24 releases

0.1.13 Jul 21, 2024
0.1.12 Dec 11, 2023
0.1.5 Mar 13, 2023
0.1.4 Feb 10, 2023
0.0.7 Dec 31, 2022

#43 in Procedural macros

Download history 32/week @ 2024-08-17 28/week @ 2024-08-24 34/week @ 2024-08-31 21/week @ 2024-09-07 52/week @ 2024-09-14 71/week @ 2024-09-21 75/week @ 2024-09-28 13/week @ 2024-10-05 17/week @ 2024-10-12 25/week @ 2024-10-19 31/week @ 2024-10-26 24/week @ 2024-11-02 2/week @ 2024-11-09 17/week @ 2024-11-16 34/week @ 2024-11-23 41/week @ 2024-11-30

94 downloads per month
Used in 17 crates

ISC license

110KB
2.5K SLoC

Terrars is a tool for building Terraform stacks in Rust. This is an alternative to the CDK.

See a working example in helloworld.

Current status: Usable, but may have some rough edges and missing features. I may continue to tweak things to improve ergonomics.

Why use this or the CDK instead of raw Terraform?

  • All stacks eventually get complicated to the point you need a full programming language to generate them. CDK and this aren't particularly more verbose than raw Terraform, so I'd always use them from the start.
  • Reuse constants and datastructures from your code (json structures, environment variables, endpoints and reverse routing functions) when defining your infrastructure
  • Autocompletion, edit-time verification via types

Why use this instead of the CDK?

  • It's Rust, which you're already using
  • More static type safety - the CDK ignores types in a number of situations and munges required and optional parameters together
  • Pre-generated bindings
  • No complicated type hierarchy with scopes, inheritance, etc.
  • Fewer layers - cdk requires terraform, a cdk CLI, Javascript tools, Javascript package directories, and depending on what language you use that language itself as well. CDK generation requires a json spec -> typescript -> generated javascript -> final language translation process. terrars only requires terraform both during generation and runtime and goes directly from the JSON spec to Rust.

Why not use this instead of the CDK?

  • You need to create your own workflow. You can create a simple build.rs file, but if you want a more complex wrapper you need to write it yourself.

Pre-generated bindings

Infrastructure

Other services

Local, software

Getting started

Note: There's a full, working example in helloworld.

  1. Add terrars and pre-generated bindings such as terrars-andrewbaxter-stripe or else generate your own (see Generation below) to your project. Enable the features you want to use in the bindings.

  2. Develop your code (ex: build.rs)

    Create a Stack and set up providers:

    let mut stack = &mut BuildStack{}.build();
    BuildProviderStripe {
        token: STRIPE_TOKEN,
    }.build(stack);
    

    The first provider instance for a provider type will be used by default for that provider's resources, so you don't need to bind it.

    Then create resources:

    let my_product = BuildProduct {
        name: "My Product".into(),
    }.build(stack);
    let my_price = BuildPrice {
        ...
    }.build(stack);
    my_price.set_product(my_product.id());
    ...
    

    Finally, write the stack out:

    fs::write("mystack.tf.json", &stack.serialize("state.json")?)?;
    
  3. Call terraform as usual in the directory you generated mystack.tf.json in

    (Stack also has methods run() and get_output() to call terraform for you. You must have terraform in your path.)

Generating bindings

While there are premade crates for some providers, you can generate code for new providers locally using terrars-generate.

  1. Install the generate cli with cargo install terrars

  2. Create a config file. As an example, to use hashicorp/aws, create a json file (ex: terrars_aws.json) with the specification of what you want to generate:

    {
      "provider": "hashicorp/aws",
      "version": "4.48.0",
      "include": [
        "cognito_user_pool",
        "cognito_user_pool_client",
        "cognito_user_pool_domain",
        "cognito_user_pool_ui_customization",
        "route53_zone",
        "route53_record",
        "aws_acm_certificate",
        "aws_acm_certificate_validation"
      ],
      "dest": "src/bin/mydeploy/tfschema/aws"
    }
    

    tfschema/aws must be an otherwise unused directory - it will be wiped when you genenerate the code. If include is missing or empty, this will generate everything (alternatively, you can use exclude to blacklist resources/datasources). Resources and datasources don't include the provider prefix (aws_ in this example). Datasources start with data_.

  3. Make sure you have terraform in your PATH. Run cargo install terrars, then terrars-generate terrars_aws.json.

  4. The first time you do this, create a src/bin/mydeploy/tfschema/mod.rs file with this contents to root the generated provider:

    pub mod aws;
    

General usage

Definitions

There are Build* structs containing required parameters and a build method for most schema items (resources, stack, variables, outputs, etc). The build method registers the item in the Stack if applicable. Optional parameters can be set on the value returned from build.

Expressions

Background: In Terraform, all fields regardless of type can be assigned a string template expression for values computed during stack application. Since all strings can potentially be templates, non-template strings must be escaped to avoid accidental interpolation.

How terrars handles it: When defining resources and calling methods, String and &str will be treated as non-template strings and appropriately escaped. To avoid the escaping, you can produce a PrimExpr object via stack.str_expr (to produce an expr that evaluates to a string) or stack.expr for other expression types. To produce the expression body you can use format!() as usual, but note - you must call .raw() on any PrimExprs you use in the new expression to avoid double-antiescaping issues.

If Terraform gives you an error about something with the text _TERRARS_SENTINEL* it means you probably missed a .raw() call on that value (some expression was double-antiescaped).

As a rule of thumb

  1. Converting from expression to string/field is OK. The expression gets turned into a sentinel value and interpolated during writing the template
  2. Converting from string/field with no sentinel values (literals, etc) to expression is OK.
  3. Converting string/field containing sentinel values -> expression is BAD. The sentinel replacement will happen twice and you'll have broken data. This can only happen if you convert an expression into a string and then back, so shouldn't happen often.

For-each

Lists, sets, and record references have a .map method which takes care of all the different "for" methods in Terraform. Specifically

  • Call .map and define a resource: does resource-level for-each (per Terraform limitations, this cannot be done on lists derived from other resources so has very limited use, you should probably just use a for loop)
  • Call .map and define a block element: does block-level for-each
  • Call .map and return an attribute reference: produces an attribute for expression

.map always produces a list reference, but this can be assgned to set fields as well. .map_rec is similar to .map but results in a record.

Vecs and maps of primitives

There's two helper macros for generating vecs and maps of primitive values:

  • primvec![v, ...] - creates a vec of primitive values, converting each value into a primitive if it is not. Use like primvec!["stringone", "stringtwo"] (easier than vec!["stringone".into(), "stringtwo".into()]).
  • primmap!{"k" = v, ...} - creates a map of strings to primitive values, converting each value into a primitive if it is not. Same as above, performs automatic conversion.

How it works

Terraform provides a method to output provider schemas as json. This tool uses that schema to generate structures that would output matching json Terraform stack files.

Expressions/template strings/interpolation/escaping

Take as an example:

format!("{}{}", my_expr, verbatim_string))

This code would somehow need to escape the pattern and verbatim_string, while leaving my_expr unescaped, and the result would need to be treated as an "expression" to prevent escaping if it's used again in another format! or something. This applies to not just format! but serde serialization (json), other methods.

For now Terrars uses a simple (somewhat dirty) hack to avoid this. All expressions are put into a replacement table, and a sentinel string (ex: _TERRARS_SENTINEL_99_) is used instead. During final stack json serialization, the strings are escaped and then the original expression text is substituted back in , replacing the sentinel text.

This way, all normal string formatting methods should retain the expected expressions.

Current limitations and warnings

  • Not all Terraform features have been implemented

    The only one I'm aware of missing at the moment is resource Provisioning.

  • ignore_changes takes strings rather than an enum

  • No variable or output static type checking

    I'd like to add a derive macro for generating variables/outputs automatically from a structure at some point.

  • Non-local deployment methods

    I think this is easy, but I haven't looked into it yet.

The name

I originally called this terrarust but then I realized it sounded like terrorist so I decided to play it safe and chopped out the u t which stands for unreal tournament.

Dependencies

~12–23MB
~350K SLoC