Kubeflow Pipelines overview in MLOps - Step-by-Step Execution

Process Flow - Kubeflow Pipelines overview

Define pipeline components

↓

Assemble components into pipeline

↓

Compile pipeline to YAML

↓

Upload pipeline to Kubeflow UI

↓

Run pipeline experiment

↓

Pipeline executes steps in order

↓

Monitor step status and logs

↓

View results and outputs

↓

Pipeline completes

Kubeflow Pipelines lets you build, run, and monitor machine learning workflows step-by-step using components assembled into a pipeline.

Execution Sample

MLOps

@dsl.component
def preprocess():
    pass

@dsl.component
def train():
    pass

@dsl.pipeline(name='simple-pipeline')
def pipeline():
    preprocess_task = preprocess()
    train_task = train().after(preprocess_task)

Defines two steps and assembles them into a simple Kubeflow pipeline.

Process Table

Step	Action	Evaluation	Result
1	Define preprocess component	Function created	preprocess component ready
2	Define train component	Function created	train component ready
3	Create pipeline function	Pipeline function defined	pipeline structure ready
4	Instantiate preprocess_task	Call preprocess()	preprocess step added
5	Instantiate train_task	Call train()	train step added
6	Compile pipeline	Convert to YAML	pipeline.yaml generated
7	Upload pipeline.yaml	Upload to Kubeflow UI	Pipeline available in UI
8	Run pipeline	Start experiment	Pipeline execution started
9	Execute preprocess step	Run preprocess	Preprocess step completed
10	Execute train step	Run train	Train step completed
11	Pipeline completes	All steps done	Pipeline run successful

💡 All pipeline steps executed successfully, pipeline run ends.

Status Tracker

Variable	Start	After Step 4	After Step 5	After Step 8	After Step 11
preprocess_task	None	preprocess component instance	preprocess component instance	running	completed
train_task	None	None	train component instance	pending	completed
pipeline_status	Not started	Not started	Not started	Running	Succeeded

Key Moments - 3 Insights

Why do we define components before assembling the pipeline?

Components are the building blocks; defining them first (see steps 1 and 2 in execution_table) allows us to reuse and organize tasks before creating the pipeline.

What happens when we compile the pipeline?

Compiling (step 6) converts the Python pipeline code into a YAML file that Kubeflow understands to run the workflow.

How does Kubeflow execute the pipeline steps?

Kubeflow runs steps in order respecting dependencies (steps 9 and 10), monitoring each step's status until completion.

Visual Quiz - 3 Questions

Look at the execution_table, what is the state of 'train_task' after step 5?

A running

B None

C train component instance

D completed

💡 Hint

Check the 'train_task' variable in variable_tracker after Step 5.

At which step does the pipeline start running in Kubeflow?

A Step 6

B Step 8

C Step 9

D Step 11

💡 Hint

Look for 'Pipeline execution started' in execution_table.

If the 'preprocess' step fails, what happens to the pipeline execution?

A Pipeline stops and marks failure

B Pipeline continues to 'train' step

C Pipeline retries 'train' step

D Pipeline skips 'preprocess' and completes

💡 Hint

Refer to the execution flow where steps run in order and depend on previous success.

Concept Snapshot

Kubeflow Pipelines let you build ML workflows by:
- Defining reusable components (steps)
- Assembling them into a pipeline function
- Compiling to YAML for Kubeflow
- Uploading and running via Kubeflow UI
- Monitoring step status and outputs
Steps run in order respecting dependencies.

Full Transcript

Kubeflow Pipelines help you create machine learning workflows by defining small tasks called components. You write these components as functions, then combine them into a pipeline function. This pipeline is compiled into a YAML file that Kubeflow understands. You upload this YAML to the Kubeflow UI and start an experiment to run the pipeline. Kubeflow runs each step in order, showing status and logs. When all steps finish, the pipeline run completes successfully.