v0.5.6

  • Allow code download based on commit in CLI: polyaxon project download --commit=...
  • Allow code download related to an experiment/job/build in CLI: e.g. polyaxon experiment -xp 123 code
  • Fix issue with experiment resume cleaning the experiment’s outputs.
  • Fix some docs refs and issues.

v0.5.5

  • Fix regression in last metric saving (Last metric should always output the last metric from all previous steps).
  • Fix regression registering wrong queue for handling remote logs.
  • Fix regression order-by metrics does not proxy to the correct field.
  • Fix issue in Polystores: GCS might include empty blob names.
  • Add eventual pod restart handling for init process.
  • Update deps with security release.
  • Add docs for elastic stack config for on-prem deployments.

v0.5.4

  • Increase internal API throttling.
  • Make throttle rates configurable.
  • Expose more verbose uwsgi logging for DEBUG and INFO levels.
  • Use monospaced font and correct whitespace in logs viewer.
  • Update notebook spawner to allow downloading artifcats/code: set allow_origin to ”*“.
  • Expose ingress’s path configuration.
  • Remove wrong order check on HP tuning Matrix schemas, should be applied only to ranges and not distributions.
  • Add platform version to footer.
  • Fix bool option’s UI dropdown not handling initial state correctly.
  • Fix issue with security context not reflected in the dockerfile template.
  • Fix issue with init container not setting auth context correctly.
  • Fix SSO docs.
  • Fix impersonate urls typos (API and clients).

v0.5.3

  • Upgrade python to 3.7.4.
  • Update default kaniko image tag to v0.9.0.
  • Add note about using pgbouncer for scaling connexions to DB.
  • Add a conf caching mechanism to prevent hitting the db frequently when requesting config options.
  • Update default admin pages for builds, jobs, experiments, groups, tensorboards, notebooks.
  • Fix issue with registry access update form.
  • Add better handling for refresh auth tokens.
  • Make more defensive confirmation dialog by emphasising entities and actions.
  • Make project deletion harder by requiring the user to enter the project name.
  • Add more CLI checks for dependencies during deployment.
  • Make code path resolving consistent between local run and platform run.
  • Expose maxConnAge configuration option for both in-cluster and external postgresql connexion.
  • Force user to re-login when token expires.
  • Make stream API internal.
  • Internal improvements.

v0.5.2

  • Add asset version context processor.
  • Remove settings dropdown condition on admin interface.
  • Update health and status checks urls to follow best practices.
  • Fix jobs artifacts logging.
  • Fix dockerfile path resolution for local runs.
  • Add a simple way to inject debug mode command (only CLI side).
  • Fix issue displaying artifacts for S3 stores.
  • Add docs for using Minio for logs/data/artifacts.

v0.5.1

  • Add possibility to specify pod annotations for jobs/experiments/builds/notebooks/tensorboards.
    • By default polyaxon will inject necessary annotations, e.g. when using TPU.
    • Users can now define default annotations for each primitive cluster wide.
    • Polyaxonfile spec allows to override the annotations per run.
  • Add possibility to specify custom pod labels for jobs/experiments/builds/notebooks/tensorboards.
    • By default polyaxon uses recommended k8s labels for all managed resources.
    • Users can now define default custom labels for each primitive cluster wide.
    • Polyaxonfile spec allows to override the labels per run.
    • N.B. The custom labels cannot override Polyaxon’s required labels.
  • strengthen statuses check and fall back to db check if keys are evicted.
  • Disable namespace monitoring by default.
  • Disable containers resources monitoring by default.
  • Expose several celery options by worker type.
  • Force rabbitmq confirmation when used as a broker by default.
  • Add gzip to list APIs by default
  • Remove parts requiring privileged mode when deploying Polyaxon.
  • Fix in-cluster redis node scheduling docs: uses master/slave.
  • Fix Helm chart validation when disabling docker-registry.
  • Fix issue detecting some local configs when running polyaxonfiles locally (use_https was not detected correctly).
  • Fix quick creation modes in UI.
  • Fix route for creating Tensorboard in UI.
  • Fix UI issues noticed in offline deployment: self-host all styling requirements.
  • Fix spelling in UI.
  • Fix polyaxonfile spec unable to handle quotes in commands.
  • Update lodash: vulnerability issue.

v0.5.0

  • Add “Cluster Level Dynamic Configuration”:
    • Move scheduling configuration to UI.
    • Move integrations to UI.
    • Move secrets and config maps catalogs to UI.
    • Move github/gitlab/bitbucket/azure authentication setup to UI.
    • Move private repos setup to UI.
    • More hardware accelerator setup to UI.
    • Add possibility to create an access catalog for registries and git repos.
    • Add possibility to create an access catalog for connections, data, logs, outputs.
    • Move kaniko, default tensorboards config, default notebooks config to UI.
  • Introduce security group with optional and configurable user UID and group GID for all Polyaxon services and scheduled runs.
  • Mount all volumes with the proper security context.
  • Add a warning when code is big.
  • Add possibility to specify a service account for jobs/experiments/builds/notebooks/tensorboards.
    • By default polyaxon uses a worker service account.
    • Users can now use a default service account for each primitive cluster wide.
    • Polyaxonfile spec allows to override the service account per run.
  • Add possibility to specify a default resources to use for jobs/experiments/builds/notebooks/tensorboards.
    • By default polyaxon schedules runs without resources.
    • Users can now set default resources to use for each primitive cluster wide.
    • Polyaxonfile spec allows to override the resources per run.
  • Add pod restarts.
    • By default polyaxon schedules runs without pod restarts.
    • Users can now set a default pod restarts to use for each primitive cluster wide.
    • Polyaxonfile spec allows to override the pod restarts per run.
  • Add minimum resources requirement for injected containers (init and sidecar).
    • This will fix the issue in clusters with pod security policy requiring that all pods must specify resources.
  • Update redis chart and support using external/managed redis instance.
  • Update rabbitmq chart and support using external/managed rabbitmq instance.
  • Update docker-registry chart and support using external/managed docker registries.
  • Add possibility to turn-off rabbitmq completely and use Redis as broker instead.
  • Make .polyaxonignore honour .gitignore syntax.
  • Rename declarations section to params.
  • Add typing to Polyaxonfiles and schemas.
  • Extract polyaxon dockerizer logic to reuse in other environments.
  • Add possibility to run polyaxonfiles and update params without changing the yaml files: polyaxon run -f polyaxonfile.yaml -P param1=value1 -P param2=value2.
  • Add possibility to run polyaxonfiles locally.
    • Allow to run experiments (non-distributed) locally either by invoking a python code with a conda/pip env or by generating a dockerfile, with out-of-the box tracking if POLYAXON_NO_OP is not set to true.
  • Deprecate docker registry authentication using user/password in favor of credential stores and credential helpers.
  • Add documentation for using ECR as Docker registry.
  • Mark kaniko stable: possibility to pull / push from any docker registry.
  • Mark jupyter lab stable.
  • Add sorting for outputs.
  • Update internal architecture to allow development and extending the platform with internal plugins.
  • Expose one single port for accessing all internal Polyaxon’s APIs.
  • Update e2e tests with different deployments configurations.
  • Extend input schema to allow users to easily convert boolean params to flags in polyaxonfiles.
  • Add caching layer when checking statuses.
  • Add encryptor for managing sensitive data in-cluster with future possibility to use vault as a backend, currently the backend uses fernet encryption.
  • Add possibility to search entities by name or description using regex.
  • Improve deployment using CLI, add new command: polyaxon admin.
  • Extend build kind with lang_env to easily export language environment.
  • Add initial work on early stopping policies for hyperparameter tuning.
  • Add possibility to install auto-completion for polyaxon CLI.
  • Add possibility to invalidate individual builds and builds under a project.
  • Update API logic to save raw content to post-resolve contexts.
  • Add admin deploy/upgrade —dry-run commands for debugging purposes.
  • Add alpha version of Polyaxon on docker-compose.
  • Allow verify_ssl to be turned on via the CLI.
  • Improve init container stability by introducing retry on ConnectionError.
  • Improve docs for ingress’s annotations.
  • Fix issues in log_artifact and log_artifacts.
  • Fix issue with default ssl path: SSL with node port service should not mount certs in /etc/ssl by default.
  • Fix issue when updating the internal API port.
  • Fix issue when using S3 as outputs storage.
  • Fix issue with the Helm chart not working correctly when RBAC is disabled (This is issue was noticed in Minikube with RBAC disabled).
  • Fix issue preventing to install Polyaxon with the latest helm version.
  • Fix issue with default build backend being ignored in some instances.
  • Fix issue breaking Jupyter lab theme in some recent versions.
  • Fix issue with scheduling S3 credentials for tensorboards.
  • Fix Experiment restart —copy creating additional folder.
  • Fix restart and resume experiments in groups not adding the new experiment to the same group.
  • Add resume button to UI.
  • Add tensorflow tracking to contrib.
  • Improve keras tracking.
  • Add fastai tracking to contrib.
  • Remove several dependencies and replace with simple utils functions.
  • Add initial data/artifacts/logs stores and catalogs.
  • Add initial documentation for Polyflow: an actions/events pipeline engine for creating and automating machine learning workflows.
    • Allow to run ops in parallel while respecting concurrency in pipelines.
    • Allow to run pipelines following schedules: 2 interfaces intervals and crons.
    • Allow to run pipelines with complex ops dependencies as DAGs.
    • All to resolve ops’ definitions, from inline templates, local templates, registries of actions/events.
    • Add initial work on supporting airflow out-of-the-box or possibility to easily port ops to Polyaxon to run container-native operations.
  • Add templating for operationalizing and reusing Polyaxon’s experiments and jobs.