xpk 0.17.1__tar.gz → 0.17.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {xpk-0.17.1/src/xpk.egg-info → xpk-0.17.2}/PKG-INFO +1 -1
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/storage.py +0 -25
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/cluster.py +1 -3
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/common.py +0 -151
- {xpk-0.17.1 → xpk-0.17.2/src/xpk.egg-info}/PKG-INFO +1 -1
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk.egg-info/SOURCES.txt +0 -2
- xpk-0.17.1/src/xpk/core/kjob.py +0 -473
- xpk-0.17.1/src/xpk/templates/volume_bundle.yaml +0 -7
- {xpk-0.17.1 → xpk-0.17.2}/.dockerignore +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/CODEOWNERS +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/PULL_REQUEST_TEMPLATE.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/actions/install-kjob/action.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/actions/install-kueue/action.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/actions/setup-test-env/action.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/release.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/README.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/build_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/build_wheels.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/cleanup.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/gemini-dispatch.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/gemini-invoke.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/gemini-review.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/gemini-scheduled-triage.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/gemini-triage.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/integration_basic_cluster_create.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/integration_legacy_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/integration_pathways_cluster_create.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/integration_ray_cluster_create.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/integration_storage_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/label-validation.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/nightly_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/periodic_release.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/release_branch_versioning.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_build_kjob.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_build_scripts.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_build_wheel.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_goldens.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_integration_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_lint_and_format.yml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_storage_create.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_storage_delete.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/reusable_unit_tests.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.github/workflows/stale.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.gitignore +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/.pre-commit-config.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/LICENSE +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/Makefile +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/README.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/backoff_retry.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/data/Dockerfile +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/code-of-conduct.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/contributing.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/installation.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/local_testing.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/permissions.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/testing.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/troubleshooting.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/advanced.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/autoprovisioning.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/clusters.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/cpu.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/docker.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/gpu.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/inspector.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/job.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/run.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/storage.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/tpu7x/clusters.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/tpu7x/recipes/flex_filestore_recipe.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/tpu7x/recipes/flex_lustre_recipe.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/tpu7x/recipes/reservation_gcs_bucket_recipe.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/tpu7x/workloads.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/docs/usage/workloads.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/batch.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/fake_training.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/job.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/llama-3.1-finetuning/check_cuda.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/llama-3.1-finetuning/requirements.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/llama-3.1-finetuning/train.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/llama-3.1-finetuning/train.slurm +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/llama-3.1-finetuning/training_data.jsonl +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/nccl/nccl-a3mega.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/nccl/nccl-a3ultra.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/nccl/nccl.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/storage/filestore-manifest-attach.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/storage/gcsfuse-manifest.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/storage/lustre-manifest-attach.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/storage/parallelstore-manifest-attach.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/examples/storage/pd-manifest-attach.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/golden_buddy.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Basic_cluster_create.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Batch.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_for_multi-host_nodepool.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_for_single-host_single-slice_TPU.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_private.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_sub-slicing.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_super-slicing.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_CPU_and_memory_limits_above_capacity.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_CPU_and_memory_limits_below_capacity.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_Managed_Lustre_driver.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_Managed_Lustre_driver_and_legacy_port.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_gb200-4.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_create_with_shared_reservation.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_delete.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Cluster_delete_force.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Job_cancel.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Job_info.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Job_list.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/NAP_cluster-create.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/NAP_cluster-create_with_pathways.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Storage_list.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_create.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_create_pathways.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_create_sub-slicing.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_create_super-slicing.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_create_with_output-manifest-file.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_delete.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens/Workload_list.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/goldens.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/pylintrc +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/pyproject.toml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/setup.cfg +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/README.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/docker_manager_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/gcluster_a3mega_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/gcluster_a3ultra_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/gcluster_a4_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/integration/gcluster_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/api/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/api/storage_crd.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3mega/config-map.yaml.tftpl +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3mega/storage_crd.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3ultra/config-map.yaml.tftpl +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3ultra/mlgru-disable.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3ultra/nccl-installer.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a3ultra/storage_crd.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a4/config-map.yaml.tftpl +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a4/nccl-rdma-installer-a4.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/blueprints/a4/storage_crd.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/batch.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/cluster.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/cluster_gcluster.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/cluster_gcluster_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/cluster_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/common.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/config.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/info.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/inspector.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/job.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/kind.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/kjob_common.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/managed_ml_diagnostics.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/managed_ml_diagnostics_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/run.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/shell.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/version.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/workload.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/commands/workload_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/blueprint_definitions.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/blueprint_generator.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/blueprint_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/testing/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/testing/data/a3_mega.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/testing/data/a3_mega_spot.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/testing/data/a3_ultra.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/blueprint/testing/data/a4.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/capacity.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/capacity_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/cluster_private.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/cluster_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/commands.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/config.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/config_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/docker_container.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/docker_image.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/docker_manager.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/docker_resources.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/filestore.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/gcloud_context.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/gcloud_context_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/gcluster_manager.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/gcsfuse.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/jobset.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/kueue_manager.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/kueue_manager_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/monitoring.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/mtc.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/nap.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/network.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/nodepool.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/nodepool_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/pathways.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/pathways_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/ray.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/remote_state/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/remote_state/fuse_remote_state.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/remote_state/remote_state_client.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/resources.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/scheduling.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/scheduling_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/storage.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/system_characteristics.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/system_characteristics_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/telemetry.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/telemetry_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/testing/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/testing/commands_tester.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/testing/commands_tester_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/updates.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/updates_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/vertex.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/rdma_decorator.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/storage_decorator.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/tcpx_decorator.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/tcpx_decorator_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_decorators/tcpxo_decorator.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/core/workload_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/main.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/batch.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/cluster.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/cluster_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/common_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/config.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/core.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/info.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/inspector.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/job.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/kind.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/run.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/shell.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/storage.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/storage_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/validators.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/version.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/workload.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/parser/workload_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/telemetry_uploader.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/arm_gpu_workload_crate.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/cluster_preheat.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/filestore-pv.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/filestore-pvc.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/filestore-sc.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/fuse-pv.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/fuse-pvc.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/kueue_config.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/kueue_gke_default_topology.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/kueue_sub_slicing_topology.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/kueue_super_slicing_topology.yaml.j2 +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/mtc-cpc.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/templates/storage.yaml +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/__init__.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/console.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/console_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/execution_context.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/feature_flags.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/file.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/gcs_utils.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/kubectl.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/kueue.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/network.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/objects.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/templates.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/topology.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/topology_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/user_agent.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/user_agent_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/user_input.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/user_input_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/validation.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/validation_test.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/versions.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk/utils/yaml.py +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk.egg-info/dependency_links.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk.egg-info/entry_points.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk.egg-info/requires.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/src/xpk.egg-info/top_level.txt +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/tools/Dockerfile-kjob +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/tools/build-kjob.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/tools/install-gke-auth-plugin.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/tools/install-xpk.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/xpk-large-scale-guide.sh +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/xpk-notebooks.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/xpk-slurm-commands.md +0 -0
- {xpk-0.17.1 → xpk-0.17.2}/xpk.py +0 -0
|
@@ -23,7 +23,6 @@ from kubernetes.client.rest import ApiException
|
|
|
23
23
|
|
|
24
24
|
from ..core import gcsfuse
|
|
25
25
|
from ..core.cluster import (
|
|
26
|
-
DEFAULT_NAMESPACE,
|
|
27
26
|
add_zone_and_project,
|
|
28
27
|
get_cluster_network,
|
|
29
28
|
setup_k8s_env,
|
|
@@ -35,12 +34,6 @@ from ..core.cluster import (
|
|
|
35
34
|
update_cluster_with_workload_identity_if_necessary,
|
|
36
35
|
)
|
|
37
36
|
from ..core.filestore import FilestoreClient, get_storage_class_name
|
|
38
|
-
from ..core.kjob import (
|
|
39
|
-
KJOB_API_GROUP_NAME,
|
|
40
|
-
KJOB_API_GROUP_VERSION,
|
|
41
|
-
KJOB_API_VOLUME_BUNDLE_PLURAL,
|
|
42
|
-
create_volume_bundle_instance,
|
|
43
|
-
)
|
|
44
37
|
from ..core.storage import (
|
|
45
38
|
GCP_FILESTORE_TYPE,
|
|
46
39
|
GCS_FUSE_TYPE,
|
|
@@ -98,9 +91,6 @@ def storage_create(args: Namespace) -> None:
|
|
|
98
91
|
|
|
99
92
|
k8s_api_client = setup_k8s_env(args)
|
|
100
93
|
create_storage_crds(k8s_api_client, args, manifest)
|
|
101
|
-
create_volume_bundle_instance(
|
|
102
|
-
k8s_api_client, args.name, manifest, args.readonly, args.mount_point
|
|
103
|
-
)
|
|
104
94
|
# Not required for Filestore. Will be uncommented when adding GCSFuse create
|
|
105
95
|
# return_code = update_cluster_with_workload_identity_if_necessary(args)
|
|
106
96
|
# if return_code > 0:
|
|
@@ -214,9 +204,6 @@ def storage_attach(args: Namespace) -> None:
|
|
|
214
204
|
|
|
215
205
|
k8s_api_client = setup_k8s_env(args)
|
|
216
206
|
create_storage_crds(k8s_api_client, args, manifest)
|
|
217
|
-
create_volume_bundle_instance(
|
|
218
|
-
k8s_api_client, args.name, manifest, args.readonly, args.mount_point
|
|
219
|
-
)
|
|
220
207
|
|
|
221
208
|
enable_csi_drivers_if_necessary(args)
|
|
222
209
|
|
|
@@ -332,18 +319,6 @@ def delete_storage_resources(k8s_api_client: ApiClient, storage: Storage):
|
|
|
332
319
|
"Storage Class",
|
|
333
320
|
)
|
|
334
321
|
|
|
335
|
-
delete_resource(
|
|
336
|
-
lambda name: api_instance.delete_namespaced_custom_object(
|
|
337
|
-
namespace=DEFAULT_NAMESPACE,
|
|
338
|
-
name=name,
|
|
339
|
-
group=KJOB_API_GROUP_NAME,
|
|
340
|
-
version=KJOB_API_GROUP_VERSION,
|
|
341
|
-
plural=KJOB_API_VOLUME_BUNDLE_PLURAL,
|
|
342
|
-
),
|
|
343
|
-
storage.name,
|
|
344
|
-
"VolumeBundle",
|
|
345
|
-
)
|
|
346
|
-
|
|
347
322
|
delete_resource(
|
|
348
323
|
lambda name: api_instance.delete_cluster_custom_object(
|
|
349
324
|
name=name,
|
|
@@ -717,10 +717,8 @@ def get_cluster_credentials(args) -> int:
|
|
|
717
717
|
location=location,
|
|
718
718
|
dns_endpoint=True,
|
|
719
719
|
)
|
|
720
|
-
if return_code != 0:
|
|
721
|
-
return return_code
|
|
722
720
|
|
|
723
|
-
if not _are_credentials_valid():
|
|
721
|
+
if return_code != 0 or not _are_credentials_valid():
|
|
724
722
|
xpk_print('Detected error. Retrying without --dns-endpoint flag...')
|
|
725
723
|
return_code = _get_credentials(
|
|
726
724
|
project=args.project,
|
|
@@ -180,157 +180,6 @@ def add_global_arguments(custom_parser_or_group: ParserOrArgumentGroup):
|
|
|
180
180
|
)
|
|
181
181
|
|
|
182
182
|
|
|
183
|
-
def add_slurm_arguments(custom_parser_or_group: ParserOrArgumentGroup):
|
|
184
|
-
"""Add Slurm job arguments to the parser.
|
|
185
|
-
|
|
186
|
-
Args:
|
|
187
|
-
custom_parser_or_group: parser or argument group to add global arguments to.
|
|
188
|
-
"""
|
|
189
|
-
custom_parser_or_group.add_argument(
|
|
190
|
-
'--ignore-unknown-flags',
|
|
191
|
-
type=bool,
|
|
192
|
-
action=argparse.BooleanOptionalAction,
|
|
193
|
-
default=False,
|
|
194
|
-
help='Ignore all the unsupported flags in the bash script.',
|
|
195
|
-
)
|
|
196
|
-
custom_parser_or_group.add_argument(
|
|
197
|
-
'-a',
|
|
198
|
-
'--array',
|
|
199
|
-
type=str,
|
|
200
|
-
default=None,
|
|
201
|
-
help=(
|
|
202
|
-
'Submit a job array, multiple jobs to be executed with identical'
|
|
203
|
-
' parameters. The indexes specification identifies what array index'
|
|
204
|
-
' values should be used. For example, "--array=0-15" or'
|
|
205
|
-
' "--array=0,6,16-32". Multiple values may be specified using a comma'
|
|
206
|
-
' separated list and/or a range of values with a "-" separator. For'
|
|
207
|
-
' example "--array=0-15%%4" will limit the number of simultaneously'
|
|
208
|
-
' running tasks from this job array to 4. The minimum index value is'
|
|
209
|
-
' 0. The maximum index value is 2147483647.'
|
|
210
|
-
),
|
|
211
|
-
)
|
|
212
|
-
custom_parser_or_group.add_argument(
|
|
213
|
-
'-c',
|
|
214
|
-
'--cpus-per-task',
|
|
215
|
-
type=str,
|
|
216
|
-
default=None,
|
|
217
|
-
help='How much cpus a container inside a pod requires.',
|
|
218
|
-
)
|
|
219
|
-
custom_parser_or_group.add_argument(
|
|
220
|
-
'--gpus-per-task',
|
|
221
|
-
type=str,
|
|
222
|
-
default=None,
|
|
223
|
-
help='How much gpus a container inside a pod requires.',
|
|
224
|
-
)
|
|
225
|
-
custom_parser_or_group.add_argument(
|
|
226
|
-
'--mem',
|
|
227
|
-
type=str,
|
|
228
|
-
default=None,
|
|
229
|
-
help='How much memory a pod requires.',
|
|
230
|
-
)
|
|
231
|
-
custom_parser_or_group.add_argument(
|
|
232
|
-
'--mem-per-task',
|
|
233
|
-
type=str,
|
|
234
|
-
default=None,
|
|
235
|
-
help='How much memory a container requires.',
|
|
236
|
-
)
|
|
237
|
-
custom_parser_or_group.add_argument(
|
|
238
|
-
'--mem-per-cpu',
|
|
239
|
-
type=str,
|
|
240
|
-
default=None,
|
|
241
|
-
help=(
|
|
242
|
-
'How much memory a container requires, it multiplies the number '
|
|
243
|
-
'of requested cpus per task by mem-per-cpu.'
|
|
244
|
-
),
|
|
245
|
-
)
|
|
246
|
-
custom_parser_or_group.add_argument(
|
|
247
|
-
'--mem-per-gpu',
|
|
248
|
-
type=str,
|
|
249
|
-
default=None,
|
|
250
|
-
help=(
|
|
251
|
-
'How much memory a container requires, it multiplies the number '
|
|
252
|
-
'of requested gpus per task by mem-per-gpu.'
|
|
253
|
-
),
|
|
254
|
-
)
|
|
255
|
-
custom_parser_or_group.add_argument(
|
|
256
|
-
'-N',
|
|
257
|
-
'--nodes',
|
|
258
|
-
type=int,
|
|
259
|
-
default=None,
|
|
260
|
-
help='Number of pods to be used at a time.',
|
|
261
|
-
)
|
|
262
|
-
custom_parser_or_group.add_argument(
|
|
263
|
-
'-n',
|
|
264
|
-
'--ntasks',
|
|
265
|
-
type=int,
|
|
266
|
-
default=None,
|
|
267
|
-
help='Number of identical containers inside of a pod, usually 1.',
|
|
268
|
-
)
|
|
269
|
-
custom_parser_or_group.add_argument(
|
|
270
|
-
'-o',
|
|
271
|
-
'--output',
|
|
272
|
-
type=str,
|
|
273
|
-
default=None,
|
|
274
|
-
help=(
|
|
275
|
-
'Where to redirect the standard output stream of a task. If not'
|
|
276
|
-
' passed it proceeds to stdout, and is available via kubectl logs.'
|
|
277
|
-
),
|
|
278
|
-
)
|
|
279
|
-
custom_parser_or_group.add_argument(
|
|
280
|
-
'-e',
|
|
281
|
-
'--error',
|
|
282
|
-
type=str,
|
|
283
|
-
default=None,
|
|
284
|
-
help=(
|
|
285
|
-
'Where to redirect std error stream of a task. If not passed it'
|
|
286
|
-
' proceeds to stdout, and is available via kubectl logs.'
|
|
287
|
-
),
|
|
288
|
-
)
|
|
289
|
-
custom_parser_or_group.add_argument(
|
|
290
|
-
'--input',
|
|
291
|
-
type=str,
|
|
292
|
-
default=None,
|
|
293
|
-
help='What to pipe into the script.',
|
|
294
|
-
)
|
|
295
|
-
custom_parser_or_group.add_argument(
|
|
296
|
-
'-J',
|
|
297
|
-
'--job-name',
|
|
298
|
-
type=str,
|
|
299
|
-
default=None,
|
|
300
|
-
help='What is the job name.',
|
|
301
|
-
)
|
|
302
|
-
custom_parser_or_group.add_argument(
|
|
303
|
-
'-D',
|
|
304
|
-
'--chdir',
|
|
305
|
-
type=str,
|
|
306
|
-
default=None,
|
|
307
|
-
help='Change directory before executing the script.',
|
|
308
|
-
)
|
|
309
|
-
custom_parser_or_group.add_argument(
|
|
310
|
-
'-t',
|
|
311
|
-
'--time',
|
|
312
|
-
type=str,
|
|
313
|
-
default=None,
|
|
314
|
-
help=(
|
|
315
|
-
'Set a limit on the total run time of the job. '
|
|
316
|
-
'A time limit of zero requests that no time limit be imposed. '
|
|
317
|
-
'Acceptable time formats include "minutes", "minutes:seconds", '
|
|
318
|
-
'"hours:minutes:seconds", "days-hours", "days-hours:minutes" '
|
|
319
|
-
'and "days-hours:minutes:seconds".'
|
|
320
|
-
),
|
|
321
|
-
)
|
|
322
|
-
custom_parser_or_group.add_argument(
|
|
323
|
-
'--priority',
|
|
324
|
-
type=str,
|
|
325
|
-
default='medium',
|
|
326
|
-
choices=['very-low', 'low', 'medium', 'high', 'very-high'],
|
|
327
|
-
help=(
|
|
328
|
-
'A priority, one of `very-low`, `low`, `medium`, `high` or'
|
|
329
|
-
' `very-high`. Defaults to `medium`.'
|
|
330
|
-
),
|
|
331
|
-
)
|
|
332
|
-
|
|
333
|
-
|
|
334
183
|
def add_tpu_type_argument(
|
|
335
184
|
custom_parser_or_group: ParserOrArgumentGroup,
|
|
336
185
|
required: bool = False,
|
|
@@ -182,7 +182,6 @@ src/xpk/core/gcloud_context_test.py
|
|
|
182
182
|
src/xpk/core/gcluster_manager.py
|
|
183
183
|
src/xpk/core/gcsfuse.py
|
|
184
184
|
src/xpk/core/jobset.py
|
|
185
|
-
src/xpk/core/kjob.py
|
|
186
185
|
src/xpk/core/kueue_manager.py
|
|
187
186
|
src/xpk/core/kueue_manager_test.py
|
|
188
187
|
src/xpk/core/monitoring.py
|
|
@@ -262,7 +261,6 @@ src/xpk/templates/kueue_sub_slicing_topology.yaml.j2
|
|
|
262
261
|
src/xpk/templates/kueue_super_slicing_topology.yaml.j2
|
|
263
262
|
src/xpk/templates/mtc-cpc.yaml
|
|
264
263
|
src/xpk/templates/storage.yaml
|
|
265
|
-
src/xpk/templates/volume_bundle.yaml
|
|
266
264
|
src/xpk/utils/__init__.py
|
|
267
265
|
src/xpk/utils/console.py
|
|
268
266
|
src/xpk/utils/console_test.py
|