ZenML Release Notes

New Databricks Step Operator

You can now run individual pipeline steps on Databricks using the new Databricks step operator PR #4648. This is useful when you want specific steps to execute in the Databricks runtime while the rest of your pipeline uses a different orchestrator. The Databricks orchestrator also now supports optional tag settings to label jobs and cluster resources for cost tracking, ownership, and governance.

Nested Dynamic Pipelines

Dynamic pipelines can now be nested, allowing you to call one dynamic pipeline from within another PR #4775. This enables more modular and reusable pipeline designs.

Enhanced Run:AI Training Workload Configuration

The Run:AI step operator now supports advanced training workload settings PR #4780, including:

• Multiple mount types (PVC, ConfigMap, Secret, NFS, S3, HostPath)
• Workload templates via workload_template_id
• Security context settings (UID/GID, non-root execution, seccomp, capabilities)
• Port declarations and external URL exposure
• Training workload parallelism and completions

Improved Kubernetes Job Failure Diagnostics

When dynamic pipeline jobs fail due to system issues (such as OOM kills), ZenML now provides richer diagnostic information PR #4800. This makes it easier to understand why Kubernetes terminated your pods.

Better Kubernetes Label Handling

Kubernetes string handling has been improved with separate sanitization for DNS-style names/keys and looser label-value rules for metadata like run, pipeline, and step IDs PR #4756. This makes it easier to navigate through runs in Kubernetes.

Increased Secret Size Limit

The maximum allowed size for ZenML secrets stored in the SQL secrets store has been increased to 64KB PR #4769. The limit applies to the combined size of all keys and values in a secret object.

Dashboard: Parent Run Display

The dashboard now displays parent run information in the run details view when available PR #1050.

Fixed

Signal handling during step execution: Signal handlers are now properly unregistered after step execution, preventing strange errors when running many sync steps or interrupting pipelines PR #4784. Dynamic pipeline steps running in isolated environments or step operators are no longer affected by signal handling from the orchestrator.

RBAC performance: Fixed redundant RBAC permission checks during response model dehydration PR #4797. Previously, when permissions were already prefetched and denied, additional RBAC requests were sent unnecessarily for each sub-model.

Keyword-only arguments in steps: Step functions can now use keyword-only arguments without causing failures PR #4798.

Kubernetes orchestrator settings: Fixed step pod configuration by replacing legacy hardcoded orchestrator.kubernetes lookups with proper orchestrator.get_settings(...) calls PR #4803. Step pods now correctly apply orchestrator settings from canonical component keys while maintaining backward compatibility.

What's Changed

• Add version 0.94.2 to legacy docs by @github-actions[bot] in #4766
• Add 0.94.3 to the migration tests by @github-actions[bot] in #4768
• Revise Run:AI service account setup and troubleshooting by @safoinme in #4772
• Update dynamic pipelines docs by @schustmi in #4776
• Misc release-testing fixes by @schustmi in #4770
• Increase the max size of zenml secrets stored in the SQL secrets store by @stefannica in #4769
• Create run metadata for isolated steps (dynamic) by @Json-Andriopoulos in #4785
• Better signal handling during step execution by @schustmi in #4784
• Bump the minor-and-patch group across 1 directory with 9 updates by @dependabot[bot] in #4795
• Fix O(N) global scan in artifact-version delete by @joaquinhuigomez in #4686
• Fix multiple RBAC calls during dehydration by @schustmi in #4797
• Clarify that DynamicPipelineRunner.wait() timeout is a soft timeout by @shaun0927 in #4741
• Fix typo in step creation error message by @shaun0927 in #4736
• Allow keyword only args in step function by @schustmi in #4798
• Kubernetes label sanitization by @bcdurak in #4756
• Allow non module-level definitions of steps and dynamic pipelines by @schustmi in #4799
• Added ability to get help with flavor specific configurations by @AlexejPenner in #4662
• Extract projects-backend release dispatch into dedicated workflow by @safoinme in #4720
• Fixing orchestrator setting resolution by @bcdurak in #4803
• Fix notify projects backend zizmor scan by @strickvl in #4808
• Fix Check Links workflow failures (runai.md + dangling preemption refs + Slack 403) by @strickvl in #4807
• Fix nightly PyPI trusted publishing by @strickvl in #4806
• Add uv.lock to .gitignore by @strickvl in #4811
• Nested dynamic pipelines by @schustmi in #4775
• Refresh agent guidance docs by @strickvl in #4817
• Fix MLflow deployment healthcheck method by @strickvl in #4816
• Update docs for recent feature drift by @strickvl in #4818
• Add advanced Run:AI training workload settings by @safoinme in #4780
• Fix deployment invocation request fields by @schustmi in #4809
• Extra info on kubernetes-killed jobs by @Json-Andriopoulos in #4800
• Introduce new Databricks step-operator and fix tags and custom-tags by @safoinme in #4648
• Fix PydanticAI agent example dependencies by @strickvl in #4823
• Refactor Databricks test cases to use temporary path for wheel directory by @safoinme in #4824
• Prepare release 0.94.4 by @github-actions[bot] in #4821

New Contributors

• @joaquinhuigomez made their first contribution in #4686
• @shaun0927 made their first contribution in #4741

Full Changelog: 0.94.3...0.94.4

What's Changed

• Add version 0.94.1 to legacy docs by @github-actions[bot] in #4701
• Bump the minor-and-patch group with 2 updates by @dependabot[bot] in #4705
• Fix zizmor SHA/version mismatch and add zizmor to lint.sh by @strickvl in #4703
• Fix broken docs header - triggers by @Json-Andriopoulos in #4707
• Misc kitaru extensions by @schustmi in #4715
• Introduce a container engine abstraction and add podman support by @stefannica in #4651
• Include wait condition resolved at in DAG response by @schustmi in #4717
• Replace darglint with pydoclint for docstring linting by @strickvl in #4721
• Bump the minor-and-patch group with 4 updates by @dependabot[bot] in #4725
• Fix empty PR number in Claude Code review workflow by @strickvl in #4735
• Resource pools by @schustmi in #4465
• Feature/event triggers by @Json-Andriopoulos in #4692
• Helper endpoint: Trigger SourceType supported events by @Json-Andriopoulos in #4745
• Don't resolve zenml pypi version on flavor model generation by @stefannica in #4744
• Document Kubernetes permissions for various roles in running k8s pipelines by @stefannica in #4746
• Add docs for on-prem deployment of optional Pro features by @stefannica in #4751
• Remove the unique constraint from the LogsSchema by @bcdurak in #4729
• Add memory_call step type by @strickvl in #4724
• Add acknowledgeable trigger dispatch status and improve CLI commands by @stefannica in #4743
• Support for max-runs stop criteria by @Json-Andriopoulos in #4752
• Fix making a secret public using CLI/client by @schustmi in #4755
• Feature/add k8s gateway support by @safoinme in #4726
• Fix run status update condition by @schustmi in #4747
• Fix broken tests by @Json-Andriopoulos in #4758
• Fix StrEnum import breaking 3.10 by @Json-Andriopoulos in #4763
• Don't block while waiting for concurrent step inputs by @schustmi in #4699
• Fix python package fetching by @schustmi in #4762
• Allowing multiple components of the same type for a stack by @bcdurak in #4671
• Server-side replays by @schustmi in #4716
• Fix pydantic materializer double encoding by @schustmi in #4760
• Prepare release 0.94.3 by @github-actions[bot] in #4765

Full Changelog: 0.94.2...0.94.3

🎨 Dashboard Enhancements

The ZenML dashboard now includes a Run Summary View that provides a comprehensive overview of your pipeline runs at a glance PR #1029. Timeline rows now automatically resize for better visualization of your pipeline execution history PR #1028.

🔧 Pipeline & Trigger Improvements

Artifact Name Substitutions for Dynamic Pipelines: You can now use artifact name substitutions in dynamic pipelines, making it easier to reference and manage artifacts programmatically PR #4668.

Enhanced Trigger Configuration: Triggered runs now follow a cleaner snapshot.source pattern. You can provide a configuration object to customize parameters for all triggered runs in an attachment, giving you more control over automated pipeline executions PR #4610.

⚠️ Deprecation Notice

Helm Chart Configuration Update: The top-level zenml: values key in the ZenML Helm chart has been renamed to server: PR #4637. Your existing values files will continue to work — the chart automatically merges both keys for backwards compatibility. If both keys are present, zenml: takes precedence for overlapping fields. The zenml: key is deprecated and will be removed in a future release; we recommend migrating to server: in your Helm configurations.

Fixed

Schedule Management: Fixed issues where deletion by ID prefix wasn't working for archived schedules, and reusing an archived schedule name was incorrectly blocked. Documentation has been updated to clarify schedule behavior post-deletion, including archived schedule renaming, hard deletion guidelines, and discovery of archived objects PR #4641.

Input Artifact Handling: Resolved a bug where a list of input artifacts containing a single item was incorrectly treated as a scalar input artifact, causing materializer type incompatibility errors PR #4679.

HTTP Request Optimization: The system now dynamically reduces chunk size when fetching many steps to avoid HTTP 414 (URI Too Long) errors PR #4553.

Dashboard Deployment Stability: Fixed a template loading issue in the FastAPI deployment app's dashboard that could cause TypeError: unhashable type: 'dict' on certain environments (e.g., Kubernetes with newer Jinja2 versions), preventing the server from starting when the dashboard was enabled PR #4649.

What's Changed

• Add version 0.94.0 to legacy docs by @github-actions[bot] in #4630
• Add 0.94.1 to the migration tests by @github-actions[bot] in #4632
• Optuna Hyperparameter Tuning with ZenML by @amitvikramraj in #4545
• Renamed helm chart top-level key from zenml to server, maintains backwards compatibility by @amitvikramraj in #4637
• Hydra Config Management with ZenML by @amitvikramraj in #4544
• Bugfix/archived schedule issues by @Json-Andriopoulos in #4641
• Follow snapshot.source pattern for triggered runs by @Json-Andriopoulos in #4610
• Only group minor/patch dependabot action updates by @strickvl in #4653
• Bump core GitHub Actions to latest major versions by @strickvl in #4655
• Bump minor/patch GitHub Action versions by @strickvl in #4654
• Dynamically reduce chunk size if running into 414 error by @schustmi in #4553
• Bump cloud provider GitHub Actions to latest major versions by @strickvl in #4656
• Bump third-party GitHub Actions to latest major versions by @strickvl in #4657
• Bump the all-actions group across 1 directory with 7 updates by @dependabot[bot] in #4666
• Fix incorrect version comments on SHA-pinned GitHub Actions by @strickvl in #4667
• Configurable dashboard files directory by @schustmi in #4660
• Add extra analytics coverage by @strickvl in #4677
• Add Kitaru mention to README by @strickvl in #4688
• Allow artifact name substitutions for dynamic pipelines by @schustmi in #4668
• Fix duplicate 'the the' typos in docstrings and log messages by @strickvl in #4681
• Fix broken filter snapshots by trigger by @Json-Andriopoulos in #4693
• Differentiate scalar input artifacts against single item lists by @schustmi in #4679
• Fix deployer dashboard HTML when Jinja2/Starlette template loading breaks by @stefannica in #4649
• Add supply chain security cooldowns for Dependabot and uv by @strickvl in #4659
• Prepare release 0.94.2 by @github-actions[bot] in #4696

New Contributors

@amitvikramraj made their first contribution in #4545

Full Changelog: 0.94.1...0.94.2