LLRT (Low Latency Runtime) is a lightweight JavaScript runtime designed to address the growing demand for fast and efficient Serverless applications. LLRT offers up to over 10x faster startup and up to 2x overall lower cost compared to other JavaScript runtimes running on AWS Lambda
It's built in Rust, utilizing QuickJS as JavaScript engine, ensuring efficient memory usage and swift startup.
Warning
LLRT is an experimental package. It is subject to change and intended only for evaluation purposes.
LLRT - DynamoDB Put, ARM, 128MB:
Node.js 20 - DynamoDB Put, ARM, 128MB:
HTTP benchmarks measured in round trip time for a cold start (why?)
Configure Lambda functions to use LLRT
Download the last LLRT release from https://github.com/awslabs/llrt/releases
Option 1: Custom runtime (recommended)
Choose Custom Runtime on Amazon Linux 2023 and package the LLRT bootstrap binary together with your JS code.
Option 2: Use a layer
Choose Custom Runtime on Amazon Linux 2023, upload llrt-lambda-arm64.zip or llrt-lambda-x64.zip as a layer and add to your function
Option 3: Package LLRT in a container image
See our AWS SAM example or:
FROM --platform=arm64 busybox
WORKDIR /var/task/
COPY app.mjs ./
ADD https://github.com/awslabs/llrt/releases/latest/download/llrt-container-arm64 /usr/bin/llrt
RUN chmod +x /usr/bin/llrt
ENV LAMBDA_HANDLER "app.handler"
CMD [ "llrt" ]That's it 🎉
Important
Even though LLRT supports ES2020 it's NOT a drop in replacement for Node.js. Consult Compatibility matrix and API for more details.
All dependencies should be bundled for a browser platform and mark included @aws-sdk packages as external.
Option 4: AWS SAM
The following example project sets up a lambda instrumented with a layer containing the llrt runtime.
Option 5: AWS CDK
You can use cdk-lambda-llrt construct library to deploy LLRT Lambda functions with AWS CDK.
import { LlrtFunction } from "cdk-lambda-llrt";
const handler = new LlrtFunction(this, "Handler", {
entry: "lambda/index.ts",
});See Construct Hub and its examples for more details.
Testing & ensuring compatibility
The best way to ensure your code is compatible with LLRT is to write tests and execute them using the built-in test runner. The test runner currently supports Jest/Chai assertions. There are two main types of tests you can create:
Unit Tests
- Useful for validating specific modules and functions in isolation
- Allow focused testing of individual components
End-to-End (E2E) Tests
- Validate overall compatibility with AWS SDK and WinterCG compliance
- Test the integration between all components
- Confirm expected behavior from end-user perspective
For more information about the E2E Tests and how to run them, see here.
Test runner
Test runner uses a lightweight Jest-like API and supports Jest/Chai assertions. For examples on how to implement tests for LLRT see the /tests folder of this repository.
To run tests, execute the llrt test command. LLRT scans the current directory and sub-directories for files that ends with *.test.js or *.test.mjs. You can also provide a specific test directory to scan by using the llrt test -d <directory> option.
The test runner also has support for filters. Using filters is as simple as adding additional command line arguments, i.e: llrt test crypto will only run tests that match the filename containing crypto.
Compatibility matrix
Note
LLRT only support a fraction of the Node.js APIs. It is NOT a drop in replacement for Node.js, nor will it ever be. Below is a high level overview of partially supported APIs and modules. For more details consult the API documentation
| Node.js | LLRT |
|
|---|---|---|
| buffer | ✔︎ | ✔︎️ |
| streams | ✔︎ | ✔︎* |
| child_process | ✔︎ | ✔︎⏱ |
| net:sockets | ✔︎ | ✔︎⏱ |
| net:server | ✔︎ | ✔︎ |
| tls | ✔︎ | ✘⏱ |
| fetch | ✔︎ | ✔︎ |
| http | ✔︎ | ✘⏱** |
| https | ✔︎ | ✘⏱** |
| fs/promises | ✔︎ | ✔︎ |
| fs | ✔︎ | ✘⏱ |
| path | ✔︎ | ✔︎ |
| timers | ✔︎ | ✔︎ |
| crypto | ✔︎ | ✔︎ |
| process | ✔︎ | ✔︎ |
| encoding | ✔︎ | ✔︎ |
| console | ✔︎ | ✔︎ |
| events | ✔︎ | ✔︎ |
| ESM | ✔︎ | ✔︎ |
| CJS | ✔︎ | ✔︎ |
| async/await | ✔︎ | ✔︎ |
| Other modules | ✔︎ | ✘ |
Using node_modules (dependencies) with LLRT
Since LLRT is meant for performance critical application it's not recommended to deploy node_modules without bundling, minification and tree-shaking.
LLRT can work with any bundler of your choice. Below are some configurations for popular bundlers:
Warning
LLRT implements native modules that are largely compatible with the following external packages. By implementing the following conversions in the bundler's alias function, your application may be faster, but we recommend that you test thoroughly as they are not fully compatible.
| Node.js | LLRT |
|---|---|
| fast-xml-parser | llrt:xml |
| uuid | llrt:uuid |
ESBuild
esbuild index.js --platform=node --target=es2020 --format=esm --bundle --minify --external:@aws-sdk --external:@smithyRollup
import resolve from "@rollup/plugin-node-resolve";
import commonjs from "@rollup/plugin-commonjs";
import terser from "@rollup/plugin-terser";
export default {
input: "index.js",
output: {
file: "dist/bundle.js",
format: "esm",
sourcemap: true,
target: "es2020",
},
plugins: [resolve(), commonjs(), terser()],
external: ["@aws-sdk", "@smithy"],
};Webpack
import TerserPlugin from "terser-webpack-plugin";
import nodeExternals from "webpack-node-externals";
export default {
entry: "./index.js",
output: {
path: "dist",
filename: "bundle.js",
libraryTarget: "module",
},
target: "web",
mode: "production",
resolve: {
extensions: [".js"],
},
externals: [nodeExternals(), "@aws-sdk", "@smithy"],
optimization: {
minimize: true,
minimizer: [
new TerserPlugin({
terserOptions: {
ecma: 2020,
},
}),
],
},
};Using AWS SDK (v3) with LLRT
LLRT includes many AWS SDK clients and utils as part of the runtime, built into the executable. These SDK Clients have been specifically fine-tuned to offer best performance while not compromising on compatibility. LLRT replaces some JavaScript dependencies used by the AWS SDK by native ones such as Hash calculations and XML parsing. V3 SDK packages not included in the list below have to be bundled with your source code. For an example on how to use a non-included SDK, see this example build script (buildExternalSdkFunction)
| Bundled AWS SDK packages |
|---|
| @aws-sdk/client-cloudwatch-events |
| @aws-sdk/client-cloudwatch-logs |
| @aws-sdk/client-cognito-identity |
| @aws-sdk/client-cognito-identity-provider |
| @aws-sdk/client-dynamodb |
| @aws-sdk/client-eventbridge |
| @aws-sdk/client-kms |
| @aws-sdk/client-lambda |
| @aws-sdk/client-s3 |
| @aws-sdk/client-secrets-manager |
| @aws-sdk/client-ses |
| @aws-sdk/client-sfn |
| @aws-sdk/client-sns |
| @aws-sdk/client-sqs |
| @aws-sdk/client-ssm |
| @aws-sdk/client-sts |
| @aws-sdk/client-xray |
| @aws-sdk/credential-providers |
| @aws-sdk/lib-dynamodb |
| @aws-sdk/lib-storage |
| @aws-sdk/s3-presigned-post |
| @aws-sdk/s3-request-presigner |
| @aws-sdk/util-dynamodb |
| @aws-sdk/util-user-agent-browser |
| @smithy |
| @aws-crypto |
Important
LLRT currently does not support returning streams from SDK responses. Use response.Body.transformToString(); or response.Body.transformToByteArray(); as shown below.
const response = await client.send(command);
// or 'transformToByteArray()'
const str = await response.Body.transformToString();Running TypeScript with LLRT
Same principle as dependencies applies when using TypeScript. TypeScript must be bundled and transpiled into ES2020 JavaScript.
Note
LLRT will not support running TypeScript without transpilation. This is by design for performance reasons. Transpiling requires CPU and memory that adds latency and cost during execution. This can be avoided if done ahead of time during deployment.
Rationale
What justifies the introduction of another JavaScript runtime in light of existing options such as Node.js, Bun & Deno?
Node.js, Bun, and Deno represent highly proficient JavaScript runtimes. However, they are designed with general-purpose applications in mind. These runtimes were not specifically tailored for the demands of a Serverless environment, characterized by short-lived runtime instances. They each depend on a (Just-In-Time compiler (JIT) for dynamic code compilation and optimization during execution. While JIT compilation offers substantial long-term performance advantages, it carries a computational and memory overhead.
In contrast, LLRT distinguishes itself by not incorporating a JIT compiler, a strategic decision that yields two significant advantages:
A) JIT compilation is a notably sophisticated technological component, introducing increased system complexity and contributing substantially to the runtime's overall size.
B) Without the JIT overhead, LLRT conserves both CPU and memory resources that can be more efficiently allocated to code execution tasks, thereby reducing application startup times.
Limitations
There are many cases where LLRT shows notable performance drawbacks compared with JIT-powered runtimes, such as large data processing, Monte Carlo simulations or performing tasks with hundreds of thousands or millions of iterations. LLRT is most effective when applied to smaller Serverless functions dedicated to tasks such as data transformation, real time processing, AWS service integrations, authorization, validation etc. It is designed to complement existing components rather than serve as a comprehensive replacement for everything. Notably, given its supported APIs are based on Node.js specification, transitioning back to alternative solutions requires minimal code adjustments.
Building from source
Clone code and cd to directory
git clone git@github.com:awslabs/llrt.git --recursive
cd llrt
Install git submodules if you've not cloned the repository with --recursive
git submodule update --init
Install rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | bash -s -- -y
source "$HOME/.cargo/env"
Install dependencies
# MacOS
brew install zig make cmake zstd node corepack
# Ubuntu
sudo apt -y install make zstd
sudo snap install zig --classic --beta
# Windows WSL2
sudo apt -y install cmake g++ gcc make zip zstd
sudo snap install zig --classic --beta
# Windows WSL2 (If Node.js is not yet installed)
sudo curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/master/install.sh | bash
nvm install --lts
Install Node.js packages
corepack enable
yarn
Install generate libs and setup rust targets & toolchains
make stdlib && make libs
Note
If these commands exit with an error that says can't cd to zstd/lib,
you've not cloned this repository recursively. Run git submodule update --init to download the submodules and run the commands above again.
Build release for Lambda
make release-arm64
# or for x86-64, use
make release-x64
Optionally build for your local machine (Mac or Linux)
make release
You should now have a llrt-lambda-arm64.zip or llrt-lambda-x64.zip. You can manually upload this as a Lambda layer or use it via your Infrastructure-as-code pipeline
Running Lambda emulator
Please note that in order to run the example you will need:
- Valid AWS credentials via a
~/.aws/credentialsor via environment variables.
export AWS_ACCESS_KEY_ID=XXX
export AWS_SECRET_ACCESS_KEY=YYY
export AWS_REGION=us-east-1- A DynamoDB table (with
idas the partition key) onus-east-1 - The
dynamodb:PutItemIAM permission on this table. You can use this policy (don't forget to modify <YOUR_ACCOUNT_ID>):
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "putItem",
"Effect": "Allow",
"Action": "dynamodb:PutItem",
"Resource": "arn:aws:dynamodb:us-east-1:<YOUR_ACCOUNT_ID>:table/quickjs-table"
}
]
}Start the lambda-server.js in a separate terminal
node lambda-server.js
Then run llrt:
make run
Environment Variables
LLRT_EXTRA_CA_CERTS=file
Load extra certificate authorities from a PEM encoded file
LLRT_GC_THRESHOLD_MB=value
Set a memory threshold in MB for garbage collection. Default threshold is 20MB
LLRT_HTTP_VERSION=value
Restrict HTTP requests to use a specific version. By default HTTP 1.1 and 2 are enabled. Set this variable to 1.1 to only use HTTP 1.1
LLRT_LOG=[target][=][level][,...]
Filter the log output by target module, level, or both (using =). Log levels are case-insensitive and will also enable any higher priority logs.
Log levels in descending priority order:
ErrorWarn | WarningInfoDebugTrace
Example filters:
warnwill enable all warning and error logsllrt_core::vm=tracewill enable all logs in thellrt_core::vmmodulewarn,llrt_core::vm=tracewill enable all logs in thellrt_core::vmmodule and all warning and error logs in other modules
LLRT_NET_ALLOW="host[ ...]"
Space-delimited list of hosts or socket paths which should be allowed for network connections. Network connections will be denied for any host or socket path missing from this list. Set an empty list to deny all connections
LLRT_NET_DENY="host[ ...]"
Space-delimited list of hosts or socket paths which should be denied for network connections
LLRT_NET_POOL_IDLE_TIMEOUT=value
Set a timeout in seconds for idle sockets being kept-alive. Default timeout is 15 seconds
LLRT_TLS_VERSION=value
Set the TLS version to be used for network connections. By default only TLS 1.2 is enabled. TLS 1.3 can also be enabled by setting this variable to 1.3
Benchmark Methodology
Although Init Duration reported by Lambda is commonly used to understand cold start impact on overall request latency, this metric does not include the time needed to copy code into the Lambda sandbox.
The technical definition of Init Duration (source):
For the first request served, the amount of time it took the runtime to load the function and run code outside of the handler method.
Measuring round-trip request duration provides a more complete picture of user facing cold-start latency.
Lambda invocation results (λ-labeled row) report the sum total of Init Duration + Function Duration.
Security
See CONTRIBUTING for more information.
License
This library is licensed under the Apache-2.0 License. See the LICENSE file.