N.B.: This release is WIP, and it’s not published yet.
V2.0.0
CLI
- New: Command to start new local sandbox
plx sandbox
. - New: Command to render local run package based on path
plx ops dashboard --offline --path path/to/offline -uid UUID
. - New: New environment variable
expot POLYAXON_HOME=path/to/context
or config settingpolyaxon config set --home=path/to/context
to allow configuring multiple contexts for managing global configurations. - New: Remove
eager
mode in favor oflocal
mode. - Enhancement: Improve
plx ops restart/resume
with a new--recompile
flag to allow passing a full polyaxonfile specification instead of an override. - Enhancement: Allow resuming and overriding the operation’s name, desc, and tags.
- Enhancement: Improve CLI errors handling and printing.
- Enhancement: Improve logging to simplify debugging and detecting issues.
- Enhancement: Improve cli
plx run -w
andplx ops statuses watch --watch
to print the correct timestamps. - Enhancement: Improve cli
plx ops
commands with offline flag to consume persisted runs. - Enhancement: Improve cli
plx admin
deploy and upgrade commands. - Fix: Issue with
plx ops run -l
andplx ops logs
for distributed runs.
Core
- Deprecate:
isList
, users should useList[...]
instead. no breaking changes. - Deprecate:
options
, users should usevalidation
instead. no breaking changes. - New: Expose full type hints on polyaxonfile specification inputs/outputs definition.
- New: Add new
set_readme
method on run client. - New: Add new notifications/alerts/events layer to the clients to allow sending mid-run events to external systems.
- New: Add new
OfflineRunClient
that allows writing, reading, and interacting with persisted runs. - New: Add new base
AsyncAPIClient
. - New: Always snapshot a temp logs version to prevent complete loss of logs in case of preemption or unhandled crashes.
- New: Add logic to enforce outputs validation.
- New: Add logic to support multiports for services.
- Breaking: Drop support for all python versions below 3.7.
- Enhancement: Make all libraries compatible with Python 3.11.
- Enhancement: Rework the schema and validation layer based on
pydantic
. - Enhancement: Improve typing and type hints.
- Enhancement: Add validation and auto-truncation for container names to prevent issues with Kubernetes 63 characters limit.
- Enhancement: Increase artifact’s name length to allow longer names.
- Enhancement: Make logs collection internal.
- Enhancement: Improve deprecation and warning logic.
- Enhancement: Update file system backends (GCS, Azure, S3, Minio, NFS, …).
- Enhancement: Improve compatibility logic and warning about CLI versions ahead of server versions.
- Enhancement: Improve inspection to include services definitions.
- Enhancement: Speed up events loading, saving, and streaming logic.
- Enhancement: Improve artifact deletion for runs managed manually.
- Enhancement: Add support for
base
in log space generator, default base is10
. - Enhancement: Move sdk inside main package and remove unneeded dependency.
- Enhancement: Improve fs watcher and avoid duplicate memory loading of mappings.
- Enhancement: Add
is_requested
indicator toParamSpec
. - Enhancement: Improve validation for mapping values context reference.
- Enhancement: Add check for additional fields for bulk insertion.
- Enhancement: Migrate
is_managed
tomanaged_by
to allow more flexible runs management. - Enhancement: Improve deletion of non managed runs to ensure that artifacts are cleaned from the artifacts store.
- Enhancement: Add auto-collecting of operation specification.
- Enhancement: Improve query manager with default filters.
- Enhancement: Update projects with stats and last update based on the last runs creation.
- Fix: Fix issue with params not updated with resume override config.
- Fix: Issue to avoid nesting folders when updating an uploaded files/directories.
- Fix: Issue with
acquisition_function
with new scipy versions.
Agent
- New: Allow multi-namespace management.
- New: Add support for managing multiple artifacts stores using a single deployment.
- New: Add support for multi-connections management and mounting of multiple volumes to upload and download artifacts to and from connections other than the artifacts store.
- New: (Beta) Add cluster and namespace monitoring:
- Show nodes’s states and health.
- Show nodes’s CPU/Memory/GPU consumption.
- New: Expose Dask runtime to all distributions.
- New: Expose Ray runtime to all distributions.
- New: Add monitoring for agent pods to quickly debug agent issues.
Tracking
- New: Add spaCy tracking callback.
- New: Add Prophet tracking callback.
- New: Expose
set_readme
method on tracking module. - New: Expose
update
method on tracking module. - Enhancement: Update PytorchLightning tracking callback.
- Enhancement: Improve debugging experience by showing more informative logs when API returns 403/401 errors.
- Enhancement: When logging a final state for a non-managed operation, the tracking module should automatically collect and end the operation without the need to manually call
.end()
. - Enhancement: Upgrade pynvml’s version for GPU tracking.
- Enhancement: Improve log processing and streaming for run managed manually.
- Enhancement: Improve tensor/array sanitization.
- Enhancement: Add support for Pandas-2.0 tracking.
- Enhancement: Fix PytorchLightning tracking callback.
UI
- New: Add metrics/params performance widget to dashboards.
- New: Add Metrics/Params correlation and importance.
- New: Add dashboard level configuration to set smoothing and sizes.
- New: Add logs and resources view to quickly check health of agent pods.
- New: Add possibility to inspect runs for finished operations to better debug errors.
- Enhancement: Add teams filter to projects, runs, models, artifacts, components table.
- Enhancement: Add stages filter to models, artifacts, components table.
- Enhancement: Improve
restart/resume
override config or full polyaxonfile specification. - Enhancement: Allow resuming and overriding the operation’s name, desc, and tags.
- Enhancement: Improve UX by adding accordions to the inspection sections and containers.
- Enhancement: Rename runs’ “artifacts” tab to “files” to reduce confusion.
- Enhancement: Show all containers logs by default.
- Enhancement: Add global step and smoothing management on dashboard level.
- Enhancement: Rename run’s sidebar artifacts tab to files.
- Enhancement: Improve logs streaming for manually managed runs.
- Enhancement: Improve managed by metadata.
- Enhancement: Update visualization packages: plotly, bokeh, vega/altair.
- Enhancement: Update vendor packages
- Enhancement: Improve graph view rendering.
- Enhancement: Allow hiding sub-paths in the graph view.
- Enhancement: Speed up loading and rendering.
- Enhancement: Add a note to the concurrency update modal to explain the behavior when increasing or decreasing the concurrency of a pipeline or a matrix.
- Enhancement: Add a note to runs transfer modal to prevent transferring active runs.
- Enhancement: Improve collapse/expand logic to freeze the sidebar without an additional click.
- Enhancement: Remove the unintuitive lock/unlock button in the sidebar and automatically remember user’s choice.
- Enhancement: Improve hovered/selected nodes in graphs.
- Enhancement: Improve performance of runs pages.
- Enhancement: Improve performance of activity-logs pages.
- Enhancement: Improve performance of analytics pages.
- Enhancement: Improve UI for dates filters.
- Fix: Edge hover color for ops in warning status.
- Fix: Issue reloading the table after a multi-run deletion when the runs count is less than 10.
Operator
- New: Update go version to 1.19 and kubebuilder version to 3.9.0.
- New: Add support for PaddlePaddle operator.
- New: Add support for Ray operator.
- New: Add support for Dask operator.
- Enhancement: Update training operator to latest version.
- Enhancement: Better handling of distributed training jobs.
- Enhancement: Improve reconciliation logic.
- Fix: Issue with race conditions.
Integrations
- New: Add gradio integration.
- New: Add mlflow integration.
Docs
- New: New guide to get started with the local sandbox server.
- New: New deployment guides using:
- pip
- Docker
- Docker compose
- Kubernetes
Deployment
- Breaking: Remove
compressedLogs
on agent config section. - New: Add support for agent based deployment in the community edition (CE) to support:
- Batch deletion
- Batch stopping
- Better cleaning of resources
- Beta: Add new sandbox debugger for running and visualizing on a local machine without any docker requirement.
- Beta: Allow rendering a local run package based on the offline flag or pulled using the CLI.
- New: Upgrade platform chart dependency to use postgres/redis/rabbitmq from bitnami.
- New: Refactor database connector to allow using Postgres and SQLite.
- New: Refactor proxies to use a single service to expose Gateway, API, and streams endpoint.
- New: Allow deployment and configuring multiple Polyaxon deployments and gateway accesses.
- Enhancement: Allow passing custom labels, annotations, priorityClassName to the Helm charts.
- Enhancement: Improve speed of several core components.
- Enhancement: Upgrade Python to v3.11 for core component images.
- Enhancement: Cleanup charts repository and remove all deprecated packages.
- Security: Fix CVE issues.
Commercial
- New: Project and organization info pages redesign.
- New: Add team views (workspaces).
- New: Add a new tab to explore unregistered artifact/component/model versions under each project.
- New: Add cross-projects artifact/component/model versions view similar to
All Runs
. - New: Add control for archive deletion interval per organization and per project.
- New: Add stats for the number of runs, artifacts, and models per project and per organization.
- New: Add logic to show active contributors per run, model, artifact, project and per organization.
- Enhancement: Improve deletion logic.
- Enhancement: Improve runs table speed and performance.
- Enhancement: Improve resuming or restarting an operation that is part of a pipeline.
- Enhancement: Improve reassignment of the main organization owner.
- Enhancement: Improve cleanup hooks.
- Enhancement: Add stats calculation and caching for projects and organizations.
- Enhancement: Make histograms the default analytics view.
- Enhancement: Improve archive and deletion logic.
- Enhancement: Always set updated_at on archive/restore logic.
- Fix: Issue with transferring pipeline or matrix runs between projects.
- Security: Fix CVE issues.