Moving my Lambda Functions to Rust

My experience changing the AWS Lambda functions for this website from Python to Rust.

7 Feb 2022
7 min read

Since I’ve started using Rust quite a bit more at work and in some personal projects, I’ve been wondering how well Rust would fair when used in AWS Lambda functions. Back in 2018, AWS published a blog post announcing a runtime for Rust on AWS Lambda. More recently, I found the comments on this reddit post to be quite interesting. Aleksandr Filichkin also published in September last year an AWS Lambda Battle that showed the performance of Rust on AWS Lambda was quite impressive. In a later post in November, Aleksandr wrote about the performance of x86 vs ARM on AWS Lambda, which again showed Rust to be quite performant, and even more so on ARM.

GitHub - awslabs/aws-lambda-rust-runtime: A Rust runtime for AWS Lambda
GitHub - awslabs/aws-lambda-rust-runtime: A Rust runtime for AWS Lambda
A Rust runtime for AWS Lambda. Contribute to awslabs/aws-lambda-rust-runtime development by creating an account on GitHub.
GitHub GitHub awslabs

Site Analytics

A few months ago I decided to change the analytics for this website over to a custom analytics implementation, replacing my use of Simple Analytics. The analytics are now provided by a couple of lambda functions: one which provides the API for collecting statistics and reporting, and another that processes DynamoDB events to provide simple aggregation.

I’ve been fairly pleased with this approach: it means I don’t really pay anything for the analytics, as AWS Lambda and DynamoDB are extremely cheap for small use cases like this. Moreover, the reporting provided the minimal amount of information that I currently desire from site analytics.

As a first attempt using Rust on AWS Lambda I felt that these two functions were a good starting point for a few reasons:

  1. The API function is behind an API Gateway proxy, and so I could make use of the lambda-http crate to handle HTTP requests and responses.
  2. The trigger function processes events from a DynamoDB stream, which is another fairly nice use-case for a Lambda function.
  3. There’s little pressure for these to be performant or stable, as it only effects this site 😆

Building Rust for AWS Lambda

I initially had a number of issues compiling Rust code for AWS Lambda using the method described in the README in the AWS Lambda Rust runtime repository. The main issue I came across was a segfault from LLVM when compiling the regex-syntax crate. This seemed to only arise when I compiled in Docker on the M1 when targeting x86-64. As AWS support for ARM in Lambda is quite recent, none of the AWS Lambda Rust build images I looked at currently seemed to support it.

GitHub - softprops/lambda-rust: 🐳 🦀 a dockerized lambda build env for rust applications
GitHub - softprops/lambda-rust: 🐳 🦀 a dockerized lambda build env for rust applications
This docker image extends lambda ci provided.al2 builder docker image, a faithful reproduction of the actual AWS 'provided.al2' Lambda runtime environment, and installs rustup and the stable rust toolchain.
GitHub GitHub awslabs

I’m sure that at some point they will support ARM, but for the time being it was necessary for me to create a Dockerfile that used the al2-arm64 image provided by AWS in the ECR. This Dockerfile simply installs a Rust version (currently 1.58.1) and copies over a build script. When executed, the build script invokes cargo to compile a release build of the lambda functions, and then bundles the compiled executables into a ZIP file, with each executable named bootstrap as required by AWS Lambda.

One caveat that I came across was the use of libssl indirectly from the fernet crate I use to encrypt the session tokens for the analytics reporting API. The libssl library appears not to be present on AWS Lambda, and therefore I needed to include it in the zip package. I considered deploying a Lambda layer for this purpose, however I thought it overkill for just one function. However, the same folder structure can be used to provide library dependencies in a zip package. The somewhat evil trick I use is to invoke ldd to find the resolved library for libssl on the Docker container and copy that file into a lib directory in the zip package.

cp $(ldd "$EXE_PATH" | grep ssl | awk '{print $3}') "$OUTPUT_DIR/lib/"

Now that my executables could be run by Lambda, I could start iterating the API and trigger functions.

Implementing the API

The initial implementation of the API in Rust has gone very easily, mostly due to the structures provided in various crates available to Rust, including the lambda-http crate. I was able to finish most of the API code on one Sunday morning.

Making the calls to DynamoDB in Rust was a very nice experience. The AWS client crate makes use of “builder” syntax, which works quite nicely. For example, to build a query for a particular path and section (the primary and sort keys for the analytics table) is quite easily comprehended, if somewhat verbose:

let res = env
    .ddb
    .query()
    .table_name(env.table_name.to_owned())
    .key_condition_expression("#P = :v1 AND begins_with(#S, :v2)")
    .expression_attribute_names("#P", "Path")
    .expression_attribute_names("#S", "Section")
    .expression_attribute_values(":v1", AttributeValue::S(path))
    .expression_attribute_values(":v2", AttributeValue::S(section))
    .send()
    .await?;

Implementing the API was not entirely a matter of simply translating the Python code. Mostly this was due to the fact that Rust tends to encourage us to better handle errors than Python. I found that there were a number of times in the Python code where I was simply assuming that a value was as I expected. As an example, when we query DynamoDB we get back a dict of attributes in Python, where each attribute is another dict containing the DynamoDB named-value pair. This tends to lead me to write Python code like:

int(item["ViewCount"]["N"])

Unfortunately I’m making a lot of assumptions in the above code, such as:

  1. The item is a dict,
  2. There is a ViewCount attribute,
  3. The ViewCount attribute is a dict,
  4. There is an N attribute in the ViewCount dictionary, and
  5. The value can be parsed as an integer with int().

Rust forced me to be more aware of this: Each item from DynamoDB is a HashMap mapping a String key to an AttributeValue. The equivalent Rust code item["ViewCount"] makes use of the Index trait for HashMap which will panic if the given key is not found in the mapping (see here). This encouraged me to use the get method to access the attributes of an item returned from DynamoDB. The AttributeValue enumeration provides a number of methods that help with unwrapping the enumeration, which all return a Result, which we need to match against (or just lazily unwrap).

if let Some(view_count_attr) = item.get("ViewCount") {
    match view_count_attr.as_n() {
        Ok(view_count) => {
            match i32::from_str_radix(view_count, 10) {
                Ok(n) => n,
                Err(_) => {
                    // We couldn't parse the string as an i32
                }
            }
        }
        Err(_) => {
            // The 'ViewCount' was not an 'N'
        }
    }
} else {
    // There is no 'ViewCount' attribute
}

This results in somewhat cleaner code. That being said, there are times where I felt justified in using unwrap and expect and allowing the Lambda function to panic.

i32::from_str_radix(item["ViewCount"].as_n().unwrap(), 10).unwrap() // 😤

Implementing the Trigger

Once I had come to understand the structures and functions in the Rust AWS client crates, I had a far easier time building the trigger function. This function simply responds to events received from a DynamoDB stream, which it receives as a serde_json::Value. Depending on the contents of the event, it performs various aggregations by updating items to increment the ViewCount attribute for weekly and monthly totals.

I was somewhat worried about the parsing of user agent strings: in Python I did this using the user-agents library. However I found a rather nice crate woothee that performs the same operation just as well for my use case.

Conclusion

I was pleasantly surprised at how well the process went. Apart from the somewhat slow start getting Rust code compiled for AWS Lambda on ARM, once I had my bearings it was quite easy going.

I am somewhat worried that the error handling is not as graceful as it should be. I’m somewhat nervous of cases where I’m using the Index trait of a HashMap, or just unwrap-ing or expect-ing a value where I should be using match and if let. I think I’d better go over the Lambda functions again some other time to round off the edges.

As before, all code is available on GitHub, with the Rust Lambda functions found in the lambda directory.

Cover image courtesy of Jay Heike (@jayrheike on unsplash).