Repository navigation

Issues

New issue

Kubeflow SDK v0.5 Release
#648 · andreyvelich opened on Jul 24, 2026
KEP-107: Support Spark with Kubeflow SDK
#597 · tariq-hasan opened on Jul 16, 2026
2
Kubeflow SDK v0.6 Release
#837 · tariq-hasan opened on Oct 8, 2026
2

Docs: quickstart and runtimes Trainer snippets fail when copied
kubeflow/sdk#939 · opened on Nov 9, 2026
ContainerBackend.get_job_logs() yields raw Docker/Podman chunks instead of log lines
kubeflow/sdk#938 · opened on Nov 9, 2026
bug(trainer): LocalProcessBackend.train() fails when runtime is omitted, contradicting docs and other backends
kubeflow/sdk#936 · opened on Nov 9, 2026
1
Spark Examples E2E uses current Spark Operator controller with older Helm chart RBAC
kubeflow/sdk#933 · opened on Nov 6, 2026
8
bug(spark): connect(base_url=..., token=...) never sends the token, so token-protected servers reject the session
kubeflow/sdk#927 · opened on Nov 4, 2026
2
bug(optimizer): GridSearch experiments fail in Katib with the default num_trials or a continuous parameter
kubeflow/sdk#924 · opened on Nov 3, 2026
2
bug(trainer): ContainerBackend.train() with an existing TrainJob's name deletes that job's workdir and network
kubeflow/sdk#922 · opened on Nov 3, 2026
1
bug(trainer): MPI TrainJobs report Created while running and Running before the launcher starts
kubeflow/sdk#920 · opened on Nov 3, 2026
test(trainer): add unit tests for PodmanClientAdapter
kubeflow/sdk#918 · opened on Nov 2, 2026
2
bug(trainer): LocalProcessBackend breaks packages_to_install entries that contain quotes, e.g. environment markers
kubeflow/sdk#916 · opened on Nov 2, 2026
2
bug(trainer): multiple RuntimePatch options produce duplicate spec.runtimePatches managers
kubeflow/sdk#911 · opened on Oct 29, 2026
chore(spark): align SDK Spark defaults with Spark Operator 4.2.0
kubeflow/sdk#909 · opened on Oct 28, 2026
2
[Bug] LocalProcessBackend passes func_args as a positional dictionary instead of keyword arguments
kubeflow/sdk#906 · opened on Oct 26, 2026
1
bug(trainer, optimizer): default namespace is resolved from the default kubeconfig instead of KubernetesBackendConfig.config_file
kubeflow/sdk#899 · opened on Oct 24, 2026
bug(spark): unsuffixed and exponent-form Kubernetes memory quantities are passed to Spark unchanged (bytes read as MiB, 1e9 rejected), and a non-string value raises TypeError
kubeflow/sdk#898 · opened on Oct 24, 2026
2
OptimizationJob status shows Created while the Katib Experiment is running
kubeflow/sdk#892 · opened on Oct 21, 2026
Migrate deprecated Pydantic class-based Config to ConfigDict
kubeflow/sdk#890 · opened on Oct 21, 2026
2
bug(trainer): LocalProcessBackend leaks the per-job virtualenv when the TrainJob fails
kubeflow/sdk#888 · opened on Oct 20, 2026
bug(trainer): ContainerPatch accepts env entries with both value and valueFrom set
kubeflow/sdk#883 · opened on Oct 20, 2026
2
fix(trainer): DataCacheInitializer accepts empty schema or table name in storage_uri
kubeflow/sdk#881 · opened on Oct 19, 2026
bug(trainer): CustomTrainer func_args with non literal values fail inside the TrainJob
kubeflow/sdk#879 · opened on Oct 19, 2026
1
TrainerClient.wait_for_job_status polls until timeout for an invalid or empty status on local backends
kubeflow/sdk#877 · opened on Oct 19, 2026
Docs: Outdated .types imports cause ImportError across Trainer and Optimizer doc guides
kubeflow/sdk#875 · opened on Oct 18, 2026
2
bug(spark): wait_for_job_status accepts invalid statuses and waits until timeout
kubeflow/sdk#873 · opened on Oct 18, 2026
2
feat(trainer): make list_jobs() send a constant number of API requests instead of ~3 per TrainJob
kubeflow/sdk#870 · opened on Oct 18, 2026
1