As of December 2020, AWS supports running Lambda functions from any container image. This is an example of deploying Julia program as AWS lambda.
The Dockerfile uses an AWS provided base image. It is more convenient because their
base image already includes the
Runtime Interface Emulator (RIE). It would be possible
to use a different base image but then the RIE has to be installed separately.
Julia is downloaded and installed from the official julialang.org web site.
Project content are copied to the standard deployment directory /var/task.
By setting JULIA_DEPOT_PATH environment variable, the precompiled files
will be stored in a known location.
Note that AWS Lambda provides a read-only file system. If needed, the JULIA_DEPOT_PATH
can be reset to include /tmp/.julia for additional precompilation during runtime.
See AWS FAQ for container images
for more information.
As required, Lambda functions must support the
Lambda Runtime API.
The main.jl file implements the general workflow of this API, which involves
fetching lambda invocation requests, calling the handle_event function,
and reporting results back to the Lambda execution environment. This runtime is normally
provided by AWS for the officially supported languages. For Julia, we're just rolling
our own.
This main program delegates lambda requests to the underlying module. In this case,
it's called JuliaLambdaExample. The module must define a function called
handle_event and the function must accept two arguments - event_data and headers.
The return value of this function is then passed back to the lambda environment.
To build your own AWS Lambda function, you can copy this repo and rename the
JuliaLambdaExample to whatever you want. Just make sure that all references
are renamed properly.
There is a convenient shell script in scripts/deploy.sh that can be used to
quickly build/tag/push a Docker image and deploy the function on AWS.
For example:
sh scripts/deploy.sh julia-lambda latest
The script does not deploy the lambda function unless it is already created. Hence, just for the first time, you must create the lambda function using your preferred approach (web interface, cloud formation, CDK, etc.)
Feel free to raise an issue or PR if you would like to contribute to this repo.