Managing steps
Managing steps in ZenML.
We have learned a lot about step configurations in the last section. Here we go into even further detail on how to manage steps and their configuration in ZenML.
Type annotations
Your functions will work as ZenML steps even if you don't provide any type annotations for their inputs and outputs. However, adding type annotations to your step functions gives you lots of additional benefits:
Type validation of your step inputs: ZenML makes sure that your step functions receive an object of the correct type from the upstream steps in your pipeline.
Better serialization: Without type annotations, ZenML uses Cloudpickle to serialize your step outputs. When provided with type annotations, ZenML can choose a materializer that is best suited for the output. In case none of the builtin materializers work, you can even write a custom materializer.
ZenML provides a built-in CloudpickleMaterializer that can handle any object by saving it with cloudpickle. However, this is not production-ready because the resulting artifacts cannot be loaded when running with a different Python version. In such cases, you should consider building a custom Materializer to save your objects in a more robust and efficient format.
Moreover, using the CloudpickleMaterializer
could allow users to upload of any kind of object. This could be exploited to upload a malicious file, which could execute arbitrary code on the vulnerable system.
It is impossible for ZenML to detect whether you want your step to have a single output artifact of type Tuple
or multiple output artifacts just by looking at the type annotation.
We use the following convention to differentiate between the two: When the return
statement is followed by a tuple literal (e.g. return 1, 2
or return (value_1, value_2)
) we treat it as a step with multiple outputs. All other cases are treated as a step with a single output of type Tuple
.
If you want to make sure you get all the benefits of type annotating your steps, you can set the environment variable ZENML_ENFORCE_TYPE_ANNOTATIONS
to True
. ZenML will then raise an exception in case one of the steps you're trying to run is missing a type annotation.
Step output names
By default, ZenML uses the output name output
for single output steps and output_0, output_1, ...
for steps with multiple outputs. These output names are used to display your outputs in the dashboard and fetch them after your pipeline is finished.
If you want to use custom output names for your steps, use the Annotated
type annotation:
If you do not give your outputs custom names, the created artifacts will be named {pipeline_name}::{step_name}::output
or {pipeline_name}::{step_name}::output_{i}
in the dashboard. See the documentation on artifact versioning and configuration for more information.
Parameters for your steps
When calling a step in a pipeline, the inputs provided to the step function can either be an artifact or a parameter. An artifact represents the output of another step that was executed as part of the same pipeline and serves as a means to share data between steps. Parameters, on the other hand, are values provided explicitly when invoking a step. They are not dependent on the output of other steps and allow you to parameterize the behavior of your steps.
In order to allow the configuration of your steps using a configuration file, only values that can be serialized to JSON using Pydantic can be passed as parameters. If you want to pass other non-JSON-serializable objects such as NumPy arrays to your steps, use External Artifacts instead.
Parameters of steps and pipelines can also be passed in using YAML configuration files. The following configuration file and Python code can work together and give you the flexibility to update configuration only in YAML file, once needed:
There might be conflicting settings for step or pipeline inputs, while working with YAML configuration files. Such situations happen when you define a step or a pipeline parameter in the configuration file and override it from the code later on. Don't worry - once it happens you will be informed with details and instructions how to fix. Example of such a conflict:
Parameters and caching
When an input is passed as a parameter, the step will only be cached if all parameter values are exactly the same as for previous executions of the step.
Artifacts and caching
When an artifact is used as a step function input, the step will only be cached if all the artifacts are exactly the same as for previous executions of the step. This means that if any of the upstream steps that produce the input artifacts for a step were not cached, the step itself will always be executed.
Using a custom step invocation ID
When calling a ZenML step as part of your pipeline, it gets assigned a unique invocation ID that you can use to reference this step invocation when defining the execution order of your pipeline steps or use it to fetch information about the invocation after the pipeline has finished running.
Control the execution order
By default, ZenML uses the data flowing between steps of your pipeline to determine the order in which steps get executed.
The following example shows a pipeline in which step_3
depends on the outputs of step_1
and step_2
. This means that ZenML can execute both step_1
and step_2
in parallel but needs to wait until both are finished before step_3
can be started.
If you have additional constraints on the order in which steps get executed, you can specify non-data dependencies by passing the invocation IDs of steps that should run before your step like this: my_step(after="other_step")
. If you want to define multiple upstream steps, you can also pass a list for the after
argument when calling your step: my_step(after=["other_step", "other_step_2"])
.
Check out the previous section to learn about the invocation ID and how to use a custom one for your steps.
This pipeline is similar to the one explained above, but this time ZenML will make sure to only start step_1
after step_2
has finished.
Enable or disable logs storing
By default, ZenML uses a special logging handler to capture the logs that occur during the execution of a step. These logs are stored within the respective artifact store of your stack.
You can display the logs in the dashboard as follows:
If you do not want to store the logs in your artifact store, you can:
Disable it by using the
enable_step_logs
parameter either with your@pipeline
or@step
decorator:Disable it by using the environmental variable
ZENML_DISABLE_STEP_LOGS_STORAGE
and setting it totrue
. This environmental variable takes precedence over the parameters mentioned above.
Last updated