention.
Keep an eye out for alchemiscale-k8s, which leverages this new capability!
Second, we added the Strategist service, which enables automated execution of Strategy objects for AlchemicalNetworks.
A Strategy proposes where to apply additional compute given the set of results currently available for a given AlchemicalNetwork.
Users can now assign a Strategy to any of their AlchemicalNetworks, and the Strategist service will periodically apply it server-side to adaptively sample Transformations until the Strategy’s stop thresholds are satisfied.
This release also delivered substantial performance improvements for AlchemicalNetwork submission and retrieval via the AlchemiscaleClient.
We optimized how AlchemicalNetwork objects are submitted to the state store, and implemented direct KeyedChain retrieval from Neo4j, making these operations notably faster for large networks.
Finally, v0.7.0 resolved deprecation issues affecting Python 3.12+ (specifically utcnow) and upgraded to Pydantic v2, ensuring alchemiscale remains compatible with modern Python environments.
v0.7.1#
Release v0.7.1 was a bugfix release addressing issues found in v0.7.0, while also improving developer experience.
A critical bug affecting the Strategist service startup process was resolved, ensuring the service initializes correctly in deployed environments.
We also added the ability for compute managers to specify service names via command-line arguments, giving operators more flexibility when deploying compute services across different environments.
On the development side, we implemented parallel Neo4j container testing, significantly improving test suite performance. This makes it faster for contributors to run the full test suite during development. We also enhanced our developer documentation with guidance on Docker setup for integration testing, making it easier for new contributors to get started.
v0.7.2#
Release v0.7.2 was an incremental release focused on bugfixes and improvements in support of autoscaling implementations, such as that featured in alchemiscale-k8s.
Key improvements included updating Traefik to v2.11 for better routing performance and security, fixing scope-related issues in the compute manager, and making adjustments to ComputeManagerID validation.
We also added the Strategist service to the docker-compose deployment configuration, making it easier to deploy complete alchemiscale instances with automated strategy execution capabilities.
what’s next?#
We are currently hard at work on the major new features coming in v0.8.0, so keep a look out for this release in the coming months!
If you are interested in trying alchemiscale out, or if you already have and want to offer ideas for improvement, please reach out! Posting in our Discussions forum is the best way to get started!