Datasets:

hjerpe

/

github-kubeflow-issues

like 0

https://api.github.com/repos/kubeflow/pipelines/issues/7455

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7455/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7455/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7455/events

https://github.com/kubeflow/pipelines/issues/7455

I_kwDOB-71UM5GPrN8

[feature] Support IR YAML format in API

closed

COLLABORATOR

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> /area backend <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? <!-- Provide a description of this feature and the user experience. --> Use YAML instead of JSON for intermediate representation (IR). ### What is the use case or pain point? <!-- It helps us understand the benefit of this feature for your use case. --> YAML is a superset of JSON and is easier to understand for a reader. It also paves the way to enhancing readability in the future. ### Is there a workaround currently? <!-- Without this feature, how do you accomplish your task today? --> NA --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7455/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7454

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7454/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7454/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7454/events

https://github.com/kubeflow/pipelines/issues/7454

I_kwDOB-71UM5GPkOu

[feature] Access to ParallelFor values and set_paralellism per Op

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> /area backend /area sdk <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? I stumbled upon a case where two features could be useful when building dynamically a pipeline that is controlled by some outside config. ### What is the use case or pain point? Let's say we have a number of datasets and number of models we want to train. Not all models should be run on all datasets, so we use a config to specify pipeline content. (Examples in the code). Moreover, we would like that each run-per-dataset is parallel to each other and some task may be CPU/RAM heavy so we would like to limit the parallelism per Op-type. This i a case when we want to run a pipeline from github CI/CD to test the newly pushed code on a set of dataset and models to ensure it's validity and performance. ### Is there a workaround currently? Semi-workaround is presented in the example as using config in a function that returns pipeline-function and iterating over the config without using `ParallelFor`, but this seems problematic in limiting parallelism per Op. For the problem of paralellism I saw a related issue https://github.com/kubeflow/pipelines/issues/4089, but also I know that Argo allows setting parallelism per Task/Step, so having that in Kubeflow would be nice. Also using separate pipelines per dateset is somewhat a working option. ```import kfp from kfp.components import func_to_container_op import kfp.dsl as dsl import json ### Functions def print_fun(calculation: str) -> str: print("Calculation Type: ",calculation) return calculation def add(a: float, b: float) -> float: return a + b def multiply(a: float, b: float) -> float: return a * b def divide(a: float, b: float) -> float: return a / b def subtract(a: float, b: float) -> float: return a - b ### Container ops print_op = func_to_container_op(print_fun) add_op = func_to_container_op(add) multiply_op = func_to_container_op(multiply) divide_op = func_to_container_op(divide) subtract_op = func_to_container_op(subtract) ### Dict to easily fetch op according to config name2operator = { "add": add_op, "multiply": multiply_op, "divide": divide_op, "subtract": subtract_op } ### Example config master_config = [ { "name": "name1", "a": 1, "b": 1, "models": ["add", "subtract", "multiply", "divide"] }, { "name": "name2", "a": 0, "b": 1, "models": ["multiply", "divide"] }, { "name": "name3", "a": 100, "b": 2, "models": ["subtract", "multiply", "divide"] }, ] ``` Semi-working solution ``` ### Workaround to use for-loop and create model-task in line with config. ### Using ParallelFor won't allow to get the config as def get_pipeline(config): @dsl.pipeline( name='Parallel pipeline mock test', description='Pipeline with for-loop and config-based operators.' ) def multi_pipeline(): for config_item in config: root_op = print_op(config_item['name']) root_op. for model_name in config_item['models']: new_op = name2operator[model_name](config_item['a'], config_item['b']) new_op.after(root_op) return multi_pipeline client = kfp.Client() client.create_run_from_pipeline_func(get_pipeline(master_config), arguments={}) ``` Example of my thought process in solving the problem – changing `ParallelFor` ``` @dsl.pipeline( name='Parallel pipeline mock test', description='Pipeline with for-loop and config-based operators.' ) def multi_pipeline( config ): root_ops = [] # I only want parallelism limit on the print_op for idx, item in enumerate(dsl.ParallelFor(config, parallelism=2)): root_op = print_op(item['name']) root_ops.append(root_op) # Now it is impossible, as config is PipelineParam and it is not iterable, but would be nice if it is for root_idx, item in enumerate(config): # item['models'] is also not iterable for model_name in item['models']: new_op = name2operator[model_name](item.a, item.b) new_op.after(root_ops[root_idx]) client = kfp.Client() client.create_run_from_pipeline_func(multi_pipeline, arguments={ "config": json.dumps(master_config) }) ``` --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7454/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7450

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7450/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7450/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7450/events

https://github.com/kubeflow/pipelines/issues/7450

I_kwDOB-71UM5GLS3J

[sdk] unarchive_run sets run.storage_state to None instead of STORAGESTATE_ACTIVE

open

NONE

### Environment * KFP version: 1.4 * KFP SDK version: 1.8.11 * All dependencies version: NA ### Steps to reproduce 1. Do a pipeline run: ``` run = client.create_run_from_pipeline_package( pipeline.yaml, run_name='my_run', experiment_name='my_exp', arguments=args ) result = run.wait_for_run_to_complete() ``` 2. Archive the run ``` client.runs.archive_run(result.run.id) ``` 3. Check that the storage state has transition to STORAGESTATE_ARCHIVED ``` run_after_archive = client.runs.get_run(result.run.id) print(run_after_archive.run.storage_state) ``` 4. Now, unarchive the run, and verify the storage state has transitioned to None ``` client.runs.unarchive_run(result.run.id) run_after_unarchive = client.runs.get_run(result.run.id) print(run_after_unarchive.run.storage_state) ``` ### Expected result When a run is unarchived, the storage state should transition back to STORAGESTATE_ACTIVE. This would be consistent with what I get when I archive, and then subsequently unarchive an experiment using similar commands (client.experiment.archive_experiment, and client.experiment.unarchive_experiment). In the experiment unarchive case, the storage state transitions back to STORAGESTATE_ACTIVE. ### Materials and Reference Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7450/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7449

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7449/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7449/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7449/events

https://github.com/kubeflow/pipelines/issues/7449

I_kwDOB-71UM5GLMFT

kubeflow pipeline run in "kubeflow" namespace instead of user namespace

open

NONE

/kind question **Question:** My team has installed kubeflow 1.4 on AKS 1.21 with multi-user mode enabled. When I created a pipeline run in a user namespace, it was run in “kubeflow” namespace, instead of the user namespace. There is no namespace specified in pipeline yaml, and when a pipeline run is created, it automatically added "namespace: kubeflow" in the pod metadata and run it there, like what is shown in below screenshot.  I tried to specify namespace in pipeline yaml file, but it gives me error. Probably because kubeflow does not allow overwrite the namespace.  If I created a notebook, it was inside the user namespace as expected though. Any idea how to fix this?

https://api.github.com/repos/kubeflow/pipelines/issues/7449/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7445

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7445/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7445/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7445/events

https://github.com/kubeflow/pipelines/issues/7445

I_kwDOB-71UM5GC3Cf

[feature] Option to disable downloading of artifacts through UI

closed

NONE

### Feature Area /area frontend /area backend ### What feature would you like to see? - Option to disable the ability to download artifacts from the front-end UI. Some form of this would limit accidentally clicking the download link. - Option to restrict access to the back-end artifact storage. Some form of this would limit "guessing" the download link (unsure if that is even possible). ### What is the use case or pain point? We work with sensitive data. It is very simple to download artifacts produced from the front-end UI and save locally. ### Is there a workaround currently? 1. I have not found any option in https://github.com/kubeflow/pipelines/blob/master/frontend/server/app.ts#L119-L150, but I might have been looking in the wrong places. 2. We are looking into blocking access between the minio server and the UI server through kubernetes.

https://api.github.com/repos/kubeflow/pipelines/issues/7445/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7444

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7444/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7444/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7444/events

https://github.com/kubeflow/pipelines/issues/7444

I_kwDOB-71UM5F_uwo

Convert PipelineSpec format from json to yaml

closed

COLLABORATOR

## Problem There are two types of Pipeline Template definition: ArgoWorkflow (for KFPv1) and PipelineSpec (for KFPv2). Currently we are using `isPipelineSpec()` to determine whether it is v1 or v2. Reference: https://github.com/kubeflow/pipelines/blob/939f81088b39ba703cc821d18e7a99523662bf5f/frontend/src/lib/v2/WorkflowUtils.ts#L48-L67 We are assuming that if the PipelineTemplate is not ArgoWorkflow, then it is PipelineSpec: see `isArgoWorkflowTemplate()`: https://github.com/kubeflow/pipelines/blob/939f81088b39ba703cc821d18e7a99523662bf5f/frontend/src/lib/v2/WorkflowUtils.ts#L22-L31 We are assuming that PipelineSpec is in JSON format. https://github.com/kubeflow/pipelines/blob/939f81088b39ba703cc821d18e7a99523662bf5f/frontend/src/lib/v2/WorkflowUtils.ts#L34-L45 Now the direction is to use YAML format instead of JSON format on PipelineSpec. We need to switch our logic to adopt YAML format and abandon JSON format. See SDK reference: https://github.com/kubeflow/pipelines/pull/7431. ## Possible solution Possible tool: https://github.com/nodeca/js-yaml ## Note We also need to update test data https://github.com/kubeflow/pipelines/tree/master/frontend/mock-backend/data/v2/pipeline

https://api.github.com/repos/kubeflow/pipelines/issues/7444/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7441

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7441/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7441/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7441/events

https://github.com/kubeflow/pipelines/issues/7441

I_kwDOB-71UM5F-aVf

[feature] Clone recurring run capability in sdk

open

CONTRIBUTOR

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> /area sdk <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? There's no way to patch an existing recurring run via the SDK (or GUI). The GUI gets around this with the clone recurring run button, but there's no equivalent in the SDK. A very common scenario, for example, is updating the cron string of an existing recurring run. This is more or less impossible via the SDK. One way to address this would be by adding a `clone_recurring_run` method to the SDK that emulates what the frontend does. <!-- Provide a description of this feature and the user experience. --> ### What is the use case or pain point? Making minor modifications to an existing recurring run via the SDK is not possible. <!-- It helps us understand the benefit of this feature for your use case. --> ### Is there a workaround currently? Only through the GUI via the clone button. <!-- Without this feature, how do you accomplish your task today? --> --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7441/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7437

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7437/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7437/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7437/events

https://github.com/kubeflow/pipelines/issues/7437

I_kwDOB-71UM5F7-W9

Why is certificates.k8s.io/v1 used in Cache Deployer instead of OpenSSL?

open

NONE

Caching is one of the most crucial features of KFP. Each time a pipeline step is the same as an already executed, the results are loaded from the cache server. Caching is accomplished in KFP via two interdependent modules: the cache deployer and the cache server. While trying to set up the modules in an enterprise cluster ([Mercedes-Benz AG](https://github.com/mercedes-benz/DnA)), it was noted that the installation couldn’t be completed. The reason was that the cache deployer is built to generate a Signed Certificate for the cache server by referring to the Kubernetes Certificate-SigningRequest API. ``` yaml ... # create server cert/key CSR and send to k8s API cat <<EOF | kubectl create -f - apiVersion: certificates.k8s.io/v1 kind: CertificateSigningRequest metadata: name: ${csrName} spec: groups: - system:authenticated request: $(cat ${tmpdir}/server.csr | base64 | tr -d '\n') signerName: kubernetes.io/kubelet-serving usages: - digital signature - key encipherment - server auth EOF .. ``` The usage of API server certificates in our enterprise environment is restricted because those allow permission escalation. The security risk is critical, as by using this API, users can order certificates that let them impersonate both Kubernetes control plane and cluster team access. To adjust the cache deployer’s certificate generation process without affecting the actual functionality to avoid loosening the security restrictions, we [used](https://github.com/mercedes-benz/DnA/blob/9c2487e111490285ee57dc241aa27778a1acc774/deployment/dockerfiles/kubeflow/kfp/backend/src/cache/deployer/webhook-create-signed-cert.sh#L98) the widely-known OpenSSL. Is there any specific reason for using the K8s API? If not, would the community be interested in an upstream contribution?

https://api.github.com/repos/kubeflow/pipelines/issues/7437/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7435

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7435/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7435/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7435/events

https://github.com/kubeflow/pipelines/issues/7435

I_kwDOB-71UM5F3n70

[feature] Add a param in create_recurring_run to support override the exists job with the same name in the same experiment

closed

CONTRIBUTOR

### Feature Area /area sdk ### What feature would you like to see? Add a param in `create_recurring_run` func to support override exists recurring job with the same name. ### What is the use case or pain point? In my use case, when I changed the logic of my job, I will remove the jobs with the same name first, and create a new one with the same name again. I do this because I don't want to create duplicated jobs. The code will looks like ```python recurring_run_name = "..." recurring_run_description = "..." jobs = kfp_client.list_recurring_runs( experiment_id=experiment_id, filter=filter_by_job_name, ... ).jobs if jobs != None: # Delete the jobs kfp_client.create_recurring_run( experiment_id=experiment_id, job_name=recurring_run_name, description=recurring_run_description, ... ) ``` So, would it accepted for you guys to add a param such as `replace_exists` to implement these logic. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7435/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7432

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7432/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7432/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7432/events

https://github.com/kubeflow/pipelines/issues/7432

I_kwDOB-71UM5F1yI9

[feature] Support for features in v2

open

CONTRIBUTOR

### Feature Area /area sdk ### What feature would you like to see? Would like to decipher wether KFP v2 will continue to support: - Recursive pipelines (https://www.kubeflow.org/docs/components/pipelines/sdk/dsl-recursion/) - ContainerOp components using python SDK directly (not python functions) - LocalClient for local execution ### What is the use case or pain point? Recursive pipelines is useful for research HPT and similar use cases where it's not clear when will a pipeline need to stop executing. ### Is there a workaround currently? This is just a request for more info since I lack seeing it in the docs themselves. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7432/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7421

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7421/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7421/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7421/events

https://github.com/kubeflow/pipelines/issues/7421

I_kwDOB-71UM5FwqbC

[backend] "Updating" a recurring run, whilst a recurring run is in progress causes it to terminate

open

CONTRIBUTOR

### Environment * How did you deploy Kubeflow Pipelines (KFP)? <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> Atop AWS using the `kubeflow/manifests` official distribution with some overlays specific to Zillow. * KFP version: <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> We use KF v1.3.1 currently (plan to update to v1.5 next month), looks like that includes KFP v1.5.1 ([link](https://github.com/kubeflow/manifests/blob/v1.3.1/apps/pipeline/upstream/base/pipeline/kustomization.yaml#L43)). * KFP SDK version: <!-- Specify the output of the following shell command: $pip list | grep kfp --> We run a forked version ([see here](https://github.com/zillow/pipelines)), I think we last pulled in at ~v1.8. ### Steps to reproduce <!-- Specify how to reproduce the problem. This may include information such as: a description of the process, code snippets, log output, or screenshots. --> The problem is that via CICD we want our scientists to be able to define a `recurring-run` and have it updated on subsequent CICD executions if they change their settings. Given the limitations of the "Job" API in KFP we delete and re-add to "update" a recurring run, (see the [API reference](https://www.kubeflow.org/docs/components/pipelines/reference/api/kubeflow-pipeline-api-spec/#tag-JobService) for more info). What happens in this case if the first recurring run has triggered a workflow then the scheduledworkflow controller deletes the `ScheduledWorkflow` and then via Kubernetes garbage collection it cascades and deletes any in-progress `Workflow` resources. Then when the "new" updated recurring run is created it needs to wait for the new trigger point to schedule the next run, it won't try to resurrect the now terminated run (and it definitely shouldn't thinking about it). ### Expected result <!-- What should the correct behavior be? --> Basically some way of keeping the run that was in progress before the "update" occurred. Potentially when deleting a recurring run if there was a flag available in the API/SDK/CLI to say to [orphan on deletion](https://kubernetes.io/docs/tasks/administer-cluster/use-cascading-deletion/#set-orphan-deletion-policy) that'd solve this issue and you'd not need a complicated update API. ### Materials and Reference <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> Apologies, maybe this is more of a "feature" request thinking about it now. Apologies for the mischaracterization! --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7421/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7420

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7420/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7420/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7420/events

https://github.com/kubeflow/pipelines/issues/7420

I_kwDOB-71UM5Fwhke

[SDK] Pipeline runs web UI shows the wrong namespace

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? As part of a full Kubeflow 1.4 install on GKE. * KFP version: 1.7.0, packaged with Kubeflow 1.4 ### Steps to reproduce 1. Use the KFP Python SDK (for example, `create_run_from_pipeline_func()` to create a pipeline from a Jupyter Notebook. Use the `namespace` argument to create this pipeline run in a namespace that's not your own, but in which you are a collaborator. 2. The notebook will display a link to the "run details" page in the web UI of the following form: https://<cluster-domain>/pipeline/#/runs/details/<uuid> 3. Open the link. 4. The pipeline run details page is shown, but the namespace selector in the top-left corner shows your own default namespace, not the namespace in which the pipeline was actually run. ### Expected result The pipeline run details should be shown, and the namespace selector in the top-left corner should show the namespace in which the pipeline was run (not your own namespace). Thank you for your help debugging this issue! --- Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7420/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7416

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7416/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7416/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7416/events

https://github.com/kubeflow/pipelines/issues/7416

I_kwDOB-71UM5Fr1JG

[bug] google-cloud-pipeline-components.readthedocs.io now points to 1.0.1 version instead of 1.0.0

closed

NONE

### What steps did you take I went to the main documentation URL: https://google-cloud-pipeline-components.readthedocs.io/ ### What happened: I was directed to this URL, which doesn't exist: https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-1.0.1/ ### What did you expect to happen: I expected to be taken to this URL, which does exist - and which worked perfectly last week: https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-1.0.0/ ### Environment: Google Chrome Version 99.0.4844.57 (Official Build) (64-bit) ### Anything else you would like to add: <!-- Miscellaneous information that will assist in solving the issue.--> ### Labels <!-- Please include labels below by uncommenting them to help us better triage issues --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area testing --> <!-- /area samples --> <!-- /area components --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7416/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7413

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7413/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7413/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7413/events

https://github.com/kubeflow/pipelines/issues/7413

I_kwDOB-71UM5FraSW

[sdk] Code style/formatting standardization

closed

MEMBER

This issue describes steps for improving the linting and formatting of our codebase. Action items: - [ ] Change source code files to be `pylint` compliant and align `.pylintrc` with the desired state of our source code files - [x] Format all files with `yapf` - [ ] Create (a) CI/CD workflow(s) to hold linting, formatting, and other code checks - [x] Provide developers with tooling to help make their PRs compliant (scripts, tox, etc.) /area sdk /cc @chensun /cc @ji-yaqi

https://api.github.com/repos/kubeflow/pipelines/issues/7413/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7389

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7389/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7389/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7389/events

https://github.com/kubeflow/pipelines/issues/7389

I_kwDOB-71UM5FVz9c

[bug] google_cloud_pipeline_components ModelBatchPredictOp()

open

NONE

### What steps did you take: I was deploying a pipeline to Vertex AI that was doing Model Upload and Batch Prediction jobs. But I got an error when the pipeline was executing the Batch Prediction Job. The code is below: ``` from kfp.v2.google.client import AIPlatformClient from google_cloud_pipeline_components import aiplatform as gcc_aip from kfp.v2 import compiler, dsl @dsl.pipeline(name="sample-pipeline", pipeline_root=pipeline_root) def pipeline(): model_upload_op = gcc_aip.ModelUploadOp( project=GCP_PROJECT_ID, display_name=model_display_name, artifact_uri=model_artifact_uri, serving_container_image_uri=PREDICTION_IMAGE, serving_container_ports = [{"container_port": 8080}], serving_container_predict_route = "/predict", serving_container_health_route = "/health", labels=model_metadata ) gcc_aip.ModelBatchPredictOp( project=GCP_PROJECT_ID, job_display_name=pipeline_id, model=model_upload_op.outputs["model"], gcs_source_uris=batch_prediction_data_uri, gcs_destination_output_uri_prefix=batch_prediction_results_uri, instances_format="csv", predictions_format="jsonl", machine_type = "n1-standard-2", starting_replica_count=1, max_replica_count=1, ) compiler.Compiler().compile(pipeline_func=pipeline, package_path="pipeline.json") api_client = AIPlatformClient(project_id=GCP_PROJECT_ID, region=GCP_REGION) response = api_client.create_run_from_job_spec( 'pipeline.json', pipeline_root=pipeline_root ) ``` <!-- A clear and concise description of what the bug is.--> ### What happened: The error messages are below: ``` Info 2022-03-07T21:29:52.083653660ZINFO:root:Job started for type: BatchPredictionJob Error 2022-03-07T21:29:52.115350265ZTraceback (most recent call last): Error 2022-03-07T21:29:52.115389275Z File "/opt/python3.7/lib/python3.7/runpy.py", line 193, in _run_module_as_main Error 2022-03-07T21:29:52.115585620Z "__main__", mod_spec) Error 2022-03-07T21:29:52.115634257Z File "/opt/python3.7/lib/python3.7/runpy.py", line 85, in _run_code Error 2022-03-07T21:29:52.115704257Z exec(code, run_globals) Error 2022-03-07T21:29:52.115712908Z File "/opt/python3.7/lib/python3.7/site-packages/google_cloud_pipeline_components/container/v1/gcp_launcher/launcher.py", line 229, in <module> Error 2022-03-07T21:29:52.115924803Z main(sys.argv[1:]) Error 2022-03-07T21:29:52.115938316Z File "/opt/python3.7/lib/python3.7/site-packages/google_cloud_pipeline_components/container/v1/gcp_launcher/launcher.py", line 225, in main Error 2022-03-07T21:29:52.116124076Z _JOB_TYPE_TO_ACTION_MAP[job_type](**parsed_args) Error 2022-03-07T21:29:52.116137398Z File "/opt/python3.7/lib/python3.7/site-packages/google_cloud_pipeline_components/container/v1/gcp_launcher/batch_prediction_job_remote_runner.py", line 95, in create_batch_prediction_job Error 2022-03-07T21:29:52.116316276Z insert_artifact_into_payload(executor_input, payload)) Error 2022-03-07T21:29:52.116362741Z File "/opt/python3.7/lib/python3.7/site-packages/google_cloud_pipeline_components/container/v1/gcp_launcher/batch_prediction_job_remote_runner.py", line 47, in insert_artifact_into_payload Error 2022-03-07T21:29:52.116380612Z job_spec = json.loads(payload) Error 2022-03-07T21:29:52.116401022Z File "/opt/python3.7/lib/python3.7/json/__init__.py", line 348, in loads Error 2022-03-07T21:29:52.116745671Z return _default_decoder.decode(s) Error 2022-03-07T21:29:52.116771230Z File "/opt/python3.7/lib/python3.7/json/decoder.py", line 337, in decode Error 2022-03-07T21:29:52.116975231Z obj, end = self.raw_decode(s, idx=_w(s, 0).end()) Error 2022-03-07T21:29:52.117005384Z File "/opt/python3.7/lib/python3.7/json/decoder.py", line 355, in raw_decode Error 2022-03-07T21:29:52.117186534Z raise JSONDecodeError("Expecting value", s, err.value) from None Error 2022-03-07T21:29:52.117214480Zjson.decoder.JSONDecodeError: Expecting value: line 1 column 210 (char 209) Info 2022-03-07T21:30:05.776338368ZJob is running. Error 2022-03-07T21:30:05.834539659ZThe replica workerpool0-0 exited with a non-zero status of 1. ``` ### What did you expect to happen: The Vertex AI Batch Prediction Job can be executed correctly through the pre-built component ModelBatchPredictOp(). ### Environment: <!-- Please fill in those that seem relevant. --> * How do you deploy Kubeflow Pipelines (KFP)? I was using Vertex AI Notebook Instance and deploy the pipeline to Vertex AI Pipelines <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: v2 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: 1.8.11 <!-- Specify the output of the following shell command: $pip list | grep kfp --> * google-cloud-pipeline-components version: 1.0.0 ### Anything else you would like to add: <!-- Miscellaneous information that will assist in solving the issue.--> ### Labels <!-- Please include labels below by uncommenting them to help us better triage issues --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area testing --> <!-- /area samples --> <!-- /area components --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7389/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7382

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7382/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7382/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7382/events

https://github.com/kubeflow/pipelines/issues/7382

I_kwDOB-71UM5FQp2_

[PROPOSAL] A Generic Protocol for Kubeflow Pipeline Template Registry

closed

CONTRIBUTOR

## Introduction We are working on a generic protocol design of Kubeflow Pipeline Template Registry. At high level, the protocol will define the generic APIs between KFP SDK client and a template registry server to - upload and download template (as YAML file or ZIP), and - organize templates as versioned objects. Then, KFP SDK will implement a registry client class that can work with any registry server that implements the APIs defined in the generic protocol. Feedbacks are welcomed :-) ## Resource Hierarchy  The protocol will enable organizing templates as versioned objects in the registry, following the hierarchy of - Host (`http://my-registry.server/kubeflow-pipelines/public/`) - Package (`mnist-demo-workflow`) - Version (`sha256:e54fda…`) - Tag (`latest`, `v2`) Note - Version name will use sha256 digest of the uploaded content. - Tags are nicknames for versions, uniquely identifying a version under a package. - Tags are unique among versions - the same tag can only apply to a single version under a package. - Multiple tags can apply to the same version. - Tags are not fixed - an existing tag can be removed from the existing version, or be updated to point to another version. ## API Design Please visit [Generic Protocol for Kubeflow Pipelines Template Registry](https://docs.google.com/document/d/1Lc4NqN0VYIO3CHUPb_6LRIDF4VAWYy23MfpHYM8SncA/edit#heading=h.x9snb54sjlu9) for the detailed design. If you don't have permission to access the doc, please join https://groups.google.com/g/kubeflow-discuss for access.

https://api.github.com/repos/kubeflow/pipelines/issues/7382/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7381

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7381/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7381/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7381/events

https://github.com/kubeflow/pipelines/issues/7381

I_kwDOB-71UM5FQP4d

[feature] Add the ability to decide if we want an Artifact to be uploaded to the remote or not

closed

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> /area frontend /area backend /area sdk ### What feature would you like to see? I would like to be able to decide if a component Output/Input should be uploaded to the remote or not. The `Artifact` would always be registered by MLMD but the upload to the remote could be optional. <!-- Provide a description of this feature and the user experience. --> We could add a flag to the kfp sdk `shouldUploadArtifactsToRemote` that controls this for a specific artifact input/output. ### What is the use case or pain point? The use case is that we have huge datasets we don't necessarily want to duplicate on the remote. We perform multiple processing on the dataset that can easily be re-done to get the exact same dataset since KFP records all the pipeline parameters that were used for the run (`random_seed`, `dataset_name`, `dataset_version` etc). We use https://github.com/iterative/ldb-resources to version our datasets. <!-- It helps us understand the benefit of this feature for your use case. --> This feature would prevent overloading the remote with duplicates of datasets, if we want to get back a dataset that was used, we can simply refer to the pipeline params and in our case we'll be good to go. It would also speed up our pipelines considerabely since we wouldn't have to wait for the re-upload between every components. This might not be the case for everyone, but we should have the choice. ### Is there a workaround currently? The only workaround I see so far would be to manually create folders for our datasets, instead of using `Output/Input[Dataset]`, to pass the `my_dataset_path: str` between components and use PVCs to share storage. This is far from ideal and we won't even use this workaround since PVCs would make the pipeline not portable. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7381/timeline

not_planned

https://api.github.com/repos/kubeflow/pipelines/issues/7379

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7379/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7379/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7379/events

https://github.com/kubeflow/pipelines/issues/7379

I_kwDOB-71UM5FMXhB

Update the feature flags in localStorage if the new version has more feature flags than the stored one

closed

COLLABORATOR

## Description Currently, we have a hidden page which allows user to enable/disable features which are under development. The URL path is `<KFP_HOST_ADDRESS>/frontend_features`. We call it Feature Flag page, refer to [Large features development](https://github.com/kubeflow/pipelines/tree/master/frontend#large-features-development) documentation. ## Problem Currently, if the feature flag map called `flags` already exists in localStorage, it will be automatically loaded to window object: Refer to https://github.com/kubeflow/pipelines/blob/7ad949106f9310a1bf3c5ff960e2c6274cd042d5/frontend/src/features.ts#L31-L34. However, we will likely be adding new flags when we release KFPv2 for users to try. That means we need to overwrite the existing flag map with `default key list` by new KFP frontend code, while keeping the current value of existing key from the local storage. ## Approach Add a logic to merge existing flag map in localStorage with the default flag map hardcoded by frontend codebase: - If a key doesn't exist in localStorage: Add this key with default value. - If a key already exist in localStorage: Keep this key and keep the existing value. - If a key exists in localStorage, but not in default flag map: Delete this key. After this merge, save the merged flag map back to localStorage and window object. *Note:* We should never change the default value of an existing key, and never add back a key which has been deprecated. Because it will cause inconsistency to user experience. We should update documentation to keep a record about past deprecated keys, and warn developers about the risk of changing default value. ## Reference - [Large features development](https://github.com/kubeflow/pipelines/tree/master/frontend#large-features-development) documentation. - [window.localStorage](https://developer.mozilla.org/en-US/docs/Web/API/Window/localStorage)

https://api.github.com/repos/kubeflow/pipelines/issues/7379/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7378

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7378/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7378/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7378/events

https://github.com/kubeflow/pipelines/issues/7378

I_kwDOB-71UM5FKYhQ

ValueError: Invalid value for `image_pull_policy` (None), must be one of ['Always', 'IfNotPresent', 'Never']

open

NONE

* KFP version: kfp 1.8.11 kfp-pipeline-spec 0.1.13 kfp-server-api 1.7.1 I am trying to create a Kubeflow pipeline using python SDK v2 and when I am trying to run the code on the pipeline I am getting "Image_Pull_Policy" error. ``` import kfp from caffe2.perfkernels.hp_emblookup_codegen import o from kfp import dsl from kfp.v2.dsl import component from kfp.v2.dsl import ( Input, Output, Artifact, Dataset, ) @component def dataloader_op(datasets:Output[Artifact]): op = kfp.components.load_component_from_text(''' name: dataloader implementation: container: image: racahu23/ml-blueprint_dataloader:8 imagePullPolicy: Always command: - python - /home/user/dataloader.py - "(--platform aws)" - "(--bucketname new-classification)" - "(--remoteDirectoryName datasets) - {outputPath: /home/user/datasets} ''') op.container.set_image_pull_policy("Always") op.container.set_termination_message_policy('FallbackToLogsOnError') return op @component() def datapreprocessor(datasets:Dataset, Data: Output[Artifact]): op = kfp.components.load_component_from_text(''' name: datapreprocessor implementation: container: image: racahu23/ml-blueprint_preprocessor:7 imagePullPolicy: Always command: - python - "$(--input_dir datasets)" - "$(--output_dir Data)" - {outputPath: /home/user/Data} ''') op.container.set_image_pull_policy("Always") op.container.set_termination_message_policy('FallbackToLogsOnError') return op @component( base_image="docker.io/kubeflowkatib/kubeflow-pipelines-launcher", output_component_file="katib_training_component.yaml" ) def training(Data) -> bool: import os import json experiment_spec = { "algorithm": { "algorithmName": "random" }, "maxFailedTrialCount": 1, "maxTrialCount": 2, "objective": { "goal": 0.99, "objectiveMetricName": "Accuracy", "type": "maximize" }, "parallelTrialCount": 1, "metricsCollectorSpec": { "source": { "fileSystemPath": { "path": "/katib/training.log", "kind": "File" } }, "collector": { "kind": "File" } }, "parameters": [ { "feasibleSpace": { "max": "32", "min": "32" }, "name": "batch_size", "parameterType": "int" } ], "trialTemplate": { "primaryContainerName": "training-container", "retain": "true", "trialParameters": [ { "description": "Number of estimators for the training model", "name": "batch_size", "reference": "batch_size" } ], "trialSpec": { "apiVersion": "batch/v1", "kind": "Job", "spec": { "template": { "metadata": { "annotations": { "sidecar.istio.io/inject": "false" } }, "spec": { "containers": [ { "command": [ "python3", "/home/user/primary_trainer.py", "--input_dir" +" " + str(Data), "--num_labels 9", "--logging_dir logs/", '--num_train_epochs 1', '--evaluation_strategy epoch', '--per_device_train_batch_size ${trialParameters.batchSize}', '--per_device_eval_batch_size 64', '--save_strategy epoch', '--logging_strategy epoch', '--eval_steps 100' ], "image": "racahu23/ml-blueprint_trainer:12", "imagePullPolicy": "Always", "name": "training-container" } ], "restartPolicy": "Never" } } } } } } experiment_spec = json.dumps(experiment_spec) exec_str = f""" python src/launch_experiment.py \ --experiment-name katib-pipeline \ --experiment-namespace <namespace> \ --experiment-spec '{experiment_spec}' \ --experiment-timeout-minutes '60' \ --delete-after-done "False" \ --output-file /tmp/outputs/Best_Parameter_Set/data \ """ os.system(exec_str) return True @dsl.pipeline(name="katib_experiment_pipeline") def training_pipeline( ): _dataloader_op = dataloader_op() _preprocessor_op = datapreprocessor((_dataloader_op.outputs["datasets"])) training_task = (training(_preprocessor_op.outputs["Data"])) arguments = { } client = kfp.Client(namespace="kubeflow", host="http://localhost:8080") client.create_run_from_pipeline_func(training_pipeline,arguments=arguments,namespace= "kubeflow",mode = kfp.dsl.PipelineExecutionMode.V2_COMPATIBLE,enable_caching = False) ``` This is the error that I am getting: <img width="1573" alt="Screen Shot 2022-03-05 at 14 15 36" src="https://user-images.githubusercontent.com/51020974/156884731-01e036b3-4cdd-48f7-b49a-74d67f887af7.png"> I have been struggling quite a bit in fixing this issue. Any help would be really appreciated.

https://api.github.com/repos/kubeflow/pipelines/issues/7378/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7370

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7370/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7370/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7370/events

https://github.com/kubeflow/pipelines/issues/7370

I_kwDOB-71UM5E-DFb

[feature] Add an endpoint that restore experiment and its runs at once

open

MEMBER

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> /area backend <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? Add an endpoint that restores Experiment and all of its Runs at once ```/apis/v1beta1/experiments/{id}/runs:unarchive``` <!-- Provide a description of this feature and the user experience. --> ### What is the use case or pain point? Provide an option to the user to restore an Experiment and all of its Runs in a single click. For more context took a look at this comment - https://github.com/kubeflow/pipelines/issues/5114#issuecomment-777932647 This endpoint will be leveraged by [this](https://github.com/kubeflow/pipelines/issues/7335) frontend issue. <!-- It helps us understand the benefit of this feature for your use case. --> ### Is there a workaround currently? Yes, do it with multiple steps. <!-- Without this feature, how do you accomplish your task today? --> --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7370/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7369

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7369/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7369/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7369/events

https://github.com/kubeflow/pipelines/issues/7369

I_kwDOB-71UM5E97VY

[feature] Create parent directories of `output_component_file`

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> /area sdk ### What feature would you like to see? I would like `output_component_file` file hierarchy to be created i.e. create parent directories if they do not exist. The example below raises `FileNotFoundError` because the parent directories do not exist. **Code example:** ```python from kfp.v2.dsl import pipeline, component @component( output_component_file="build/components/test_component.yaml", ) def test_component() -> None: ... @pipeline(name="test-pipe") def test_pipeline() -> None: task = test_component() if __name__ == "__main__": from kfp.v2 import compiler compiler.Compiler().compile( pipeline_func=test_pipeline, package_path="build/pipeline.json" ) ``` **Traceback:** ```plain Traceback (most recent call last): File "test.py", line 7, in <module> def test_component() -> None: File "/home/gmdev/.cache/pypoetry/virtualenvs/whatever-Ez6qCvWa-py3.8/lib/python3.8/site-packages/kfp/v2/components/component_decorator.py", line 104, in component return component_factory.create_component_from_func( File "/home/gmdev/.cache/pypoetry/virtualenvs/whatever-Ez6qCvWa-py3.8/lib/python3.8/site-packages/kfp/v2/components/component_factory.py", line 440, in create_component_from_func component_spec.save(output_component_file) File "/home/gmdev/.cache/pypoetry/virtualenvs/whatever-Ez6qCvWa-py3.8/lib/python3.8/site-packages/kfp/components/_structures.py", line 494, in save with open(file_path, 'w') as f: FileNotFoundError: [Errno 2] No such file or directory: 'build/components/test_component.yaml' ``` ### What is the use case or pain point? Kubeflow SDK will try to generate the component file definition automatically, so if the value of `output_component_file` contains directories that do not exist it will raise `FileNotFoundError`. ### Is there a workaround currently? Yes, you can create the directory hierarchy just before `kfp` is imported to avoid the issue, but this solution does not seem very clean to me. ```python from pathlib import Path Path("build/components").mkdir(exist_ok=True, parents=True) from kfp.v2.dsl import pipeline, component ... ``` Maybe it would be a good idea to add in the method below to check if the parent directories of `file_path` exist and if they don't, then create them. https://github.com/kubeflow/pipelines/blob/6965dbac2faee0411fc0ca9565fd9a9d7ef8e2bf/sdk/python/kfp/components/_structures.py#L486-L495 I would love to start contributing to this repo, and if this feature proposal is finally accepted I would love to create a PR. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7369/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7368

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7368/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7368/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7368/events

https://github.com/kubeflow/pipelines/issues/7368

I_kwDOB-71UM5E9z8Q

[backend] DeletePipeline does not clean PipelienVersions data from Minio

open

MEMBER

When a Pipeline has Pipeline Versions for each version there is a record in the Object Store (Minio). On delete Pipeline all the records (Pipeline and its Versions) are deleted from the database(MySQL), but not all are deleted from Minio. Minio records for Pipeline Versions are not deleted. ### Environment Linux Ubuntu - 20.04.3 LTS kind v0.11.1 go1.16.4 linux/amd64 kustomize version: {KustomizeVersion:3.2.0 GitCommit:a3103f1e62ddb5b696daa3fd359bb6f2e8333b49 BuildDate:2019-09-18T16:26:36Z GoOs:linux GoArch:amd64} Docker Engine 20.10.12 * How did you deploy Kubeflow Pipelines (KFP)? Full Kubeflow deployment ```while ! kustomize build example | kubectl apply -f -; do echo "Retrying to apply resources"; sleep 10; done``` with manifest [commit](https://github.com/kubeflow/manifests/tree/bcc0e4f8bd54977aa62f9dae1c41ad47d62524c5) <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: Kubeflow Pipelines apps/pipeline/upstream [1.7.0](https://github.com/kubeflow/pipelines/tree/1.7.0/manifests/kustomize) **But the problem exists with the latest Pipeline version as well!** * KFP SDK version: <!-- Specify the output of the following shell command: $pip list | grep kfp --> ### Steps to reproduce 1. Create a Pipeline with Pipeline versions 2. Delete the Pipeline 3. Check Minio - Minio records associated with the Pipeline Versions are not deleted. <!-- Specify how to reproduce the problem. This may include information such as: a description of the process, code snippets, log output, or screenshots. --> ### Expected result Minio records associated with the Pipeline Versions to be deleted as well <!-- What should the correct behavior be? --> ### Materials and Reference In [delete Pipeline function](https://github.com/kubeflow/pipelines/blob/6965dbac2faee0411fc0ca9565fd9a9d7ef8e2bf/backend/src/apiserver/resource/resource_manager.go#L224) there is a comment that clearly says - that the current implementation of this func supports only Pipelines with a **single** Version. Today, **multiple** Pipeline Versions are supported and it is time to be handled appropriately when Deleting a Pipeline. ``` // Delete pipeline file and DB entry. // Not fail the request if this step failed. A background run will do the cleanup. // https://github.com/kubeflow/pipelines/issues/388 // TODO(jingzhang36): For now (before exposing version API), we have only 1 // file with both pipeline and version pointing to it; so it is ok to do // the deletion as follows. After exposing version API, we can have multiple // versions and hence multiple files, and we shall improve performance by // either using async deletion in order for this method to be non-blocking // or or exploring other performance optimization tools provided by gcs. err = r.objectStore.DeleteFile(r.objectStore.GetPipelineKey(fmt.Sprint(pipelineId))) if err != nil { glog.Errorf("%v", errors.Wrapf(err, "Failed to delete pipeline file for pipeline %v", pipelineId)) return nil } err = r.pipelineStore.DeletePipeline(pipelineId) ``` <!-- Help us debug this issue by providing resources such as sample code, background context, or links to references. --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7368/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7363

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7363/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7363/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7363/events

https://github.com/kubeflow/pipelines/issues/7363

I_kwDOB-71UM5E4jrH

[sample] Error running xgboost_training_cm.py

closed

COLLABORATOR

### Environment * How did you deploy Kubeflow Pipelines (KFP)? GCP marketplace <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: 1.7.1 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: <!-- Specify the output of the following shell command: $pip3 list | grep kfp --> ``` kfp 1.8.11 kfp-pipeline-spec 0.1.13 kfp-server-api 1.8.1 ``` ### Steps to reproduce <!-- Specify how to reproduce the problem. This may include information such as: a description of the process, code snippets, log output, or screenshots. --> Follow the steps given in the following link to deploy `xgboost_training_cm.py` https://github.com/kubeflow/pipelines/tree/master/samples/core/xgboost_training_cm ### Expected result <!-- What should the correct behavior be? --> There shows an error with `dataproc-create-cluster` (the second block) ``` Traceback (most recent call last): File "/usr/local/lib/python3.7/runpy.py", line 193, in _run_module_as_main "__main__", mod_spec) File "/usr/local/lib/python3.7/runpy.py", line 85, in _run_code exec(code, run_globals) File "/ml/kfp_component/launcher/__main__.py", line 45, in <module> main() File "/ml/kfp_component/launcher/__main__.py", line 42, in main launch(args.file_or_module, args.args) File "/ml/kfp_component/launcher/launcher.py", line 45, in launch return fire.Fire(module, command=args, name=module.__name__) File "/usr/local/lib/python3.7/site-packages/fire/core.py", line 127, in Fire component_trace = _Fire(component, args, context, name) File "/usr/local/lib/python3.7/site-packages/fire/core.py", line 366, in _Fire component, remaining_args) File "/usr/local/lib/python3.7/site-packages/fire/core.py", line 542, in _CallCallable result = fn(*varargs, **kwargs) File "/ml/kfp_component/google/dataproc/_create_cluster.py", line 76, in create_cluster client = DataprocClient() File "/ml/kfp_component/google/common/_utils.py", line 170, in __init__ self._build_client() TypeError: _build_client() takes 0 positional arguments but 1 was given ``` ### Materials and Reference <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> N/A --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7363/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7361

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7361/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7361/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7361/events

https://github.com/kubeflow/pipelines/issues/7361

I_kwDOB-71UM5E0iQO

[backend] "Signal command failed: command terminated with exit code 1" when terminating a pipeline

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)?: As part of a full Kubeflow 1.4 deployment on GKE. * KFP version: 1.7.0, packaged with Kubeflow 1.4 * KFP SDK version: N/A ### Steps to reproduce 1. Create a new pipeline run. 2. Click the "Terminate" link in the KFP web UI. 3. Observe that the currently-running pod runs to completion before termination. ### Expected result In step 3 above, the currently-running pod should immediately terminate. ### Materials and Reference I have been able to determine the following so far: 1. When the "Terminate" button is clicked, KFP adds `activeDeadlineSeconds: 0` to the spec of the workflow being terminated. This is happening as expected. 2. The Argo Worfklows controller notices this and attempts to kill the currently-running pod by executing `sh -c kill -s USR2 $(pidof argoexec)` in the `main` container of the running pod. This causes the Argo Workflows controller to log the following error: ``` time="2022-02-25T21:54:54.186Z" level=info msg="Applying sooner Workflow Deadline for pod emenendez-taxi-x5q66-345673846 at: 2022-02-25 21:53:08 +0000 UTC" namespace=emenendez workflow=emenendez-taxi-x5q66 time="2022-02-25T21:54:54.186Z" level=info msg="Updating execution control of emenendez-taxi-x5q66-345673846: {\"deadline\":\"2022-02-25T21:53:08Z\"}" namespace=emenendez workflow=emenendez-taxi-x5q66 time="2022-02-25T21:54:54.244Z" level=info msg="Patch pods 200" time="2022-02-25T21:54:54.246Z" level=info msg="Signalling emenendez-taxi-x5q66-345673846 of updates" namespace=emenendez workflow=emenendez-taxi-x5q66 time="2022-02-25T21:54:54.247Z" level=info msg="https://7.255.204.1:443/api/v1/namespaces/emenendez/pods/emenendez-taxi-x5q66-345673846/exec?command=sh&command=-c&command=kill+-s+USR2+%24%28pidof+argoexec%29&container=main&stderr=true&stdout=true&tty=false" time="2022-02-25T21:54:54.292Z" level=info msg="Create pods/exec 101" time="2022-02-25T21:54:54.390Z" level=warning msg="Signal command failed: command terminated with exit code 1" namespace=emenendez workflow=emenendez-taxi-x5q66 ``` This appears to be a bug with the Argo Workflows controller -- instead of executing `sh -c kill -s USR2 $(pidof argoexec)` in the `main` container, it should execute that command in the *`wait`* container. It appears this bug was introduced in https://github.com/argoproj/argo-workflows/pull/5099, and fixed as part of an unrelated refactor in https://github.com/argoproj/argo-workflows/pull/6022, which is included in Argo Workflows 3.2.0. The specific buggy code is [this block](https://github.com/argoproj/argo-workflows/blob/release-3.1/workflow/controller/exec_control.go#L80-L96), which iterates over the main containers in a pod and sends the USR2 signal there. Interestingly enough, I could not find an issue related to this bug in the Argo Workflows project. Questions: 1. Are there any other workarounds or fixes for the inability to terminate a running pipeline pod other than upgrading Argo Workflows? 2. If not, could Argo Workflows be updated to a version with the fix? Thanks so much! --- Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7361/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7358

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7358/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7358/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7358/events

https://github.com/kubeflow/pipelines/issues/7358

I_kwDOB-71UM5Exj5P

[bug] Component input "project_id" with type "GCPProjectID"

open

NONE

### What happened: I getting this error when use the Dataproc component, there is a way to pass the project_id like a string? ``` TypeError: Passing value "saasbcn-98ce0" with type "String" (as "Parameter") to component input "project_id" with type "GCPProjectID" (as "Artifact") is incompatible. Please fix the type of the component input. ``` <!-- /area sdk --> <!-- /area components --> Best regards, Victor

https://api.github.com/repos/kubeflow/pipelines/issues/7358/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7356

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7356/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7356/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7356/events

https://github.com/kubeflow/pipelines/issues/7356

I_kwDOB-71UM5EvL7z

Best data versioning tool/framework with kfp for seamless integration.

closed

NONE

Hi, our team has started implementing kfp for mlops, and we are researching on best methods to data version. We primarily work on **images** but not limited to it. Which would be the best data versioning framework which works along with kfp with seamless integration and featureset.

https://api.github.com/repos/kubeflow/pipelines/issues/7356/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7347

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7347/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7347/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7347/events

https://github.com/kubeflow/pipelines/issues/7347

I_kwDOB-71UM5EdBdF

GCP Vertex AI, deploying model to existing endpoint

open

NONE

### Feature Area /area sdk /area components I currently have a simple pipeline that deploys to Vertex AI that takes an existing model container image from Artifact Registry, uploads it to the Vertex AI model store, creates an endpoint and deploys the model to that endpoint. So far so good. Now say I have a new revision of that model and I want to deploy that to the same endpoint with a 50/50 traffic split between the old and new version. This is achievable relatively simply in the Vertex AI gui but how do I do this in a kubeflow pipeline? How am I able to get either an existing model (already uploaded) or an already created endpoint to pass to the ModelDeployOp? The following code snippet does not work as the returned endpoint type doesnt plug into the operation. ``` def pipeline( project: str = PROJECT_ID, model_display_name: str = MODEL_DISPLAY_NAME, serving_container_image_uri: str = IMAGE_URI, ): train_task = print_op("No training to be done here!") model_upload_op = gcc_aip.ModelUploadOp( project=project, location=REGION, display_name=model_display_name, # artifact_uri=WORKING_DIR, serving_container_image_uri=serving_container_image_uri, serving_container_ports=[{"containerPort": 8000}], serving_container_predict_route="/hello_world", serving_container_health_route="/health", ) endpoints = aip.Endpoint.list( filter=f"display_name={ENDPOINT_NAME}", order_by="create_time" ) existing_endpoint = endpoints[0] existing_endpoint = existing_endpoint model_deploy_op = gcc_aip.ModelDeployOp( endpoint=existing_endpoint, model=model_upload_op.outputs["model"], deployed_model_display_name=model_display_name, dedicated_resources_machine_type="n1-standard-4", dedicated_resources_min_replica_count=1, dedicated_resources_max_replica_count=1, traffic_split=traffic_split, ) ``` <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7347/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7346

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7346/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7346/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7346/events

https://github.com/kubeflow/pipelines/issues/7346

I_kwDOB-71UM5EbiWi

[components] Google Cloud Custom Job missing option in create_custom_training_job_op_from_component

closed

NONE

Looks like option reserved_ip_ranges is missing in create_custom_training_job_op_from_component() in utils. It was only added to components.yaml.

https://api.github.com/repos/kubeflow/pipelines/issues/7346/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7345

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7345/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7345/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7345/events

https://github.com/kubeflow/pipelines/issues/7345

I_kwDOB-71UM5EbcXA

[sdk] Can't use create_component_from_func with pip packages when running as non-root

open

NONE

### Environment * KFP version: 1.7 (KF 1.4) * KFP SDK version: 1.6.6 * All dependencies version: ``` kfp 1.6.6 kfp-pipeline-spec 0.1.13 kfp-server-api 1.7.1 ``` ### Steps to reproduce Background: * Due to security concerns it's a bad idea to run containers as root. * For composability and maintenance, it's a good idea to define small & modular KFP components. With this in mind, I wish to report that `create_components_from_func` does not work as expected when the container is run as a non-root user and when the `packages_to_install` parameter is used to add some runtime dependencies. To reproduce, see attached pipeline definition at the bottom. When this pipeline is run, the following output is seen in Kubeflow: ``` WARNING: The directory '/.cache/pip' or its parent directory is not owned or is not writable by the current user. The cache has been disabled. Check the permissions and owner of that directory. If executing pip with sudo, you should use sudo's -H flag. ERROR: Could not install packages due to an OSError: [Errno 13] Permission denied: '/.local' Check the permissions. WARNING: The directory '/.cache/pip' or its parent directory is not owned or is not writable by the current user. The cache has been disabled. Check the permissions and owner of that directory. If executing pip with sudo, you should use sudo's -H flag. ERROR: Could not install packages due to an OSError: [Errno 13] Permission denied: '/.local' Check the permissions. Error: exit status 1 ``` ### Expected result The correct behaviour here would be for packages to be installed in a location that's writable by non-root users. As a direct consequence, that location would also to have to be added to PYTHONPATH. With the attached pipeline definition, `kfp.components._python_op._get_packages_to_install_command` today produces the following yaml: ``` - sh - -c - (PIP_DISABLE_PIP_VERSION_CHECK=1 python3 -m pip install --quiet --no-warn-script-location 'tqdm' || PIP_DISABLE_PIP_VERSION_CHECK=1 python3 -m pip install --quiet --no-warn-script-location 'tqdm' --user) && "$0" "$@" - sh - -ec - | program_path=$(mktemp) printf "%s" "$0" > "$program_path" python3 -u "$program_path" "$@" - | def hello_world(): import tqdm print("Hello world!") import argparse _parser = argparse.ArgumentParser(prog='Hello world', description='') _parsed_args = vars(_parser.parse_args()) _outputs = hello_world(**_parsed_args) ``` I propose that `kfp.components._python_op._get_packages_to_install_command` is changed to instead output: ``` - sh - -c - (PIP_DISABLE_PIP_VERSION_CHECK=1 python3 -m pip install --quiet --no-warn-script-location 'tqdm' || PIP_DISABLE_PIP_VERSION_CHECK=1 PYTHONUSERBASE=/tmp/pip python3 -m pip install --quiet --no-warn-script-location --cache-dir /tmp/pip-cache 'tqdm' --user) && "$0" "$@" - sh - -ec - | PIP_CUSTOM_FOLDER=$(realpath /tmp/pip/lib/*/site-packages) program_path=$(mktemp) printf "%s" "$0" > "$program_path" PYTHONPATH=$PYTHONPATH:$PIP_CUSTOM_FOLDER python3 -u "$program_path" "$@" - | def hello_world(): import tqdm print("Hello world") import argparse _parser = argparse.ArgumentParser(prog='Hello world', description='') _parsed_args = vars(_parser.parse_args()) _outputs = hello_world(**_parsed_args) ``` This change would accomplish two things: * Allow non-root users to install pip packages on the fly * Allow non-root users to install packages from cache A side note: I don't have the historical context of why KFP first tries to install packages as root and on failure as the current user with `--user`, IMO doing it with `--user` from the beginning would make more sense. But might be missing something :) If you agree with the structure of my proposal, I can work on the change - seems like a pretty small fix. Thanks! ### Materials and Reference #### Pipeline definition ``` import argparse import kfp import kubernetes def hello_world(): import tqdm print("Hello world!") def hello_world_op(): return kfp.components.create_component_from_func(func=hello_world, packages_to_install=['tqdm'])() def pipeline(): component = hello_world_op() user_sc = kubernetes.client.models.V1SecurityContext(run_as_user=1234) component.set_security_context(user_sc) def get_args(): parser = argparse.ArgumentParser() parser.add_argument('--yaml', type=str, required=True) return parser.parse_args() def main(): args = get_args() kfp.compiler.Compiler().compile( pipeline_func=pipeline, package_path=args.yaml) if __name__ == "__main__": main() print(kfp.__version__) ``` --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7345/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7343

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7343/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7343/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7343/events

https://github.com/kubeflow/pipelines/issues/7343

I_kwDOB-71UM5EV6Ux

[feature][question]When output artifacts, is there a way to disable compression?

open

NONE

We have one kubeflow pipeline to train a model and then output a model to Minio storage. Then we are using KServe to fetch the model from Minio and deploy it. But by default, the model is compressed as tgz file in Mino, which makes KServe couldn't load the model. Any guidance on how to solve this? Thanks!!!

https://api.github.com/repos/kubeflow/pipelines/issues/7343/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7341

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7341/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7341/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7341/events

https://github.com/kubeflow/pipelines/issues/7341

I_kwDOB-71UM5ESsb5

[feature] System utilization for the pipeline and its steps

open

NONE

### Feature Area Not sure ### What feature would you like to see? Logging of system utilization to KFP metrics. Pipeline statistics of CPU/RAM/GPU usage over the course of pipeline run could be helpful in experiment comparison from performance point of view. Example how the final results could look like when plotted nicely [wandb-example](https://wandb.ai/borisd13/char-RNN/runs/cw9gnx9z/system ) ### What is the use case or pain point? Monitoring resource utilization by the pipeline and by each step. Would allow to compare the runs in terms of performance/resource requirements. ### Is there a workaround currently? Manual logging or Prometheus? I encountered some Prometheus traces in the repo, but looks like deprecated or abandoned. If there is an existing solution to that problem, please let me know. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7341/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7340

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7340/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7340/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7340/events

https://github.com/kubeflow/pipelines/issues/7340

I_kwDOB-71UM5EPo-H

[feature] Parametrize pipeline using current date and time

open

NONE

Is there any way to parametrize a pipeline with current date and time? I'd like to do something like ```python @pipeline() def my_pipeline( date: str = datetime.now().strftime("%Y_%m_%d-%H:%M:%S.%f")[:-4], ): ... ``` By default I want to set that input argument to the datetime when the pipeline is run. However, for reusability purposes, I'd also like to be able to set it manually (pointing to a previous run). I can go into further details of why I want to do this if needed. Any way I can achieve this? Thanks a lot in advance!

https://api.github.com/repos/kubeflow/pipelines/issues/7340/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7338

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7338/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7338/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7338/events

https://github.com/kubeflow/pipelines/issues/7338

I_kwDOB-71UM5EN_qG

[release] Backend 1.8.1 tracker

closed

COLLABORATOR

Need to cherry-pick the following PRs to release-1.8 branch before making the official release: - [x] https://github.com/kubeflow/pipelines/pull/7337

https://api.github.com/repos/kubeflow/pipelines/issues/7338/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7336

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7336/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7336/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7336/events

https://github.com/kubeflow/pipelines/issues/7336

I_kwDOB-71UM5ELwNV

[backend] Use `default-editor` as serviceAccountName in multi-user mode

closed

COLLABORATOR

### Environment This is an issue I found with KFP backend 1.8.0 on Full fledged Kubeflow. ### Steps to reproduce Run a KFP pipeline on full Kubeflow, the error is ``` This step is in Error state with this message: task 'conditional-execution-pipeline-with-exit-handler-6gfkq.exit-handler-1.flip-coin-op' errored: pods "conditional-execution-pipeline-with-exit-handler-6gfkq-1717566668" is forbidden: error looking up service account jamxl/pipeline-runner: serviceaccount "pipeline-runner" not found ``` ### Materials and Reference We are always using the `pipeline-runner` service account in this file: https://github.com/kubeflow/pipelines/blob/3e734ed19146f569e910f75627d12239ec2e86dc/backend/src/apiserver/template/template.go#L191-L202. Instead, it should use `default-editor` for multi-user mode, the util to check multi-user mode is https://github.com/kubeflow/pipelines/blob/2e945750cb1758eea6db8453b437e57e68152b4a/backend/src/apiserver/common/config.go#L94 <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7336/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7335

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7335/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7335/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7335/events

https://github.com/kubeflow/pipelines/issues/7335

I_kwDOB-71UM5EHzg0

[feature] When restoring an Experiment, add an option to restore all of its Runs

open

MEMBER

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> /area frontend <!-- /area backend --> <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? When restoring an experiment, a popup should show 3 options for the user to choose: `Cancel`, `Restore Experiment` and `Restore Experiment and All Runs`. `Restore Experiment` needs to be highlighted because that is the default behavior. <!-- Provide a description of this feature and the user experience. --> ### What is the use case or pain point? see this [comment](https://github.com/kubeflow/pipelines/issues/5114#issuecomment-777735817) <!-- It helps us understand the benefit of this feature for your use case. --> ### Is there a workaround currently? Yes, currently the user could Restore the Experiment and then restore its Runs. <!-- Without this feature, how do you accomplish your task today? --> --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7335/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7334

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7334/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7334/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7334/events

https://github.com/kubeflow/pipelines/issues/7334

I_kwDOB-71UM5EHM8Z

[feature] Add deletePipelineByName endpoint

open

MEMBER

### Feature Area /area backend <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? Add new endpoint - deletePipelineByName In the case of Pipeline standalone: `/apis/v1beta1/namespaces/-/pipelines/{name}` In the case of full-fledged deployment: `/apis/v1beta1/namespaces/{namespace}/pipelines/{name}` <!-- Provide a description of this feature and the user experience. --> ### What is the use case or pain point? Simplify user experience and keep consistent with other tools like Kubernetes, DAGs in Argo, Airflow that use names as a reference instead of UUID This issue is inspired from #https://github.com/kubeflow/pipelines/issues/3360. <!-- It helps us understand the benefit of this feature for your use case. --> ### Is there a workaround currently? Yes, use UUIDs instead <!-- Without this feature, how do you accomplish your task today? --> --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7334/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7330

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7330/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7330/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7330/events

https://github.com/kubeflow/pipelines/issues/7330

I_kwDOB-71UM5EA2CH

[feature] allow jobs to fail

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> /area backend ### What feature would you like to see? <!-- Provide a description of this feature and the user experience. --> _**Allow failure of jobs**_ - If an operation fails, do not fail the pipeline. Allow the pipeline to continue to the next stage, and there it may fail if that does not have the pre-requisites. ### What is the use case or pain point? <!-- It helps us understand the benefit of this feature for your use case. --> In the machine learning pipelines, it is fairly common to run multiple models, or possibly different configurations of same model, and this possibly runs on a subset of training data. After these are trained, usually they are compared using some metric, the best model is chosen, and that is run on the entire training data to have the final trained model. If someone uses [`kfp.dsl.ParallelFor`](https://kubeflow-pipelines.readthedocs.io/en/latest/source/kfp.dsl.html#kfp.dsl.ParallelFor) to run the different models, failure in one of them causes the entire pipeline to fail and possibly successful training of other models are lost. But if the next stage, the one to compare using metric supports comparison of the available (i.e. successful) models, the pipeline failure costs the time to train those models, as one have to restart. If we support the requested feature, the failed operations will display an warning (may be ⚠️), and will go on to final training step. Then depending on whether that supports comparison of subset of all models, it will proceed as if the failed models were not there. If not, it'll fail there. Very similar functionality in available in few CI tools. For example, Gitlab CI has [`allow_failure`](https://docs.gitlab.com/ee/ci/yaml/#allow_failure), Travis CI has [`allow_failures`](https://docs.travis-ci.com/user/customizing-the-build/#jobs-that-are-allowed-to-fail), etc. ### Is there a workaround currently? <!-- Without this feature, how do you accomplish your task today? --> It is possible to do very broad top level exception handling to suppress failures. However, in this way the fact that it failed is hidden in the logs and not displayed in the pipeline dashboard. In scheduled pipelines where no one really go through the logs of all "successful" pipelines, these failures will go unnoticed. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7330/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7329

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7329/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7329/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7329/events

https://github.com/kubeflow/pipelines/issues/7329

I_kwDOB-71UM5D_uS2

[backend] Some executions on Kubeflow MLMD have no values

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? Full Kubeflow deployment * KFP version: 1.5 * KFP SDK version: 1.6.0 ### Steps to reproduce Some output values of Kubeflow MLMD `mlmd.get_executions_by_context(context.id)` ([reference](https://www.tensorflow.org/tfx/ml_metadata/api_docs/python/mlmd/MetadataStore#get_executions_by_context)) just include the field names and no values, e.g. ``` custom_properties { key: "input:output_artifacts" value { string_value: "[\"saved_model_uri\", \"model_name\", \"user_data\"]" } } ``` ### Expected result Output values of components should include the field names as well as the values ### Materials and Reference --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7329/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7328

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7328/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7328/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7328/events

https://github.com/kubeflow/pipelines/issues/7328

I_kwDOB-71UM5D_uLU

[backend] Only 1 component gets logged into MLMD for some pipeline runs

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? Full Kubeflow deployment * KFP version: 1.5 * KFP SDK version: 1.6.0 ### Steps to reproduce After running several pipelines multiple times, only 1 component gets logged into MLMD for some pipeline runs. This is problematic as we can't query for pipeline artifacts reliably from MLMD. ### Expected result For every run we expect all components log information into Kubeflow MLMD ### Materials and Reference <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7328/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7327

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7327/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7327/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7327/events

https://github.com/kubeflow/pipelines/issues/7327

I_kwDOB-71UM5D_twP

[feature] Remove custom hash suffix to end of KFP pipeline name in executions

open

NONE

### Feature Area /area backend ### What feature would you like to see? Remove custom hash suffix to end of KFP pipeline name in executions ### What is the use case or pain point? KFP adds a custom hash to the end of the pipeline name similar to how kubernetes adds a unique suffix to pod names. This makes it hard to query runs and metadata of pipelines after the execution. ### Is there a workaround currently? Users have to query for the correct execution or context and get the relevant id then get the desired pipeline which is counter intuitive and troublesome. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7327/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7325

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7325/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7325/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7325/events

https://github.com/kubeflow/pipelines/issues/7325

I_kwDOB-71UM5D_hDF

[sdk] kfp.dsl.RUN_ID_PLACEHOLDER confusion: is this different than Argo's {{workflow.uid}}?

closed

CONTRIBUTOR

### Environment * KFP version: 1.7.1 * KFP SDK version: 1.6.2 * All dependencies version: ``` kfp 1.6.2 kfp-pipeline-spec 0.1.13 kfp-server-api 1.7.1 ``` ### Steps to reproduce My team uses `kfp.dsl.RUN_ID_PLACEHOLDER` as a way to reference a particular KFP run. We use this as part of the GCS path for storing artifacts, and we save it in BigQuery in a `kfp_run_id` column. I wanted to change our system so we use a short version of this ID in some places. For example, if the full ID is `ee7b084b-5051-4a08-b007-b075b2d8bd08`, I wanted to use the first segment `ee7b084b` (kind of like using short git commit hashes). I saw that Argo supports some expressions ([docs for Argo workflow variables](https://github.com/argoproj/argo-workflows/blob/master/docs/variables.md)), so I tried using `{{=sprig.substr(0, 8, workflow.uid)}}`. This appeared to work, but the 8 characters didn't match the full KFP run ID. ### Expected result I expected `{{=sprig.substr(0, 8, workflow.uid}}` to be the first 8 characters of `{{workflow.uid}}`. ### Materials and Reference I was debugging by passing `plain:{{workflow.uid}} substr:{{=sprig.substr(0, 36, workflow.uid}}` as an argument to a test component, and I noticed that the workflow object on k8s replaces the "plain" version but not the "substr" version (based on `kubectl get wf`). Based on the discussions here, it seems that KFP's backend is searching for the exact string `{{workflow.uid}}` and replacing it with its own UUID. Is that correct? - https://github.com/kubeflow/pipelines/issues/3681 - https://github.com/kubeflow/pipelines/pull/4995 - https://github.com/kubeflow/pipelines/pull/3709 - https://github.com/kubeflow/pipelines/issues/5474 Is there any way to work around this? Thanks. --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7325/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7310

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7310/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7310/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7310/events

https://github.com/kubeflow/pipelines/issues/7310

I_kwDOB-71UM5D21_K

[sdk] PipelineParams are not serialized

closed

NONE

### Environment * KFP SDK version: 1.8.11 (v2) This is related to: #2206 I also commented there, but I am not sure if adding a new issue is the better way to go. I try to pass a dict containing pipeline parameters to a component (as shown by numerology in #2206 ), but the serialization of the pipeline params does not work. ```python import fire from utils.pipeline_cli import PipelineCLI from kfp.v2 import dsl @dsl.component(base_image="python:3.8-slim") def print_op(params: dict): print(params) @dsl.pipeline( name="test", description="test-pipeline.", ) def pipeline(test: str = "test"): print_op(params=dict(foo=test)) if __name__ == "__main__": cli = PipelineCLI(pipeline=pipeline) fire.Fire(cli) ``` where PipelineCLI implements the compile and run subcommands. even the following minimal example does not work. ```bash python3 -c 'import kfp, json;print(json.dumps(kfp.dsl.PipelineParam("aaa")))' ``` both yield: ```bash Traceback (most recent call last): File "<string>", line 1, in <module> File "/opt/conda/lib/python3.7/json/__init__.py", line 231, in dumps return _default_encoder.encode(obj) File "/opt/conda/lib/python3.7/json/encoder.py", line 199, in encode chunks = self.iterencode(o, _one_shot=True) File "/opt/conda/lib/python3.7/json/encoder.py", line 257, in iterencode return _iterencode(o, 0) File "/opt/conda/lib/python3.7/json/encoder.py", line 179, in default raise TypeError(f'Object of type {o.__class__.__name__} ' TypeError: Object of type PipelineParam is not JSON serializable ```

https://api.github.com/repos/kubeflow/pipelines/issues/7310/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7302

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7302/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7302/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7302/events

https://github.com/kubeflow/pipelines/issues/7302

I_kwDOB-71UM5DwVDE

[feature] Multi-user support for local standalone deployment

closed

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> /area backend <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? <!-- Provide a description of this feature and the user experience. --> I would like to be able to run pipelines in specific namespaces that I have set up as a cluster administrator without having to deploy the full Kubeflow deployment. ### What is the use case or pain point? <!-- It helps us understand the benefit of this feature for your use case. --> Right now, I have followed [these instructions](https://www.kubeflow.org/docs/components/pipelines/installation/localcluster-deployment/) and everything works, but when I try to create a run from a pipeline func and specify a namespace, I see ```text HTTP response body: {"error":"Validate experiment request failed.: Invalid input error: In single-user mode, CreateExperimentRequest shouldn't contain resource references.","code":3,"message":"Validate experiment request failed.: Invalid input error: In single-user mode, CreateExperimentRequest shouldn't contain resource references.","details":[{"@type":"type.googleapis.com/api.Error","error_message":"In single-user mode, CreateExperimentRequest shouldn't contain resource references.","error_details":"Validate experiment request failed.: Invalid input error: In single-user mode, CreateExperimentRequest shouldn't contain resource references."}]} ``` My use case is that I have an existing kubernetes cluster, and there are existing namespaces that everyone uses, so it would be convenient if I could just set the namespace and have that be accepted by the KFP backend. ### Is there a workaround currently? <!-- Without this feature, how do you accomplish your task today? --> Not that I am aware of. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7302/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7301

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7301/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7301/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7301/events

https://github.com/kubeflow/pipelines/issues/7301

I_kwDOB-71UM5DmgId

Reuse artifact in different pipeline/run

open

NONE

My pipeline consists of three components: `dataset_preparation`, `training` and `evaluation`. The first one reads all files in my data lake and generates a dataset artifact (stored to S3) that is used in the following components. We keep adding new files to the datalake on a daily basis, so by default I want to train on all the data available. Therefore, I don't want to cache that component. However, in some cases I might want to re-run that pipeline (with different hyperparameters for instance). To be able to compare runs, I'd like to run it on an existing dataset artifact. Thus I'd like to provide the artifact as an input parameter to the pipeline and skip the execution of `dataset_preparation`. **Summary**: I'd like to be able to provide a dataset artifact as an optional input to the pipeline. If not provided, it will be created by the `dataset_preparation` component. If provided, `dataset_preparation` will be skipped. Is there any way to accomplish this? Thanks a lot in advance!

https://api.github.com/repos/kubeflow/pipelines/issues/7301/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7339

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7339/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7339/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7339/events

https://github.com/kubeflow/pipelines/issues/7339

I_kwDOB-71UM5EPXwL

Component Visualizations are not shown in Run Output

open

NONE

I have uploaded the metrics_visualization pipeline in the Kubeflow Console and when I invoke a run I am able to see the visualizations yet they are not seen in the run output   Do we need to make any configuration changes to show the visualizations in the Run Output ? Used the below code to compile the pipeline : compiler.Compiler(mode=kfp.dsl.PipelineExecutionMode.V2_COMPATIBLE).compile( pipeline_func=pipeline, package_path=__file__.replace('.py', '.yaml')) As per the documentation : https://www.kubeflow.org/docs/components/pipelines/sdk/output-viewer/ The Run output tab shows the visualizations for all pipeline steps in the selected run. To open the tab in the Kubeflow Pipelines UI: Click Experiments to see your current pipeline experiments. Click the experiment name of the experiment that you want to view. Click the run name of the run that you want to view. Click the Run output tab.

https://api.github.com/repos/kubeflow/pipelines/issues/7339/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7296

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7296/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7296/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7296/events

https://github.com/kubeflow/pipelines/issues/7296

I_kwDOB-71UM5Dd-Zy

[feature].need support for file list

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? i want to merge multi files in one pipeline task,so i need support list[InputPath] such like this ### What is the use case or pain point? if i have 50 tasks and output 50 files i need write a function with 50 params , it is painfull ### Is there a workaround currently? i actually write 50 params function --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7296/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7294

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7294/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7294/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7294/events

https://github.com/kubeflow/pipelines/issues/7294

I_kwDOB-71UM5DXfUN

[frontend] args sort in ui doesn't follow py-func pipeline

closed

NONE

### Environment Google vertex ### Deployment If we define ```py @component def test(a: str, b: str, c: str, d:str) -> str: kwargs = locals() return str(kwargs) @pipeline( name=name ) def pipeline( a: str = 'a', b: str = 'b', c: str = 'c', d: str = 'd'): test_task = test(a, b, c,d) ``` Then the UI, doesn't display the args in order (a,b,c,d). In fact, the order seems pretty random (d,a,c,b). There's any reason behind this?

https://api.github.com/repos/kubeflow/pipelines/issues/7294/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7285

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7285/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7285/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7285/events

https://github.com/kubeflow/pipelines/issues/7285

I_kwDOB-71UM5DR5l8

[frontend] Make JSON structured logs more readable

open

CONTRIBUTOR

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> /area frontend ### What feature would you like to see? Something that makes JSON logs more readable. Perhaps JSON blocks that can be expanded and collapsed. ### What is the use case or pain point? I use Google Cloud Logging with [JSON structured logs](https://cloud.google.com/logging/docs/structured-logging). Large JSON logs in the Kubeflow log viewer UI aren't very human readable. Lines are long. Users have to [scroll horizontally](https://github.com/kubeflow/pipelines/issues/5366) to view them.  ### Is there a workaround currently? Not really * Scrolling horizontally and squinting * Make my brain parse JSON better and increase my short term memory * Copy-paste JSON string into JSON prettifier --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7285/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7283

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7283/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7283/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7283/events

https://github.com/kubeflow/pipelines/issues/7283

I_kwDOB-71UM5DO_sB

The exception is raised when non ascii code characters are passed as a argument to pipeline component.

open

NONE

### What steps did you take Run the following pipeline example code on notebook to log multi-byte string "あ". ```python import kfp from kfp import dsl from kfp.v2.dsl import component @component(base_image="python:3.7") def display_str_op(var: str): print(var) @dsl.pipeline( name="simple-pipeline" ) def simple_pipeline(): display_str_op("あ") client = kfp.Client("http://ml-pipeline.kubeflow:8888") run_result = client.create_run_from_pipeline_func( simple_pipeline, experiment_name = 'test', run_name = 'simple pipeline', arguments = {}, namespace = 'kubeflow-user-example-com', mode = kfp.dsl.PipelineExecutionMode.V2_COMPATIBLE ) ``` ### What happened: Following Exception is raised in the pipeline running. ``` ApiException: (500) Reason: Internal Server Error HTTP response headers: HTTPHeaderDict({'content-type': 'application/json', 'date': 'Wed, 05 Jan 2022 02:42:42 GMT', 'content-length': '775', 'x-envoy-upstream-service-time': '83', 'server': 'envoy'}) HTTP response body: {"error":"Failed to create a new run.: InternalServerError: Failed to store run simple-pipeline-vfzfz to table: Error 1366: Incorrect string value: '\\xEF\\xBC\\xA1\",\"...' for column 'WorkflowRuntimeManifest' at row 1","code":13,"message":"Failed to create a new run.: InternalServerError: Failed to store run simple-pipeline-vfzfz to table: Error 1366: Incorrect string value: '\\xEF\\xBC\\xA1\",\"...' for column 'WorkflowRuntimeManifest' at row 1","details":[{"@type":"type.googleapis.com/api.Error","error_message":"Internal Server Error","error_details":"Failed to create a new run.: InternalServerError: Failed to store run simple-pipeline-vfzfz to table: Error 1366: Incorrect string value: '\\xE3\\x81\\x82\",\"...' for column 'WorkflowRuntimeManifest' at row 1"}]} ``` ### What did you expect to happen: Result should have been completed successfully. ### Environment: * How do you deploy Kubeflow Pipelines (KFP)? - Install Kubeflow v1.4.0 with https://github.com/kubeflow/manifests#install-with-a-single-command * KFP version: - 1.7.0 * KFP SDK version: - 1.6.3 ### Anything else you would like to add: The version of kubernetes is v1.21.1 ### Labels <!-- Please include labels below by uncommenting them to help us better triage issues --> <!-- /area frontend --> <!-- /area backend --> <!-- /area sdk --> <!-- /area testing --> <!-- /area samples --> <!-- /area components --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7283/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7274

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7274/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7274/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7274/events

https://github.com/kubeflow/pipelines/issues/7274

I_kwDOB-71UM5DKyBN

[backend] Argo Workflows created for ScheduledWorkflows are missing common metadata

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? Kustomize * KFP version: 1.3.0 - but the same behaviour is present in current versions * KFP SDK version: 1.8.10 ### Steps to reproduce - Create a Pipeline run - Create a Recurring Run for the same pipeline - Compare `.metadata` of the resulting Argo Workflows Example created as a single Pipeline run: ```yaml apiVersion: argoproj.io/v1alpha1 kind: Workflow metadata: annotations: > pipelines.kubeflow.org/kfp_sdk_version: 1.8.10 > pipelines.kubeflow.org/pipeline_compilation_time: '2022-02-01T15:41:19.456019' > pipelines.kubeflow.org/pipeline_spec: '{"description": "Constructs a Kubeflow pipeline.", "inputs": [{"name": "pipeline-root"}], "name": "training-pipeline"}'} labels: > pipelines.kubeflow.org/kfp_sdk_version: 1.8.10 workflows.argoproj.io/completed: "true" workflows.argoproj.io/phase: Succeeded pipeline/persistedFinalState: "true" pipeline/runid: 96455954-6f96-4933-9ee1-a85cd676b8c6 ... ``` Example created by a recurring run: ```yaml apiVersion: argoproj.io/v1alpha1 kind: Workflow metadata: annotations: {} labels: pipeline/persistedFinalState: "true" pipeline/runid: bd44f525-7576-4ca8-ae82-b0eb050d1ae9 scheduledworkflows.kubeflow.org/isOwnedByScheduledWorkflow: "true" scheduledworkflows.kubeflow.org/scheduledWorkflowName: training-run-configurr8fjn scheduledworkflows.kubeflow.org/workflowEpoch: "1644166800" scheduledworkflows.kubeflow.org/workflowIndex: "2" workflows.argoproj.io/completed: "true" workflows.argoproj.io/phase: Succeeded ... ``` You can see that the workflow for the recurring run has an entirely new set of labels and no annotations. Specifically, the compiled argo workflow contains `pipelines.kubeflow.org/*` fields which get removed for the scheduled run. ### Expected result Both workflows should have the same common metadata - most importantly labels and annotations. We use these fields for automation after a pipeline has finshed. ### Materials and Reference After some investigation I found out that the `ScheduledWorkflow` CRD doesn't contain a metadata field: https://github.com/kubeflow/pipelines/blob/master/backend/src/apiserver/template/argo_template.go#L92 and https://github.com/kubeflow/pipelines/blob/master/backend/src/crd/pkg/apis/scheduledworkflow/v1beta1/types.go#L48 and that the original Workflow's Spec gets copied without its metadata: https://github.com/kubeflow/pipelines/blob/master/backend/src/crd/controller/scheduledworkflow/util/scheduled_workflow.go#L164 --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7274/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7271

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7271/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7271/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7271/events

https://github.com/kubeflow/pipelines/issues/7271

I_kwDOB-71UM5DI3tw

[backend] ml-pipeline deployment readiness probe failed

open

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? AWS EKS <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: 1.2 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: 1.2 <!-- Specify the output of the following shell command: $pip list | grep kfp --> ### Steps to reproduce <!-- Specify how to reproduce the problem. This may include information such as: a description of the process, code snippets, log output, or screenshots. --> ### Expected result <!-- What should the correct behavior be? --> ### Materials and Reference <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> <details> <summary> The liveness probe of the `ml-pipeline` deployment failed. </summary> ``` $ k describe -n kubeflow pod ml-pipeline-5f465d4c56-7xcs8 Name: ml-pipeline-5f465d4c56-7xcs8 Namespace: kubeflow Priority: 0 Node: ip-10-0-3-78.us-west-2.compute.internal/10.0.3.78 Start Time: Mon, 07 Feb 2022 11:05:22 -0800 Labels: app=ml-pipeline application-crd-id=kubeflow-pipelines pod-template-hash=5f465d4c56 Annotations: kubectl.kubernetes.io/restartedAt: 2022-02-06T17:31:44-08:00 kubernetes.io/psp: eks.privileged sidecar.istio.io/inject: false Status: Running IP: 10.0.3.52 IPs: IP: 10.0.3.52 Controlled By: ReplicaSet/ml-pipeline-5f465d4c56 Containers: ml-pipeline-api-server: Container ID: docker://6659ead43604634288ebe7987ba5f41e892e06c568645b2883547b3c26cdb167 Image: gcr.io/ml-pipeline/api-server:1.2.0 Image ID: docker-pullable://gcr.io/ml-pipeline/api-server@sha256:6553e9855e6d38eb5a70beeea39a2c37ac85b60f26a5c061b5e5e2adfffd960b Ports: 8888/TCP, 8887/TCP Host Ports: 0/TCP, 0/TCP State: Running Started: Mon, 07 Feb 2022 11:05:23 -0800 Ready: False Restart Count: 0 Liveness: exec [wget -q -S -O - http://localhost:8888/apis/v1beta1/healthz] delay=3s timeout=2s period=5s #success=1 #failure=3 Readiness: exec [wget -q -S -O - http://localhost:8888/apis/v1beta1/healthz] delay=3s timeout=2s period=5s #success=1 #failure=3 Environment: AUTO_UPDATE_PIPELINE_DEFAULT_VERSION: <set to the key 'autoUpdatePipelineDefaultVersion' of config map 'pipeline-install-config-d42hc87dh2'> Optional: false POD_NAMESPACE: kubeflow (v1:metadata.namespace) OBJECTSTORECONFIG_SECURE: false OBJECTSTORECONFIG_BUCKETNAME: <set to the key 'bucketName' of config map 'pipeline-install-config-d42hc87dh2'> Optional: false DBCONFIG_USER: <set to the key 'username' in secret 'mysql-secret-fd5gktm75t'> Optional: false DBCONFIG_PASSWORD: <set to the key 'password' in secret 'mysql-secret-fd5gktm75t'> Optional: false DBCONFIG_DBNAME: <set to the key 'pipelineDb' of config map 'pipeline-install-config-d42hc87dh2'> Optional: false DBCONFIG_HOST: <set to the key 'dbHost' of config map 'pipeline-install-config-d42hc87dh2'> Optional: false DBCONFIG_PORT: <set to the key 'dbPort' of config map 'pipeline-install-config-d42hc87dh2'> Optional: false OBJECTSTORECONFIG_ACCESSKEY: <set to the key 'accesskey' in secret 'mlpipeline-minio-artifact'> Optional: false OBJECTSTORECONFIG_SECRETACCESSKEY: <set to the key 'secretkey' in secret 'mlpipeline-minio-artifact'> Optional: false Mounts: /var/run/secrets/kubernetes.io/serviceaccount from ml-pipeline-token-zvqgd (ro) Conditions: Type Status Initialized True Ready False ContainersReady False PodScheduled True Volumes: ml-pipeline-token-zvqgd: Type: Secret (a volume populated by a Secret) SecretName: ml-pipeline-token-zvqgd Optional: false QoS Class: BestEffort Node-Selectors: <none> Tolerations: node.kubernetes.io/not-ready:NoExecute op=Exists for 300s node.kubernetes.io/unreachable:NoExecute op=Exists for 300s Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal Scheduled 12m default-scheduler Successfully assigned kubeflow/ml-pipeline-5f465d4c56-7xcs8 to ip-10-0-3-78.us-west-2.compute.internal Normal Pulled 12m kubelet Container image "gcr.io/ml-pipeline/api-server:1.2.0" already present on machine Normal Created 12m kubelet Created container ml-pipeline-api-server Normal Started 12m kubelet Started container ml-pipeline-api-server Warning Unhealthy 6s (x2 over 6m12s) kubelet Readiness probe errored: rpc error: code = DeadlineExceeded desc = context deadline exceeded Warning Unhealthy 4s (x2 over 6m10s) kubelet Liveness probe errored: rpc error: code = DeadlineExceeded desc = context deadline exceeded ``` </details> Logs of the pod ``` $ k logs -n kubeflow ml-pipeline-5f465d4c56-7xcs8 I0207 19:05:23.824447 9 client_manager.go:140] Initializing client manager I0207 19:05:23.824841 9 config.go:56] Config DBConfig.ExtraParams not specified, skipping ``` Executing the health check from the pod receives no response ``` k exec -n kubeflow ml-pipeline-5f465d4c56-7xcs8 -- wget -q -S -O - http://localhost:8888/apis/v1beta1/healthz ``` <details> <summary> Deployment yaml of the `ml-pipeline` </summary> ``` $ k get deploy -n kubeflow ml-pipeline -o yaml apiVersion: apps/v1 kind: Deployment metadata: annotations: deployment.kubernetes.io/revision: "5" kubectl.kubernetes.io/last-applied-configuration: | {"apiVersion":"apps/v1","kind":"Deployment","metadata":{"annotations":{},"labels":{"app":"ml-pipeline","application-crd-id":"kubeflow-pipelines"},"name":"ml-pipeline","namespace":"kubeflow"},"spec":{"selector":{"matchLabels":{"app":"ml-pipeline","application-crd-id":"kubeflow-pipelines"}},"template":{"metadata":{"labels":{"app":"ml-pipeline","application-crd-id":"kubeflow-pipelines"}},"spec":{"containers":[{"env":[{"name":"AUTO_UPDATE_PIPELINE_DEFAULT_VERSION","valueFrom":{"configMapKeyRef":{"key":"autoUpdatePipelineDefaultVersion","name":"pipeline-install-config-d42hc87dh2"}}},{"name":"POD_NAMESPACE","valueFrom":{"fieldRef":{"fieldPath":"metadata.namespace"}}},{"name":"OBJECTSTORECONFIG_SECURE","value":"false"},{"name":"OBJECTSTORECONFIG_BUCKETNAME","valueFrom":{"configMapKeyRef":{"key":"bucketName","name":"pipeline-install-config-d42hc87dh2"}}},{"name":"DBCONFIG_USER","valueFrom":{"secretKeyRef":{"key":"username","name":"mysql-secret-fd5gktm75t"}}},{"name":"DBCONFIG_PASSWORD","valueFrom":{"secretKeyRef":{"key":"password","name":"mysql-secret-fd5gktm75t"}}},{"name":"DBCONFIG_DBNAME","valueFrom":{"configMapKeyRef":{"key":"pipelineDb","name":"pipeline-install-config-d42hc87dh2"}}},{"name":"DBCONFIG_HOST","valueFrom":{"configMapKeyRef":{"key":"dbHost","name":"pipeline-install-config-d42hc87dh2"}}},{"name":"DBCONFIG_PORT","valueFrom":{"configMapKeyRef":{"key":"dbPort","name":"pipeline-install-config-d42hc87dh2"}}},{"name":"OBJECTSTORECONFIG_ACCESSKEY","valueFrom":{"secretKeyRef":{"key":"accesskey","name":"mlpipeline-minio-artifact"}}},{"name":"OBJECTSTORECONFIG_SECRETACCESSKEY","valueFrom":{"secretKeyRef":{"key":"secretkey","name":"mlpipeline-minio-artifact"}}}],"image":"gcr.io/ml-pipeline/api-server:1.2.0","imagePullPolicy":"IfNotPresent","livenessProbe":{"exec":{"command":["wget","-q","-S","-O","-","http://localhost:8888/apis/v1beta1/healthz"]},"initialDelaySeconds":3,"periodSeconds":5,"timeoutSeconds":2},"name":"ml-pipeline-api-server","ports":[{"containerPort":8888,"name":"http"},{"containerPort":8887,"name":"grpc"}],"readinessProbe":{"exec":{"command":["wget","-q","-S","-O","-","http://localhost:8888/apis/v1beta1/healthz"]},"initialDelaySeconds":3,"periodSeconds":5,"timeoutSeconds":2}}],"serviceAccountName":"ml-pipeline"}}}} creationTimestamp: "2021-01-15T22:01:56Z" generation: 15 labels: app: ml-pipeline application-crd-id: kubeflow-pipelines name: ml-pipeline namespace: kubeflow ownerReferences: - apiVersion: app.k8s.io/v1beta1 blockOwnerDeletion: true controller: false kind: Application name: pipeline uid: ea8a9b37-0c16-439e-bc49-3399051aca6e resourceVersion: "532602378" selfLink: /apis/apps/v1/namespaces/kubeflow/deployments/ml-pipeline uid: 908e252d-c7c6-49f2-88e0-dcf568097b14 spec: progressDeadlineSeconds: 600 replicas: 1 revisionHistoryLimit: 10 selector: matchLabels: app: ml-pipeline application-crd-id: kubeflow-pipelines strategy: rollingUpdate: maxSurge: 25% maxUnavailable: 25% type: RollingUpdate template: metadata: annotations: kubectl.kubernetes.io/restartedAt: "2022-02-06T17:31:44-08:00" sidecar.istio.io/inject: "false" creationTimestamp: null labels: app: ml-pipeline application-crd-id: kubeflow-pipelines spec: containers: - env: - name: AUTO_UPDATE_PIPELINE_DEFAULT_VERSION valueFrom: configMapKeyRef: key: autoUpdatePipelineDefaultVersion name: pipeline-install-config-d42hc87dh2 - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace - name: OBJECTSTORECONFIG_SECURE value: "false" - name: OBJECTSTORECONFIG_BUCKETNAME valueFrom: configMapKeyRef: key: bucketName name: pipeline-install-config-d42hc87dh2 - name: DBCONFIG_USER valueFrom: secretKeyRef: key: username name: mysql-secret-fd5gktm75t - name: DBCONFIG_PASSWORD valueFrom: secretKeyRef: key: password name: mysql-secret-fd5gktm75t - name: DBCONFIG_DBNAME valueFrom: configMapKeyRef: key: pipelineDb name: pipeline-install-config-d42hc87dh2 - name: DBCONFIG_HOST valueFrom: configMapKeyRef: key: dbHost name: pipeline-install-config-d42hc87dh2 - name: DBCONFIG_PORT valueFrom: configMapKeyRef: key: dbPort name: pipeline-install-config-d42hc87dh2 - name: OBJECTSTORECONFIG_ACCESSKEY valueFrom: secretKeyRef: key: accesskey name: mlpipeline-minio-artifact - name: OBJECTSTORECONFIG_SECRETACCESSKEY valueFrom: secretKeyRef: key: secretkey name: mlpipeline-minio-artifact image: gcr.io/ml-pipeline/api-server:1.2.0 imagePullPolicy: IfNotPresent livenessProbe: exec: command: - wget - -q - -S - -O - '-' - http://localhost:8888/apis/v1beta1/healthz failureThreshold: 3 initialDelaySeconds: 3 periodSeconds: 5 successThreshold: 1 timeoutSeconds: 2 name: ml-pipeline-api-server ports: - containerPort: 8888 name: http protocol: TCP - containerPort: 8887 name: grpc protocol: TCP readinessProbe: exec: command: - wget - -q - -S - -O - '-' - http://localhost:8888/apis/v1beta1/healthz failureThreshold: 3 initialDelaySeconds: 3 periodSeconds: 5 successThreshold: 1 timeoutSeconds: 2 resources: {} terminationMessagePath: /dev/termination-log terminationMessagePolicy: File dnsPolicy: ClusterFirst restartPolicy: Always schedulerName: default-scheduler securityContext: {} serviceAccount: ml-pipeline serviceAccountName: ml-pipeline terminationGracePeriodSeconds: 30 status: conditions: - lastTransitionTime: "2022-02-07T18:58:32Z" lastUpdateTime: "2022-02-07T18:58:32Z" message: Deployment does not have minimum availability. reason: MinimumReplicasUnavailable status: "False" type: Available - lastTransitionTime: "2022-02-07T19:16:46Z" lastUpdateTime: "2022-02-07T19:16:46Z" message: ReplicaSet "ml-pipeline-5f465d4c56" has timed out progressing. reason: ProgressDeadlineExceeded status: "False" type: Progressing observedGeneration: 15 replicas: 2 unavailableReplicas: 2 updatedReplicas: 1 ``` </details> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7271/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7270

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7270/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7270/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7270/events

https://github.com/kubeflow/pipelines/issues/7270

I_kwDOB-71UM5DISBo

[bug] No _back_quoted_if_needed(model_name) in bigquery_predict_model_job

open

NONE

### What steps did you take Run google_cloud_pipeline_components.experimental.bigquery.BigqueryPredictModelJobOp job on pipeline. Pass in parameter model_name = "\`project.dataset.model\`" WITH BACKTICKS as I did on google_cloud_pipeline_components.experimental.bigquery.BigqueryEvaluateModelJobOp. Pipeline run fails due to error Syntax error: Invalid empty identifier at [1:32]. I can see that this is due to https://github.com/kubeflow/pipelines/blob/master/components/google-cloud/google_cloud_pipeline_components/container/experimental/gcp_launcher/bigquery_job_remote_runner.py#L393 where it is written as 'SELECT * FROM ML.PREDICT(MODEL \`%s\`, %s%s)'. Note that there is '\`%s\`', which means that if I pass in model_name with backticks already included there will be two backticks now Where as in https://github.com/kubeflow/pipelines/blob/master/components/google-cloud/google_cloud_pipeline_components/container/experimental/gcp_launcher/bigquery_job_remote_runner.py#L599 It has 'SELECT * FROM ML.EVALUATE(MODEL %s%s%s)' % ( _back_quoted_if_needed(model_name), input_data_sql, threshold_sql) ### What happened: Pipeline job failed as I included backticks in model_name parameter ### What did you expect to happen: That the pipeline would run successfully ### Environment: kfp 1.8.11 google_cloud_pipeline_components 0.2.2 There just needs to be consistency between components, either backticks should not be included in both, or the _back_quoted_if_needed(model_name) should be done in both. <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7270/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7266

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7266/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7266/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7266/events

https://github.com/kubeflow/pipelines/issues/7266

I_kwDOB-71UM5DDCcp

[backend] v1 caching broken in KFP 1.8 release branch

closed

COLLABORATOR

The root cause is https://github.com/argoproj/argo-workflows/pull/6022 introduced a behavior change that the content of Argo workflow template is no longer logged under `pod.metadata.annotations`, instead it puts the template in container env. The change was released in Argo workflow [v3.2.0-rc1](https://github.com/argoproj/argo-workflows/blob/master/CHANGELOG.md#v320-rc1-2021-08-19). And https://github.com/kubeflow/pipelines/pull/6920 upgraded Argo workflow used in KFP from `3.1.14` to `3.2.3`. The change broke v1 caching implementation because it relies on Argo workflow template present in `pod.metadata.annotations`.

https://api.github.com/repos/kubeflow/pipelines/issues/7266/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7263

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7263/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7263/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7263/events

https://github.com/kubeflow/pipelines/issues/7263

I_kwDOB-71UM5DCXkm

[release] Backend 1.8.0 tracker

closed

COLLABORATOR

## Cherry pick PRs Make sure to cherry pick the following PRs to `release-1.8` branch before making the official release: - [x] https://github.com/kubeflow/pipelines/pull/7252 - [x] https://github.com/kubeflow/pipelines/pull/7267 - [x] https://github.com/kubeflow/pipelines/pull/7227 - [x] https://github.com/kubeflow/pipelines/pull/7273 - [x] https://github.com/kubeflow/pipelines/pull/7155 - [x] https://github.com/kubeflow/pipelines/pull/7311 Validate the new release candidate before official release. ## Instruction https://github.com/kubeflow/pipelines/blob/master/RELEASE.md#release-manager-instructions (Note, we need to skip some steps)

https://api.github.com/repos/kubeflow/pipelines/issues/7263/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7261

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7261/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7261/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7261/events

https://github.com/kubeflow/pipelines/issues/7261

I_kwDOB-71UM5DCAxE

Possible to set max concurrent runs on an experiment?

open

NONE

Hi folks, we have a use case where we have multiple training model pipelines and one evaluation pipeline. We'd like to block on the evaluation pipeline so that there is no race condition when comparing model performances. e.g. consider the case where model A has much better performance than model B, and model B is slightly better than production model. If we run evaluation pipeline for both model A <-> production and then model B <-> production at the same time, it is possible that model B will end up in production in the end. It doesn't seem possible to set max concurrent runs = 1 if it's not a recurring run. Is there a way to set this for experiments and non-recurring runs, or what's the recommendation here? Having the user involve some sort of distributed lock seems way too complicated.

https://api.github.com/repos/kubeflow/pipelines/issues/7261/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7257

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7257/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7257/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7257/events

https://github.com/kubeflow/pipelines/issues/7257

I_kwDOB-71UM5DAfMW

[bug] Artifacts and Lineage explorer don't work with KFP SDK V2

closed

NONE

### What steps did you take 1. Install Kubeflow 1.4 2. Create a pipeline with multiple artifacts type to illustrate the issue better (use Input[T] and Output[T]) 3. Run pipeline 4. Navigate to the Artifacts page ### What happened: - No pipeline workspace and pipeline name is visible. - All pipeline runs are grouped under this unnamed group - Lineage explorer is not accessible/visible  ### What did you expect to happen: I expect to get the same result as when using KFP SDK V1. Meaning we see all pipelines as groups and all artifacts under their respective pipeline group and we can access the lineage explorer. ### Environment: * How do you deploy Kubeflow Pipelines (KFP)? As part of Kubeflow manifest * KFP version: The one from Kubeflow 1.4 (backend is 1.7.1 I think) * KFP SDK version: 1.8.6 ### Labels /area frontend /area backend /area sdk <!-- /area testing --> <!-- /area samples --> <!-- /area components --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7257/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7250

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7250/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7250/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7250/events

https://github.com/kubeflow/pipelines/issues/7250

I_kwDOB-71UM5C6mkV

[feature] Use Artifact metadata in a pipeline condition

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> /area sdk <!-- /area samples --> /area components ### What feature would you like to see? <!-- Provide a description of this feature and the user experience. --> Use the value of a metadata field from an Artifact in a `dsl.Condition`. For example: ```python with dsl.Condition( component_op.metadata["n_rows"] < 20, name="check-missing-vectors", ): ``` Or, allow for a component to return both a basic output and an Artifact output. Which would facilitate: ```python with dsl.Condition( component_op.output < 20, name="check-missing-vectors", ): ``` ### What is the use case or pain point? I have been using the `dsl.Condition` to skip components based on the number of entries in a dataset. To do this, I used the Dataset's metadata attribute (`dataset.metadata["n_rows"]`). I then had to use another component to get the value and pass it to `dsl.Condition` and compare integers. This extra component seems unnecessary, and acting on metadata could make the condition logic really powerful. <!-- It helps us understand the benefit of this feature for your use case. --> ### Is there a workaround currently? As mentioned, I currently use a separate lightweight component that retrieves the metadata value. ```python @component def check_df_length_op(null_vectors: Input[Dataset]) -> int: return null_vectors.metadata["n_rows"] ``` <!-- Without this feature, how do you accomplish your task today? --> --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7250/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7242

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7242/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7242/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7242/events

https://github.com/kubeflow/pipelines/issues/7242

I_kwDOB-71UM5C0Vu5

[feature] Deduplicate component templates

open

CONTRIBUTOR

### Feature Area /area sdk ### What feature would you like to see? I would like the KFP compiler to reuse templates when calling the same component multiple times. In other words, each component only appears as a template once in the `spec.templates` pipeline yaml spec, and called with different input parameters each time they're used. Currently, each time a component is used in a pipeline, the compiler emits a new copy of the component definition in the pipeline YAML spec. I tried decorating my component functions with `@kfp.dsl.graph_component`, and refactored things so that each component function only takes `PipelineParam` inputs, but it seems like it didn't work: I still have multiple copies of each component. Perhaps I'm just using `graph_component` incorrectly? ### What is the use case or pain point? An important use case for my team is to run a single pipeline that trains/scores/QCs multiple models and then runs a reporting step comparing the results from each model. The size of the pipeline YAML spec scales linearly with the number of models we include in the pipeline. This is preventing us from comparing all the models we would like to. (Related: #4170) ### Is there a workaround currently? Not that I know of. This is blocking us from running pipelines as big as we'd like. We are already using the workaround suggested here to shrink the yaml output: https://github.com/kubeflow/pipelines/issues/4170#issuecomment-655764762 (I suppose we could try to deduplicate the generated pipeline yaml... but that seems quite complex.) --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7242/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7241

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7241/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7241/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7241/events

https://github.com/kubeflow/pipelines/issues/7241

I_kwDOB-71UM5CyJdb

[feature] Allow specifying inputs that are ignored by cache

closed

CONTRIBUTOR

It would be nice if we could control which inputs are taken into consideration when deciding whether to use cached results or not, e.g. allow setting a boolean flag in an input's definition to tell Kubeflow to ignore it when checking the cache such that if its value changes between runs, the cached output is still used. The use case is that components may have inputs that do not semantically change the resulting outputs, and so it would be better to take the output from cache rather than running the component again. To give an example, if your component runs a Beam job, you may not want to rerun the component if only the Beam pipeline options change (this is actually how TFX's caching works). As another example, if your component runs an EMR job and tags it with the workflow ID for traceability, it will rerun every time due to the workflow ID being unique on every run, despite it not changing the output at all. Even more granular control over caching would also be nice, such as still using the cache when the image changes.

https://api.github.com/repos/kubeflow/pipelines/issues/7241/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7238

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7238/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7238/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7238/events

https://github.com/kubeflow/pipelines/issues/7238

I_kwDOB-71UM5CwxM0

[SDK] KFP SDK V2 Namespace Plan

closed

CONTRIBUTOR

## Background Previously on **KFP SDK v1.x**, we had two modes of import, signaling two different compilers with different compilation results. * <code>import kfp</code> imports KFP v1 and v2-compatible mode compiler * <code>import kfp.v2</code> imports KFP v2 mode compiler Two modes of import are confusing to the users. Although users are in SDK v1.x, they have <code>kfp</code> and <code>kfp.v2</code> to choose from, where normally the package name (kfp) indicates the default versioning. As KFP SDK evolves to v2.0 version, we want to set a standard way of importing KFP namespace, for easier usage and also cleaner code for maintenance. ## Proposed Namespace for KFP SDK v2 Since the package name indicates the default versioning, we propose in **KFP v2** and possible future iterations, we will always use `kfp` as the official namespace. * <code>import kfp</code> imports KFP v2 mode compiler For v1-related features, users need to go back to SDK v1.x and use <code>import kfp</code>. In this way, users don't need to change imports when they are experiencing major version updates. For existing GA customers, we will provide an alias of `kfp.v2` to `kfp`, and throw a warning if user uses kfp.v2. Full set of warnings will be documented in the migration guide. ### Note: 1. v2-compatible mode was launched as an experimental feature, so it won't appear in v2 SDK and follows cloud deprecation policy. 2. For KFP SDK v1.x features, we will maintain it for 1 year after GA of SDK v2.0. After that, we will be providing security patches and major bug fixes for any bug that results in user code breaking until usage on GCP (Vertex and KFP) is 0. ## User Impact on Namespace 1. KFP V1 users migration (NO changes) <table> <tr> <td> Current </td> </tr> <tr> <td style="background-color: #f0f0f0"> ``` from kfp import dsl from kfp import compiler @dsl.pipeline(name='my-pipeline') def pipeline(): ... compiler.Compiler().compile(pipeline, 'path') ``` </td> </tr> </table> 2. Vertex Pipeline users (primarily removing v2 from the imports). Note: there are other changes suggested for Vertex Pipeline users which will be documented in the migration doc. <table> <tr> <td>Current </td> <td>Anticipated change </td> </tr> <tr> <td style="background-color: #f0f0f0"> ``` from kfp.v2 import dsl from kfp.v2 import compiler @dsl.pipeline(name='my-pipeline') def pipeline(): ... compiler.Compiler().compile(pipeline, 'path') ``` </td> <td style="background-color: #f0f0f0"> ``` from kfp import dsl from kfp import compiler @dsl.pipeline(name='my-pipeline') def pipeline(): ... compiler.Compiler().compile(pipeline, 'path') ``` </td> </tr> </table> 3. Component authors: libraries with dependencies on KFP (e.g. GCPC) <table> <tr> <td>Current </td> <td>Anticipated change </td> </tr> <tr> <td style="background-color: #f0f0f0"> ``` from kfp.v2 import dsl @dsl.component( packages_to_install=[ 'google-cloud-aiplatform', 'google-cloud-pipeline-components', 'protobuf' ], base_image='python:3.7') def GetTrialsOp(gcp_resources: str) -> list: … ``` </td> <td style="background-color: #f0f0f0"> ``` from kfp import dsl @dsl.component( packages_to_install=[ 'google-cloud-aiplatform', 'google-cloud-pipeline-components', 'protobuf' ], base_image='python:3.7') def GetTrialsOp(gcp_resources: str) -> list: … ``` </td> </tr> </table>

https://api.github.com/repos/kubeflow/pipelines/issues/7238/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7234

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7234/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7234/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7234/events

https://github.com/kubeflow/pipelines/issues/7234

I_kwDOB-71UM5CuDmb

[components] custom training job from component bug

closed

NONE

### Environment * KFP version: kfp==1.8.10 * google_cloud_pipeline_components SDK version: google-cloud-pipeline-components==0.2.2 ### Steps to reproduce 1) create a python function based component, e.g.: ``` @component def process( output_dataset_one: Output[Dataset], ): output_dataset_one.metadata["hello"] = "there" ``` 2) import custom job utils in pipeline code: ``` from google_cloud_pipeline_components.experimental.custom_job import utils as custom_job_utils ``` Transform component into training job: ``` process_op = custom_job_utils.create_custom_training_job_op_from_component( process, display_name = 'test-component', machine_type = 'n1-standard-4', boot_disk_type = 'pd-ssd', boot_disk_size_gb = "150" ) ``` 3) Run pipeline: ``` @dsl.pipeline( name='test', description="Pipeline to test custom training job component" ) def test_pipeline(): _ = ( process_op( project=<PROJECT_ID>, location=<REGION>, ) .set_display_name('test') ) ``` ### Expected result The artifact`output_dataset_one` will have a property/metadata `hello` if component is run normally, but when the component is transformed into a custom training job, the resulting artifact will not contain any properties/metadata. This seems like a bug. ### Materials and Reference https://github.com/kubeflow/pipelines/blob/master/components/google-cloud/google_cloud_pipeline_components/experimental/custom_job/utils.py --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7234/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7230

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7230/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7230/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7230/events

https://github.com/kubeflow/pipelines/issues/7230

I_kwDOB-71UM5Cnl3L

No graph to show in ML Pipeline UI kubeflow 1.3 on Openshift 4.8

open

NONE

### What steps did you take <!-- A clear and concise description of what the bug is.--> No graph to show in ML Pipeline UI kubeflow 1.3 on Openshift 4.8 We have used https://opendatahub.io/docs/kubeflow/installation.html for deploying Kubelfow 1.3 ### What happened: We are unable to get Graph view of the pipeline we see something like "No graph to show" I have edited destination rules in Kubeflow ml-pipeline and ml-pipeline-ui and ml-pipeline-visualizationserver from ISTIO_MUTUAL to DISABLE ### What did you expect to happen: We must see Graph for pipeline s we create ### Environment: <!-- Please fill in those that seem relevant. --> * How do you deploy Kubeflow Pipelines (KFP)? https://opendatahub.io/docs/kubeflow/installation.html <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: Kubeflow Pipelines Tekton | v0.8.0 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: <!-- Specify the output of the following shell command: $pip list | grep kfp --> ### Anything else you would like to add: <!-- Miscellaneous information that will assist in solving the issue.--> I have restarted all the PODS also the controller-manager-pod restarts and logs are too excessive to take a look at. Please advise on what could be done? --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7230/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7225

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7225/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7225/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7225/events

https://github.com/kubeflow/pipelines/issues/7225

I_kwDOB-71UM5Cgsrf

[feature] Should be able to set _DEFAULT_LAUNCHER_IMAGE using environment variable in notebook pod

open

NONE

### Feature Area <!-- /area sdk --> ### What feature would you like to see? Currently the default launcher image is set using a constant: `sdk/python/kfp/compiler/v2_compat.py:_DEFAULT_LAUNCHER_IMAGE = "gcr.io/ml-pipeline/kfp-launcher:1.8.7"`, however, if we want to use a private registry when launching a pipeline in Kubeflow notebook, we may need to use an environment variable to set this, so that we can just configure that in the Kubeflow notebook YAML file. ### Is there a workaround currently? I can set it directly in the function call here: https://kubeflow-pipelines.readthedocs.io/en/latest/source/kfp.client.html#kfp.Client.create_run_from_pipeline_func, but I'd like to configure a default for all users. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7225/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7224

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7224/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7224/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7224/events

https://github.com/kubeflow/pipelines/issues/7224

I_kwDOB-71UM5CgZaV

[bug] Unable to obtain the output of a graph component for downstream components

open

NONE

### What steps did you take In a folder, created the following: c1.yaml ```yaml name: Dummy Component A description: Manually Created Component Spec inputs: - name: Arg Input type: String outputs: - name: Arg Output type: String implementation: container: image: dummy-image:1.0 command: [ "dummy", "command", "A" ] args: [ "--arg-input", {inputValue: Arg Input}, "--arg-output", {outputPath: Arg Output} ] ``` c2.yaml ```yaml name: Dummy Component B description: Manually Created Component Spec inputs: - name: Arg Input type: String outputs: - name: Arg Output type: String implementation: container: image: dummy-image:1.0 command: [ "dummy", "command", "B" ] args: [ "--arg-input", {inputValue: Arg Input}, "--arg-output", {outputPath: Arg Output} ] ``` build_dummy_pipeline.py ```python import kfp.components import kfp import kfp.compiler c1_func = kfp.components.load_component("c1.yaml") c2_func = kfp.components.load_component("c2.yaml") def generate_graph_component_func(): @kfp.dsl.graph_component def c1_graph_component(input_arg): c1_op_a = c1_func(arg_input=input_arg) c1_op_b = c1_func(arg_input=c1_op_a.output) return {"Graph Output": c1_op_b.output} return c1_graph_component c1_graph_comp_func = generate_graph_component_func() @kfp.dsl.pipeline( name="Dummy Pipeline", description="Dummy Pipeline" ) def dummy_pipeline(input_arg: str): graph_comp_op = c1_graph_comp_func(input_arg) c2_op_a = c2_func(arg_input=graph_comp_op.outputs["Graph Output"]) return c2_op_a.output kfp.compiler.Compiler().compile(dummy_pipeline, "dummy_pipeline.yaml") ``` ```Command run: python3 build_dummy_pipeline.py ``` ### What happened: ```bash Traceback (most recent call last): File "build_dummy_pipeline.py", line 28, in <module> kfp.compiler.Compiler().compile(dummy_pipeline, "dummy_pipeline.yaml") File "/Users/achandrashekaran/PycharmProjects/kf-model-trainer/.venv/lib/python3.7/site-packages/kfp/compiler/compiler.py", line 1064, in compile package_path=package_path) File "/Users/achandrashekaran/PycharmProjects/kf-model-trainer/.venv/lib/python3.7/site-packages/kfp/compiler/compiler.py", line 1119, in _create_and_write_workflow pipeline_conf) File "/Users/achandrashekaran/PycharmProjects/kf-model-trainer/.venv/lib/python3.7/site-packages/kfp/compiler/compiler.py", line 900, in _create_workflow pipeline_func(*args_list, **kwargs_dict) File "build_dummy_pipeline.py", line 25, in dummy_pipeline c2_op_a = c2_func(arg_input=graph_comp_op.outputs["Graph Output"]) KeyError: 'Graph Output' ``` ### What did you expect to happen: On running the above python script, I expected the output of the graph component to be passed to c2_op_a and the pipeline to compile successfully. ### Environment: <!-- Please fill in those that seem relevant. --> * How do you deploy Kubeflow Pipelines (KFP)? <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: 1.7.1 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: -NA- <!-- Specify the output of the following shell command: $pip list | grep kfp --> ### Anything else you would like to add: <!-- Miscellaneous information that will assist in solving the issue.--> ### Labels <!-- Please include labels below by uncommenting them to help us better triage issues --> <!-- /area frontend --> <!-- /area backend --> /area sdk <!-- /area testing --> <!-- /area samples --> /area components --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7224/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7220

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7220/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7220/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7220/events

https://github.com/kubeflow/pipelines/issues/7220

I_kwDOB-71UM5Cdnmp

[SDK] AttributeError: 'ComponentStore' object has no attribute 'uri_search_template'

open

NONE

Hi, I am trying to load some prebuit gcp kubeflow components using `kfp.components.ComponentStore`. However I am getting this error: ``` line 180, in _load_component_spec_in_component_ref if self.uri_search_template: AttributeError: 'ComponentStore' object has no attribute 'uri_search_template' ``` when at this line of code: `mlengine_train_op = component_store.load_component('ml_engine/train')` ### Environment * How did you deploy Kubeflow Pipelines (KFP)? I am running this locally. * KFP version: ``` kfp 1.8.10 kfp-pipeline-spec 0.1.13 kfp-server-api 1.7.1 ``` ### Steps to reproduce ``` import kfp from kfp.components import func_to_container_op COMPONENT_URL_SEARCH_PREFIX = "https://raw.githubusercontent.com/kubeflow/pipelines/1.7.1/components/gcp/" component_store = kfp.components.ComponentStore( local_search_paths=None, url_search_prefixes=[COMPONENT_URL_SEARCH_PREFIX]) mlengine_train_op = component_store.load_component('ml_engine/train') mlengine_deploy_op = component_store.load_component('ml_engine/deploy') ``` --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7220/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7215

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7215/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7215/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7215/events

https://github.com/kubeflow/pipelines/issues/7215

I_kwDOB-71UM5CbfUy

[sdk] [question] How can I write test code for the code that uses the sdk?

open

NONE

/kind question ### Environment * KFP version: <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP SDK version: ``` kfp==1.8.9 ``` <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * All dependencies version: ``` kfp 1.8.9 kfp-pipeline-spec 0.1.13 kfp-server-api 1.7.1 ``` <!-- Specify the output of the following shell command: $pip list | grep kfp --> I implemented a python code to get the list of updated pipeline versions after a certain time offset using kfp.Client . This code implemented in a way that uses the methods of kfp.Client. And now I'm trying to test this code, but this code holds bunch of kfp_server_api & kfp.Client methods. It's very inconvenient to mock these SDK code to write test code. Is there anyone who has tested the code using kfp.Client? Please share some good ideas thanks!

https://api.github.com/repos/kubeflow/pipelines/issues/7215/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7214

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7214/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7214/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7214/events

https://github.com/kubeflow/pipelines/issues/7214

I_kwDOB-71UM5CbLYz

[bug] Bucket param missing when getting pipeline artifacts causing 403 in UI

open

NONE

### What steps did you take 1. Navigated to view a pipeline in the UI 2. Clicked a component to view details 3. Clicked the "Visualizations" tab 4. Saw 403 responses in the network tab in browser:  5. Port forwarded the `ml-pipeline-ui-artifact` service to `localhost:3000`: ``` kubectl port-forward svc/ml-pipeline-ui-artifact 3000:80 ``` 6. Copied request url which 403'd from screenshot: ``` https://test.kubeflow-platform.spotify.net/pipeline/artifacts/get?source=minio&namespace=hyperkube&bucket=&key=artifacts%2Ftaxi-example-pipeline-52s9w%2Ftaxi-example-pipeline-52s9w-1020558034%2Fmlpipeline-ui-metadata.tgz ``` 7. Curl url with domain replaced by `localhost:3000` to target port-forwarded service, which resulted in error message: ``` curl localhost:3000/pipeline/artifacts/get\?source\=minio\&namespace\=hyperkube\&bucket\=\&key\=artifacts%2Ftaxi-example-pipeline-52s9w%2Ftaxi-example-pipeline-52s9w-1020558034%2Fmlpipeline-ui-metadata.tgz Storage bucket is missing from artifact request% ``` 8. Re-did Step 7 again, but filled in bucket with expected bucket name `kf-test-artifact-store`, which resulted in a successful response: ``` curl localhost:3000/pipeline/artifacts/get\?source\=minio\&namespace\=hyperkube\&bucket\=kf-test-artifact-store\&key\=artifacts%2Ftaxi-example-pipeline-52s9w%2Ftaxi-example-pipeline-52s9w-1020558034%2Fmlpipeline-ui-metadata.tgz -i HTTP/1.1 200 OK ``` ### What happened: It doesn't seem that the `bucket=` query param is not being populated and I'm not sure why. We use Minio GCS Gateway for artifact storage and a Multi User install of KFP. Visualizations don't load as they seem to make the request described above on mount to the React DOM. In addition, component output strings are not loaded (unsure if that's related). Visualizations not loading:  Component output strings not loading:  ### What did you expect to happen: Visualizations to load:  ### Environment: * How do you deploy Kubeflow Pipelines (KFP)? <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> Via argo app of apps based off of GCP Blueprints * KFP version: 1.3.0 <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> * KFP SDK version: 1.4.0 <!-- Specify the output of the following shell command: $pip list | grep kfp --> ### Anything else you would like to add: <!-- Miscellaneous information that will assist in solving the issue.--> ### Labels <!-- Please include labels below by uncommenting them to help us better triage issues --> <!-- /area frontend --> /area backend <!-- /area sdk --> <!-- /area testing --> <!-- /area samples --> <!-- /area components --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7214/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7213

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7213/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7213/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7213/events

https://github.com/kubeflow/pipelines/issues/7213

I_kwDOB-71UM5CaD6x

[backend] Executor v2 : incorrect use of 0644 file mask when creating local dirs while retrieving artifacts from object storage ?

closed

NONE

### Environment * How did you deploy Kubeflow Pipelines (KFP)? deployed Kubeflow on a GKE cluster <!-- For more information, see an overview of KFP installation options: https://www.kubeflow.org/docs/pipelines/installation/overview/. --> * KFP version: <!-- Specify the version of Kubeflow Pipelines that you are using. The version number appears in the left side navigation of user interface. To find the version number, See version number shows on bottom of KFP UI left sidenav. --> kubeflow v1.4 branch * KFP SDK version: <!-- Specify the output of the following shell command: $pip list | grep kfp --> n/a ### Steps to reproduce We discovered that when local copying Minio artifacts, the Kubeflow Executor V2 creates local dirs with a `0644` mask, preventing the non-root users to access artifact dirs - and that's the case when using Kaniko to build some Docker image generated by the BentoML Framework (see https://github.com/bentoml/BentoML/issues/2199) **Is is a required behaviour or is it possible to use the `0744` mask for dirs ?** https://github.com/kubeflow/pipelines/blob/627b37c3edf23ee5efc1c31edcdcdd4ff5944f64/v2/objectstore/object_store.go#L266 more details: When launching a pipeline step in k8s that uses [Kaniko](https://github.com/GoogleContainerTools/kaniko) the `/kaniko/executor` command is wrapped by `/kfp-launcher/launch` command (i.e. the executor V2). This command processes the args to resolve pipeline artifacts (namely thoses begining with `minio://` ) i.e. it copy from the minio bucket to the local FS of the container (under `/minio/` directory. Note it is *not* a mount point/k8s volume) . It turns out that the dir is in `drw-r--r--` mode (`0644`), so preventing to perform `chown` operations. modifying it to `drwxr--r--` before launching kaniko fix the bug w/o tweaking the Dockerfile. <!-- Specify how to reproduce the problem. This may include information such as: a description of the process, code snippets, log output, or screenshots. --> ### Expected result use `0744` file mask when creating local dirs while downloading artifacts from `minio` storage. <!-- What should the correct behavior be? --> ### Materials and Reference <!-- Help us debug this issue by providing resources such as: sample code, background context, or links to references. --> --- <!-- Don't delete message below to encourage users to support your issue! --> Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7213/timeline

completed

https://api.github.com/repos/kubeflow/pipelines/issues/7204

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7204/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7204/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7204/events

https://github.com/kubeflow/pipelines/issues/7204

I_kwDOB-71UM5CXBiy

[sdk] KFP Compiler doesn't propagate ParallelFor inputs

open

CONTRIBUTOR

### Environment * KFP version: 1.8.2 * KFP SDK version: 1.8.2 * All dependencies version: ``` kfp 1.8.2 kfp-pipeline-spec 0.1.13 kfp-server-api 1.3.0 ``` ### Steps to reproduce Try to compile this code: ``` import kfp from kfp.components import func_to_container_op @func_to_container_op def produce_items() -> str: return '["first item", "second item"]' @func_to_container_op def do_nothing(): pass @kfp.dsl.pipeline() def parallelfor_nested_pipeline_param_resolving(fname1: str, fname2: str): items_op = produce_items() items = items_op.output with kfp.dsl.ParallelFor(items) as outer_loop_item: with kfp.dsl.ParallelFor(items) as inner_loop_item: do_nothing() if __name__ == '__main__': import kfp.compiler as compiler compiler.Compiler().compile(parallelfor_nested_pipeline_param_resolving, __file__ + '.yaml') ``` It fails with a `KeyError: 'produce-items'` exception from within `fix_big_data_passing` function (not the actual bug location). ### Expected result It should compile. ### Materials and Reference The minimal code to reproduce is provided above. Based on my preliminary research, it seems that `compiler/compiler.py` simply doesn't have logic to recursively propagate inputs/outputs of `OpsGroup`s as it does for regular `Op`s except for several very specific cases, and `ParallelFor` is not among them. The exception it is failing the compilation with is not the actual bug location. It just surfaces the fact that certain parameter was not propagated through nested `OpsGroups`. The code above can also be used as a unit test once the issue is fixed. --- Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7204/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7201

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7201/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7201/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7201/events

https://github.com/kubeflow/pipelines/issues/7201

I_kwDOB-71UM5CWdSw

[feature] Allow user to specify which input parameters to use in cache key

open

NONE

### Feature Area <!-- Uncomment the labels below which are relevant to this feature: --> <!-- /area frontend --> <!-- /area backend --> /area sdk <!-- /area samples --> <!-- /area components --> ### What feature would you like to see? Right now, the only caching customization that exists on an operator level in v1 KFP (please correct me if I am wrong), is: ``` task = some_op() task.execution_options.caching_strategy.max_cache_staleness ``` Something that would be greatly helpful for my use case is if there was a way to specify which input parameters should be used to generate the cache key, such as follows: ``` # assuming my input params are as follows: data_input_dir: str, model_output_dir: str, s3_path: str, task = some_op() task.execution_options.caching_strategy.include_input_parameters([data_input_dir, model_output_dir]) ``` The expected behavior here would be that if there was a previous run of the given task with identical `data_input_dir` and `model_output_dir` inputs, but a different `s3_path`, the previously-cached task would be valid and be used. ### What is the use case or pain point? We are trying to build full model training pipelines using KFP. The training jobs take a long time, and we would like them to use the cache whenever possible. However, each run has a _unique_ `s3_path` which we sync training checkpoints to during the job, which is the only way the jobs can finish successfully. Thus, we must provide `s3_path` as an input parameter, but it should not be used to determine whether the model has been trained with the same inputs before (ie. whether to use the cache). Rather, in our case, this should only depend on the data used (artifacts from the upstream task), and the config parameters (other input parameters to the task). Therefore, we would like to be able to have the caching mechanism ignore this unique `s3_path` parameter. ### Is there a workaround currently? I have tried experimenting with injecting the S3 path as an env var, but it is still considered an input parameter when I do this. Unfortunately, there does not seem to be a real workaround at this time; we are just unable to take advantage of the cache on these jobs. --- <!-- Don't delete message below to encourage users to support your feature request! --> Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

https://api.github.com/repos/kubeflow/pipelines/issues/7201/timeline

https://api.github.com/repos/kubeflow/pipelines/issues/7196

https://api.github.com/repos/kubeflow/pipelines

https://api.github.com/repos/kubeflow/pipelines/issues/7196/labels{/name}

https://api.github.com/repos/kubeflow/pipelines/issues/7196/comments

https://api.github.com/repos/kubeflow/pipelines/issues/7196/events

https://github.com/kubeflow/pipelines/issues/7196

I_kwDOB-71UM5CVYgR

[frontend] Unable to upload pipeline by URL in KF 1.4

open