- Enhancement: Allow CE users to list public components and their versions.
- Enhancement: Update logs viewer to better handle logs from main containers in TFJob and PytorchJob.
- Enhancement: Use subdued color for stopped status.
- Enhancement: Use custom titles for all pages.
- Enhancement: Improve several guides and fix several issues.
- Enhancement: Reorganize quick-start.
- Enhancement: Several enhancements.
- Enhancement: Enable basic activities and analytics in all plans.
- Enhancement: Add logic to publish
polyaxon-cli:latestdocker image to the release pipeline.
- Enhancement: Improve container’s sanitization.
- Fix: Remove errors interceptions for API/Healthz/Streams.
- Enhancement: Add
MaxConcurrentReconcilesoption to allow concurrent reconciles, default is 1.
- Enhancement: Improve condition transition and update time reporting.
- Enhancement: Add handling for issues with init containers.
- Fix: Issue with unschedulable operations not reported correctly.
- Fix: Issue with pod errors not detected correctly.
- Fix: Issue with pending, error, warning phases not reconciled correctly.
- Fix: Issue with unsupported type error when a generic error is raised.
Deployment & Setup
- Enhancement: Expose
operator.maxConcurrentReconcilesHelm option with default value 1.
- Enhancement: Do not show edit button to users without edit rights.
- New: Add new wait time field:
- Wait time is the duration from the time a run is created until it starts, it includes creation, compilation, queueing, and scheduling.
- wait time + run duration represent the total lifecycle of a run.
- Add possibility to filter and sort by wait time.
- Enhancement: Improve resume and restart status reasons.
- Enhancement: Ensure unique init container names.
- New: Add pipeline progress to the run overview page.
- New: Show wait time on the run overview page and the comparison table.
- New: Add clones lineage tab.
- Enhancement: Consolidate all pipeline related information (children runtimes, concurrency, progress, …) in a new overview section.
- Enhancement: Improve run’s overview page.
- Enhancement: Hide project panels’ shortcuts on mobile and small screens.
- Enhancement: Move filter buttons’ icons to the left side to have consistent UX.
- Fix: Regression when resuming runs; the modal does not close.
- Fix: Regression with column switches for inputs/outputs not displayed immediately.
- Fix: Some rendering issues.
- Enhancement: Typos and several improvements.
- Enhancement: Improve exp backoff when queue state is full.
- Enhancement: Significant improvements to reduce wait time and operation processing.
- Fix: Regression when resuming a run when it’s part of a pipeline.
- Enhancement: Improve type checking and allow compatible types.
- Enhancement: Allow opting out of type checks by using the type
- Enhancement: Add check for last transition time when saving status conditions and respect chronology.
- Enhancement: Consolidate statuses reason.
- Enhancement: Add exception handling for non compiler errors.
- Fix: Regression in init artifacts container for mount paths.
- Fix: Bug with
container.env. Values containing
intare casted. e.g.
- Fix: Bug with
- Enhancement: Add cursor pointer for statuses and stages.
- New: Add new how-tos section to document recurrent use-cases and frequent questions about leveraging Kubernetes abstractions in Polyaxon.
- New: Add new guide about conditional scheduling.
- New: Add new guide about manual/human approval.
- New: Add new guide about operations caching.
- New: Add new guide about concurrency management.
- New: Add new guide about external invocation.
- New: Add new guide about manual volume, secrets, and configmaps mounting.
- New: Add new guide about service accounts usage.
- New: Add new guide about security context usage.
- New: Add more information about the compiler and variables resolution in the context section.
- Fix: Typos.
- Fix Move scheduled status logging to the agent controller.
- Enhancement: Allow requesting all inputs, outputs, artifacts as dictionaries without specifying values by keys.
- Enhancement: Improve handling of unexpected compilation errors, and set the operation to failed to notify the user.
- Enhancement: Allow to specify path from and path to in the init artifacts specification.
- Enhancement: Improve restart with copy mode, the artifacts are now initialized under the new run.
- Enhancement: Improve several APIs response time.
- Enhancement: Allow passing variables to the concurrency key in matrix algorithms and dag.
- Enhancement: Add
uuidtype parser and validator to the V1IO.
- Enhancement: Improve mount path for artifacts connections in init containers.
- Enhancement: Remove restriction on parameters with
- Enhancement: Use
-rfor recursive copy, a regression was introduced when the base image was updated.
- Fix: Issue in init container logs when the artifacts are coming from the default artifacts store.
- Fix: Typos in protos files.
- Enhancement: Use ellipsis,
..., when a filter is very long.
- Enhancement: Improve the overview headers to use a smaller fonts for titles/names to not break the page when the entity name is long.
- Enhancement: The comparison table shows a list of default columns based on the kind, for instance it will include the concurrency and runtime for Matrix runs.
- Enhancement: Improve edit filter by adding a spaces after
:and between operators like
- Enhancement: Improve and consolidate tables padding.
- Fix: Issue with search input not allowing spaces.
- Enhancement: Upgrade operator:
- Use latest stable apimachinery.
- Use latest stable k8s client-go.
- Use latest stable controller-runtime.
- Enhancement: Add new Tensorboard version with upstream lineage.
- Enhancement: Add new Tensorboard version with performance based join.
- Enhancement: Move notification component to component hub.
- Enhancement: Improve Slack notifier to send extra information, such as inputs, outputs, and custom information.
- Enhancement: Upgrade go version.
- Enhancement: Move context docs to a separate section.
- Fix: Typos.
- New: Add
JoinParamspecifications. As opposed to the
matrixsection which allows to parallelize operations using hyperparameter tuning or mapping, the
- to filter, aggregate, and annotate inputs/outputs/artifacts from multiple or parallel upstream runs based on query/sort/limit/offset specification.
- to initialize an operation with all artifacts from several upstream runs dynamically.
- New: Implement logic for internal
events. Users can now run full event based workflows, and in future iterations, it will be possible to start operations based on external events and internal alerts.
- New: Implement logic for
hooks. Hooks can be used:
- to send webhooks
- to send notifications and alerts to slack, discord, github, …
- to integrate with external systems
- to initiate a test/eval operation after training.
- New: Add new lineage tabs for: connections, actions & hooks, upstream/downstream runs.
- Enhancement: Expose namespace and artifacts store connection used on the run overview page.
- Enhancement: Allow filtering runs by the artifacts store connection name used.
- Enhancement: Use
statusand other general values from upstream operations, similar to normal operation context to have more consistent interface.
- Enhancement: Add several heartbeat processes to clean distressed pipelines and runs.
- Enhancement: Improve auditing and activity logs.
- Enhancement: Optimize events access to the database.
- Enhancement: Mark operations with false conditions as
- Enhancement: Improve the run edge collection logic to conserve metadata, types, and reference.
- Enhancement: Improve the agent controller’s querying logic and traversal speed.
- Enhancement: Improve schedules reconciliation by flagging all runs on schedule to avoid race conditions.
- Enhancement: Improve connections compilation.
- Fix: Issue with pipelines getting two final states (
- Fix: Issue with agent controller not scheduling DAGs and Matrix when a schedule is attached.
- Fix: Issue when a schedule is attached to a matrix, the metadata is missing and the matrix icon not showing in the sidebar.
- Fix: Issue with schedule not stopping properly and keep scheduling new runs.