Skip to contents

Get started with r2lambda

Source: vignettes/articles/r2lambda.qmd

This article walks through the whole life of a Lambda function: build a Docker image, test it locally, deploy it, invoke it, run it on a schedule and delete it. It assumes Docker, curl and AWS credentials are set up as described in the README.

Build a docker image for the lambda function

The examples use two files installed with the package: parity.R, a runtime script whose parity() function tells whether a number is even or odd, and renv.lock, an example lockfile. system.file() returns their paths. Replace them with your own script and lockfile.

runtime_function <- "parity"
runtime_path <- system.file("parity.R", package = "r2lambda")
renvlock_path <- system.file("renv.lock", package = "r2lambda")
dependencies <- NULL

# Might take a while, its building a docker image
build_lambda(
  tag = "parity1",
  runtime_function = runtime_function,
  runtime_path = runtime_path,
  renvlock_path = renvlock_path,
  dependencies = dependencies
)

build_lambda() has a few more options:

  • support_path: files or folders the runtime script needs. Each is copied next to the runtime script under its own name, so support_path = "R" lets the script call source("R/helpers.R").
  • system_deps: Amazon Linux 2023 packages to install with dnf, for R packages that need system libraries, e.g. system_deps = "udunits2" for units. Amazon Linux 2023 has no EPEL, so some libraries, such as GDAL, are not available.
  • architecture: "x86_64" (default) or "arm64". deploy_lambda() creates the function with the image’s architecture. Building for an architecture other than your machine’s needs Docker’s QEMU emulation.

Test the lambda docker image locally 

payload <- list(number = 2)
test_lambda(tag = "parity1", payload)

Deploy to AWS Lambda 

# Might take a while, its pushing it to a remote repository
deploy_lambda(tag = "parity1")

Lambda stops a function after 3 seconds by default, which an R container may not meet, especially on its first call. Pass a longer Timeout, in seconds, and other create_function() settings through ...:

deploy_lambda(tag = "parity1", Timeout = 30, MemorySize = 512)

The function runs under a new IAM role with basic execution permissions. To let it call other AWS services, attach policies to that role, or use a role you already have:

deploy_lambda(
  tag = "parity1",
  policy_arns = "arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess"
)

deploy_lambda(tag = "parity1", role_arn = "arn:aws:iam::123456789012:role/my-role")

Inside the function, paws picks up the role’s temporary credentials, so no AWS keys go into the image or the function’s settings.

Invoke deployed lambda 

invoke_lambda(
  function_name = "parity1",
  invocation_type = "RequestResponse",
  payload = list(number = 2),
  include_logs = FALSE
)

#> Lambda response payload:
#> {"parity":"even"}

Run on a schedule

schedule_lambda() runs a deployed function at a rate() or cron() schedule expression:

schedule_lambda("parity1", "rate(5 minutes)")

A scheduled run doesn’t take a payload. lambdr passes the detail of the EventBridge event to the handler as arguments, and for a schedule that is empty, so the handler is called with no arguments. parity() needs number, so each scheduled run of parity1 fails. Give the arguments of a scheduled handler defaults, e.g. parity <- function(number = 2).

Clean up

delete_lambda() removes what the steps above created: the function and its schedules, the execution role, the ECR repository with its images, the function’s log group, and the local Docker image.

delete_lambda("parity1")

# Keep the images, for example to deploy them again later
delete_lambda("parity1", keep_ecr_image = TRUE, keep_local_image = TRUE)

A role you passed with deploy_lambda(role_arn = ) is never deleted.