v0.7.0 release

Features

• support cleanPodPolicy is Running, same as tf operator (#288, @jiaqianjing)
• Migrate pytorch-operator to go modules (#272, @Jeffwan)
• feat(init_container): Add init container image CLI argument (#265, @gaocegege)
• SDK supports getting PyTorchJob training process or logs (#252, @jinchihe)
• Add watch function for PyTorchJob python Client API (#246, @jinchihe)
• Add more APIs for SDK (#240, @jinchihe)
• feat(deletePodsAndServices):only delete master service (#233, @leileiwan)
• Generate Kubeflow PyTorchJob SDK (#227, @jinchihe)
• feat: Replace common with kubeflow/common (#225, @gaocegege)
• feat: Use golanglint (#226, @gaocegege)
• Removing v1beta2 support (#222, @johnugeorge)
• feat: Support running although it is uesless (#194, @gaocegege)
• use init container for worker pod to wait master pod ready (#187, @zlcnju)
• fix: Fix the comments (#193, @gaocegege)
• Minor fix to add CoreV1 to scheme (#184, @johnugeorge)
• update release script; fix post submit (#189, @johnugeorge)

Bug fixed

• Fix Unit Tests (#293, @andreyvelich)
• Fix minor OpenShift issues - resource requests, Dockerfile (#276, @vpavlin)
• Fix the link to run_e2e_workflow.py script (#266, @terrytangyuan)
• fix: Add resource limits for init container (#253, @gaocegege)
• fix the reconcile flow (#242, @ChanYiLin)
• fix(*) rm work service in controller_test.go (#235, @leileiwan)
• fix(job_test) test case should not include worker service (#231, @leileiwan)

Chores

• Change mnist example to use FashionMNIST (#327, @Jeffwan)
• pytorch-operator: Consolidate manifests (#323, @yanniszark)
• Temporarily disable mnist test case (#326, @Jeffwan)
• PyTorch Operator: Move manifests development upstream (#320, @yanniszark)
• Migrate to new test-infra (#316, @PatrickXYS)
• update pytorch-operator deployment manifests file (#295, @myonlyzzy)
• Add @andreyvelich to approvers (#309, @andreyvelich)
• Reuse Common Scripts for Creating / Deleting EKS clusters (#308, @PatrickXYS)
• Add Jeffwan@ to OWNERS (#306, @Jeffwan)
• Move PyTorch Operator e2e tests to AWS Prow (#305, @Jeffwan)
• Update openapi-gen to not rely on vendor (#274, @Jeffwan)
• Update README.md (#290, @pingsutw)
• Update CRD link (#289, @pingsutw)
• Adds notes and example annotation for pytorch job (#285, @shawnzhu)
• chore: Update OWNERS (#286, @gaocegege)
• fix Dockerfile-mpi download miniconda.sh (#277, @jiaqianjing)
• Update swagger-codegen-cli URL (#280, @jinchihe)
• pin kubenertes client version to work around a bug (#262, @jinchihe)
• Added The Pytorch GPU Docker under the appropriate folder (#255, @MATRIX4284)
• Copy third party vendor source code to Docker image (#251, @johnugeorge)
• Add third party license info (#250, @johnugeorge)
• ConvertPyTorchJobToUnstructured uses function ToUnstructured to convert PyTorchJob to Unstructured instead of json (#241, @leileiwan)
• replace gopkg.in/yaml.v2 with github.com/kubernetes-sigs/yaml repo (#238, @xrmzju)
• Update tf operator branch dep (#223, @johnugeorge)
• Avoiding unnecessary status update (#220, @johnugeorge)
• Removing unnecessary rbac permissions (#221, @johnugeorge)
• add mnist example dockerfile for ppc64le (#218, @zheddie)
• Fix nslookup cannot work well in initContainerTemplate (#216, @hougangliu)
• Minor change in log (#213, @johnugeorge)
• Delete v1beta2 code (#212, @johnugeorge)
• Add qps and burst options (#210, @ohmystack)
• Set pytorchjob defaults in test utils (#208, @ohmystack)
• Update codegen and verify in CI (#207, @ohmystack)
• Update manifest to v0.6.0 (#200, @hougangliu)
• Common label changes with K8s upgrade to 1.12.3 (#204, @johnugeorge)
• Use multi-build to build pytorch-operator image (#198, @hmtai)
• add total suffix in counter metrics (#201, @yeya24)
• add kubeconfig flag (#192, @yeya24)
• Remove unnecessary services for worker (#191, @hougangliu)

v1.0.0-rc.0

pytorch-operator pre graduate

v0.6.0 release

Merged pull requests:

• set annotation automatically when EnableGangScheduling is set to true #181 (zlcnju)
• fix wrong api version when delete pytorchjob #179 (wackxu)
• Moving crd to manifests #178 (johnugeorge)
• Adds developer guide and sample CRD for v1 #177 (krishnadurai)
• Update image base to UBI8 GA #176 (johnugeorge)
• PyTorch Operator Prometheus Metrics #175 (krishnadurai)
• Set start timestamp #170 (johnugeorge)
• Skip condition update when succeeded #173 (johnugeorge)
• Sync PodGroup fix #172 (johnugeorge)
• Check pending status for pastBackoffLimitOnFailure #171 (johnugeorge)
• Making ResyncPeriod configurable #169 (johnugeorge)
• add uuid to id for leader election #168 (fisherxu)
• Polish documentation for PyTorch V1 #167 (richardsliu)
• Remove v1beta1 code #166 (johnugeorge)
• Adding tests for operator v1 api #165 (johnugeorge)
• Adding examples for v1 api #164 (johnugeorge)
• Implementation of Pytorch operator v1 API #162 (johnugeorge)
• Revise API version to v1beta2 #159 (krishnadurai)
• set CompletionTime first when pytorchjob exceeds limit #158 (wackxu)
• Prune owners file #157 (johnugeorge)
• Set tf operator version to v0.5.0 #156 (johnugeorge)
• Minor format changes #155 (johnugeorge)
• Adding cleaner base image for operator builds #153 (johnugeorge)

v0.5.1 release

Sync PodGroup fix (#172)

v0.5.0 release

Closed issues:

• Ensuring CRD requires cluster-level authority #144
• Label naming style inconsistent #140
• Pytorch operator v1beta2 API #134
• Support gang-scheduling by kube-batch #129
• Pytorch workers keep crashing if master is not up yet. #125
• Support cross compile for image build. #42
• Deprecate v1alpha2 API #135
• Distribution across multi-gpu nodes #128
• Upgrade examples to Pytorch 1.0 #123
• Double gradient reduction in examples? #122

Merged pull requests:

• Implement ActiveDeadlineSeconds and BackoffLimit #151 (johnugeorge)
• Use podGroup instead of PDB in v1beta2 #150 (johnugeorge)
• Use kube-batch as scheduler by default when gang-scheduling is enabled #149 (johnugeorge)
• Remove usage of crd client #148 (johnugeorge)
• Update tests to have single operator deployment for v1beta1 and v1beta2 API #147 (johnugeorge)
• Renaming labels to consistent format #146 (johnugeorge)
• Workers are created only when the master is in running phase #145 (johnugeorge)
• Adding tests for v1beta2 #143 (johnugeorge)
• Change cluster version to 1.11 #142 (andreyvelich)
• Update OWNERS #141 (andreyvelich)
• Adding status subresource #139 (johnugeorge)
• Adding v1beta2 API implementation #138 (johnugeorge)
• Upgrading k8s to 1.11 #137 (johnugeorge)
• Removing v1alpha2 API #136 (johnugeorge)
• Adding detailed events/messages to PyTorch Jobs #133 (johnugeorge)
• Skip status reinit when job is succeeded #132 (johnugeorge)
• Travis build fix #131 (johnugeorge)
• Rework example and e2e test script #126 (TimZaman)
• Change Distributed Data Parallel example #124 (andreyvelich)

v0.5.0-rc.1 release

Adding v1beta2 API implementation (#138)

* Adding v1beta2 API implementation

* Build v1beta2

v0.4.0 release

Closed issues:

• Delete v1alpha1 API and controller from the repository #105
• Create v1beta1 Pytorch operator docker image #104
• Implement v1beta1 controller #96
• MPI backend mnist gpu example error: "No space left on device" #91
• pytorch-operator should ensure that CRD exists #87
• Refactor E2E tests #86
• [discussion] Refactor pytorch operator APIs #84

Merged pull requests:

• Updated gcloud build related code. #121 (ltomes)
• Adding Operator deployment manifests #119 (johnugeorge)
• Adding distributed mnist example with summaries #118 (johnugeorge)
• Add master role label for PyTorchJob #116 (johnugeorge)
• Minor fixes #115 (johnugeorge)
• GetCondition func fix #114 (johnugeorge)
• Update k8s cluster version to 1.10 #113 (johnugeorge)
• Pytorch Katib example #112 (johnugeorge)
• Updating pkg version to 1.10.1 #111 (johnugeorge)
• Delete v1alpha1 API and Controller #110 (johnugeorge)
• Delete v1alpha1 tests #109 (johnugeorge)
• gopkg: Use version instead of branch #107 (gaocegege)
• Keeping default project as kubeflow-ci #106 (johnugeorge)
• Adding v1beta1 binary to the docker image #103 (johnugeorge)
• Fix registry override for image builds #102 (johnugeorge)
• Adding e2e tests for v1beta1 #101 (johnugeorge)
• Release image to kubeflow images public #100 (johnugeorge)
• Ensure that PyTorch CRD exists #99 (johnugeorge)
• Adding examples for v1beta1 API #98 (johnugeorge)
• Adding controller for v1beta1 api #97 (johnugeorge)
• Updating common JobController vendor #95 (johnugeorge)
• Validation test for V1beta1 apis #94 (johnugeorge)
• Pytorch operator v1beta1 APIs #93 (johnugeorge)
• Fix MPI backend mnist gpu example error: "No space left on device" #92 (jwwandy)
• vendor: Update to 1.10 #90 (gaocegege)
• Add richardsliu to OWNERS #85 (richardsliu)

v0.4.0-rc.1 release

Updating pkg version to 1.10.1 (#111)

* Updating k8s to 1.10.1

* Update to 1.10.1

v0.3.0 Release

v0.3.0 release of PyTorch operator