Configure execution caching

When Vertex AI Pipelines runs a pipeline, it checks to see whether or not an execution exists in Vertex ML Metadata with the interface (cache key) of each pipeline step.

The step's interface is defined as the combination of the following:

  1. The pipeline step's inputs. These inputs include the input parameters' value (if any) and the input artifact id (if any).

  2. The pipeline step's output definition. This output definition includes output parameter definition (name, if any) and output artifact definition (name, if any).

  3. The component's specification. This specification includes the image, commands, arguments and environment variables being used, as well as the order of the commands and arguments.

Additionally, only the pipelines with the same pipeline name will share the cache.

If there is a matching execution in Vertex ML Metadata, the outputs of that execution are used and the step is skipped. This helps to reduce costs by skipping computations that were completed in a previous pipeline run.

You can turn off execution caching at task level by setting the following:

eval_task.set_caching_options(False)

You can turn off execution caching for an entire pipeline job. When you run a pipeline using PipelineJob(), you can use the enable_caching argument to specify that this pipeline run does not use caching. All steps within the pipeline job will not use caching. Learn more about creating pipeline runs.

Use the following sample to turn off caching:

pl = PipelineJob(
    display_name="My first pipeline",

    # Whether or not to enable caching
    # True = enable the current run to use caching results from previous runs
    # False = disable the current run's use of caching results from previous runs
    # None = defer to cache option for each pipeline component in the pipeline definition
    enable_caching=False,

    # Local or Cloud Storage path to a compiled pipeline definition
    template_path="pipeline.json",

   # Dictionary containing input parameters for your pipeline
    parameter_values=parameter_values,

    # Cloud Storage path to act as the pipeline root
    pipeline_root=pipeline_root,
)

The following limitations apply to this feature:

  • The cached result doesn't have a time-to-live (TTL), and can be reused as long as the entry is not deleted from the Vertex ML Metadata. If the entry is deleted from Vertex ML Metadata, the task will rerun to regenerate the result again.