Tidy Tuesday dataset Lambda

library(r2lambda)
library(jsonlite)
library(magrittr)

Use `{r2lambda}` to download Tidytuesday dataset

In this exercize, we’ll create an AWS Lambda function that downloads the tidytuesday data set for the most recent Tuesday (or most recent Tuesday from a date of interest).

Runtime function

The first step is to write the runtime function. This is the function that will be executed when we invoke the Lambda function after it has been deployed. To download the Tidytuesday data set, we will use the tidytuesdayR package. In the runtime script, we define a function called tidytyesday_lambda that takes one optional argument date. If date is omitted, the function returns the data set(s) for the most recent Tuesday, otherwise, it looks up the most recent Tuesday from a date of interest and returns the corresponding data set(s).

library(tidytuesdayR)

tidytuesday_lambda <- function(date = NULL) {
  if (is.null(date))
    date <- Sys.Date()
  
  most_recent_tuesday <- tidytuesdayR::last_tuesday(date = date)
  tt_data <- tidytuesdayR::tt_load(x = most_recent_tuesday)
  data_names <- names(tt_data)
  data_list <- lapply(data_names, function(x) tt_data[[x]])
  return(data_list)
}

tidytuesday_lambda("2022-02-02")

R script to build the lambda

To build the lambda image, we need an R script that sources any required code, loads any needed libraries, defines a runtime function, and ends with a call to lambdr::start_lambda(). The runtime function does not have to be defined in this file. We could, for example, source another script, or load a package and set a loaded function as the runtime function in the subsequent call to r2lambda::build_lambda (see below). We save this script to a file and record the path:

r_code <- "
  library(tidytuesdayR)

  tidytuesday_lambda <- function(date = NULL) {
    if (is.null(date))
      date <- Sys.Date()
    
    most_recent_tuesday <- tidytuesdayR::last_tuesday(date = date)
    tt_data <- tidytuesdayR::tt_load(x = most_recent_tuesday)
    data_names <- names(tt_data)
    data_list <- lapply(data_names, function(x) tt_data[[x]])
    return(data_list)
  }
  
  lambdr::start_lambda()
"

tmpfile <- tempfile(pattern = "ttlambda_", fileext = ".R")
write(x = r_code, file = tmpfile)

Build, test, and deploy the lambda function

1. Build

We set the runtime_function argument to the name of the function we wish the docker container to run when invoked. In this case, this is tidytuesday_lambda. This adds a CMD instruction to the Dockerfile
We set the runtime_path argument to the path we stored the script defining our runtime function.
We set the dependencies argument to c("tidytuesdayR")because we need to have the tidytuesdayR package installed within the docker container if we are to download the dataset. This steps adds a RUN instruction to the Dockerfile that calls install.packages to install tidytuesdayR from CRAN.
Finally, the tag argument sets the name of our Lambda function which we’ll use later to test and invoke the function. The tag argument also becomes the name of the folder that r2lambda will create to build the image. This folder will have two files, Dockerfile and runtime.R. runtime.R is our script from runtime_path, renamed before it is copied in the docker image with a COPY instruction.

runtime_function <- "tidytuesday_lambda"
runtime_path <- tmpfile
dependencies <- "tidytuesdayR"

r2lambda::build_lambda(
  tag = "tidytuesday3",
  runtime_function = runtime_function,
  runtime_path = runtime_path,
  dependencies = dependencies
)

2. Test

To make sure our Lambda docker container works as intended, we start it locally, and invoke it to test the response. The response is a list of three elements:

response <- r2lambda::test_lambda(tag = "tidytuesday3", payload = list(date = Sys.Date()))

status, should be 0 if the test worked,
stdout, the standard output stream of the invocation, and
stderr, the standard error stream of the invocation

stdout and stderr are raw vectors that we need to parse, for example:

rawToChar(response$stdout)

If the stdout slot of the response returns the correct output of our function, we are good to deploy to AWS.

3. Deploy

The deploy step is simple, in that all we need to do is specify the name (tag) of the Lambda function we wish to push to AWS ECR. The deploy_lambda function also accepts ..., which are named arguments ultimately passed onto paws.compute:::lambda_create_function. This is the function that calls the Lambda API. To see all available arguments run ?paws.compute:::lambda_create_function.

The most important arguments are probably Timeout and MemorySize, which set the time our function will be allowed to run and the amount of memory it will have available. In many cases it will make sense to increase the defaults of 3 seconds and 128 mb.

r2lambda::deploy_lambda(tag = "tidytuesday3", Timeout = 30)

4. Invoke

If all goes well, our function should now be available on the cloud awaiting requests. We can invoke it from R using invoke_lambda. The arguments are:

function_name – the name of the function
invocation_type – typically RequestResponse
include_log – whether to print the logs of the run on the console
payload – a named list with arguments sent to the runtime_function. In this case, the runtime function, tidytuesday_lambda has a single argument date, so the corresponding list is list(date = Sys.Date()). As our function can be called without any argument, we can also send and empty list as the payload.

response <- r2lambda::invoke_lambda(
  function_name = "tidytuesday3",
  invocation_type = "RequestResponse",
  payload = list(),
  include_logs = TRUE
)

Just like in the local test, the response payload comes as a raw vector that needs to be parsed into a data.frame:

tidytuesday_dataset <- response$Payload |> 
  rawToChar() |> 
  jsonlite::fromJSON(simplifyDataFrame = TRUE)

tidytuesday_dataset[[1]][1:5, 1:5]

Use {r2lambda} to download Tidytuesday dataset