Pulumi is a powerful tool for managing infrastructure as code, and its flexibility across different languages makes it a popular choice among developers. While Pulumi's TypeScript syntax offers a clean and convenient way to handle Outputs and Inputs, translating these features to Python can be challenging. This article explores the nuances of using pulumi.interpolate in TypeScript and how to achieve similar functionality in Python.

Pulumi Interpolate

In the TypeScript syntax of Pulumi, there is a clean approach for concatenating Outputs. It leverages tagged template literals, which are not available in Python. As per the Pulumi reference docs, interpolate is similar to concat but is designed to be used as a tagged template expression. For example:

// 'server' and 'loadBalancer' are both resources that expose [Output] properties.
let val: Output = pulumi.interpolate `http://${server.hostname}:${loadBalancer.port}`

As with concat, the 'placeholders' between ${} can be any Inputs, i.e., they can be Promises, Outputs, or just plain JavaScript values.

Having done most of my Pulumi work in TypeScript, I frequently used the pulumi.interpolate tagged template literal whenever I needed to pass an Input into a new resource. Without giving it much thought, I used it extensively without comparing it deeply to pulumi.concat or apply. However, when I started working with Pulumi in Python and reached for pulumi.interpolate, I realized it was missing.

This prompted a deeper dive into understanding what it means to be an Output vs. an Input and how to translate:

pulumi.interpolate`http://${server.hostname}:${loadBalancer.port}`

to:

pulumi.concat('http://', server.hostname, ':', loadBalancer.port)

Output

Outputs are values from resources that may be populated or will resolve and be populated in the future. Because an Output is associated with the resource it comes from, an edge can be created when it's passed as an Input to pulumi.interpolate or pulumi.concat, and later used to create another resource. The dependency graph between resources, created by the nodes (resources) and their edges (Output -> Input), allows Pulumi to create resources in the correct order and ensures that Outputs are populated when needed by the next resource in the graph.

Input

An input can be a raw value, a promise, or an Output. If an Input to a resource is an Output, then you have a reference to the resource where the Output was originally created. The fact that an Input can be an Output enables it to trace its dependencies.

Here's its type definition:

type Input = T | Promise | OutputInstance;

Tagged Template Literals in 30 Seconds

Here’s an example of how we could uppercase just the values (the "placeholders" between ${}), without altering the literal string portion of the template literal:

function uppercaseValues(strings, ...values) {
  const result = [];
  strings.forEach((string, i) => {
    result.push(string);
    if (i 




  
  
  Implementing pulumi.interpolate


Without knowing the exact source code, and expanding from the example above, we can imagine how to implement pulumi.interpolate on our own. It might look something like this:


function interpolate(strings, ...values) {
  const result = [];
  strings.forEach((string, i) => {
    result.push(string);
    if (i 



All we did was replace the final join call with a call to pulumi.concat. If this were the implementation, we'd perform checks on whether raw strings need to be unwrapped from Output types, instead of operating just on the placeholders, which is what the real implementation does.

Its function definition in TypeScript is:


function interpolate(literals: TemplateStringsArray, ...placeholders: Input[]): Output;




which is very similar to concat:


function concat(...params: Input[]): Output



The lightbulb moment comes when you realize that you're really just forwarding along Output values and wrapping them in parent Outputs.


  
  
  Back to Python


You can make some silly mistakes when porting interpolate over to concat. Let’s demonstrate with an example.

In TypeScript, I would have done this:


function get_image_name(imageRegistry: Repository, name: string, version: Input) {
    return pulumi.interpolate`${image_registry.repository_id}/${name}:${version}`
}




When porting to Python, I might end up with this:


def get_image_tag(image_registry: Repository, name: str, version: Input[str]):
    return pulumi.Output.concat(
        image_registry.repository_id,
        f"/{name}:{version}"
    )




However, interpolate was iterating over every placeholder individually to create dependencies and resolve outputs. With our Python code, we’ve subtly lost that connection with the version argument. We need to break up our Outputs manually and surface them as individual arguments to pulumi.Output.concat.

The corrected code would look like this:


def get_image_tag(image_registry: Repository, name: str, version: Input[str]):
    return pulumi.Output.concat(
        image_registry.repository_id,
        f"/{name}:",
        version
    )




Now, the version will be correctly included in the dependency graph, and we’ll be error-free!


  
  
  Conclusion


Translating pulumi.interpolate from TypeScript to Python requires a deeper understanding of how Outputs and Inputs work in Pulumi. While Python does not support tagged template literals, using pulumi.concat effectively allows us to achieve similar functionality. By manually managing dependencies and ensuring all Output values are properly handled, we can ensure our Pulumi code in Python is just as robust and efficient as in TypeScript.