You can use the Kubeflow Pipelines UI to complete tasks like creating, listing, and managing different resources, and viewing and comparing the outputs of experiments. This guide discusses the different operations you can perform using the UI.
The landing page of the UI is a list of all available pipelines, with management controls such as uploading new pipelines, deleting existing ones, or creating an experiment out of a selected pipeline.
From a pipeline’s details page, click Create experiment. This takes you to the new experiment page, where you can specify the experiment’s name and description.
On the experiment’s details page, click Create new run. Fill out the run’s name and description, pick which pipeline it will execute.
The next section is the pipeline’s parameters form. This form is generated automatically based on the selected pipeline. You can try out different pipelines using the pipeline picker dialog.
After you fill in the parameters (some are filled in automatically, if the pipeline author provides default values), click the Create button to start the run. This takes you back to the experiment’s details page.
In the experiment’s details page, you can also click the Create recurring run button. This takes you through a similar flow as creating a run, but here you can choose the kind of trigger that should be used to spawn the runs. You can specify Maximum concurrent runs to limit the number of runs launched in parallel. This can be helpful if the pipeline is expected to run for a long period of time, and is triggered to run frequently.
The experiment’s details page shows all of the runs created or generated in it by default. The page also has a widget to show the number of active recurring runs in this experiment, and you can click the manage button to see a list of recurring run configs, enable and disable them, or see any config’s details.
This page offers two views into the runs scheduled in the cluster:
Clicking a run in the run list (either inside an experiment, or in the top-level flat list of runs) takes you to the Run Details page in the Kubeflow Pipelines UI.
A run is a single execution of a pipeline. You can examine the configuration, status, outputs, execution graph, and logs on the run details page.
The run details page shows the following information:
The breadcrumb at the top of the page shows the following information: Experiments > [experiment name] > [run name], with an icon to the left of the run name indicating its status. For example, Succeeded, Running, Failed. Each of these elements of the breadcrumb is clickable and can be used to navigate.
The Clone button takes you to the Create new run form with the same pipeline selected and all of the parameters set to the same values as the run you are looking at. You can edit everything, but only need to fill in the run name and make sure a pipeline is selected before deploying.
The Refresh button retrieves the latest information from the back end, updating all elements on the page.
There are two main views in the run details page: the Graph view, and the Config view.
Pipelines generally consist of a number of components, and the graph shows the steps that this run has executed so far with arrows indicating parent/child relationships. The graph is viewable as soon as the run begins. You can update it using the Refresh button at the top of the page. Each node within the graph corresponds to a step within the pipeline and is labeled accordingly.
At the top right of each node is an icon indicating its status. For example, Succeeded, Running, Failed, or Skipped. A node can be skipped when its parent contains a conditional and is indicated by a gray triangle pointing towards a vertical line.
Clicking on a node highlights it and opens a resizable panel from the right of the page displaying the name of the run and containing its own three tabs: Artifacts, Input/Output, and Logs.
Artifacts are any viewers associated with a given component. There are many kinds of viewers including:
This simple view includes the inputs and outputs of the node, if any.
This tab displays the Kubernetes pod logs associated with the selected component and may be useful for debugging.
This is a main view within the run details page and consists of some basic information about the run such as its status, creation time, start time, finish time, duration, and the pipeline parameters it was run with.