Einführung in Plumber und REST

Was ist eine REST-API?

# REST API concepts in HTTP terms:
# GET /api/model/predict?x=5      -> read a prediction
# POST /api/model/train           -> create a new model
# GET /api/data/summary           -> read data summary
# DELETE /api/cache/flush         -> remove cached results

# Plumber maps R functions to these HTTP endpoints
cat('REST: stateless, resource-oriented, JSON responses')

Annotationssyntax von plumber

# plumber.R
library(plumber)

#* @get /ping
function() {
  list(status = 'ok', time = Sys.time())
}

#* @get /add
#* @param a:int First number
#* @param b:int Second number
function(a, b) {
  list(result = as.integer(a) + as.integer(b))
}

pr() – Einen Plumber-Router erstellen

library(plumber)

# Create a router from a plumber file
api <- pr('plumber.R')

# Inspect registered routes
print(api$routes)

# Alternatively, define inline without a file:
api <- pr() |>
  pr_get('/ping', function() list(status = 'ok')) |>
  pr_post('/echo', function(req) req$body)

pr_run() – Den Server starten

library(plumber)

api <- pr('plumber.R')

# Start server on localhost port 8000
# pr_run(api, host = '127.0.0.1', port = 8000)

# For Docker/remote access, bind to all interfaces
# pr_run(api, host = '0.0.0.0', port = 8000)

# View auto-generated Swagger docs in browser
# (automatically available at /docs or /__docs__/ endpoint)
cat('Swagger UI auto-generated at http://localhost:8000/__docs__/')

@get-Annotation

# plumber.R

#* Greet a user by name
#* @param name:str The name to greet
#* @get /greet
function(name = 'World') {
  list(
    message = paste('Hello,', name),
    timestamp = format(Sys.time(), '%Y-%m-%d %H:%M:%S')
  )
}
# GET /greet?name=Alice
# -> {"message":"Hello, Alice","timestamp":"2026-01-01 12:00:00"}

@post-Annotation

# plumber.R

#* Run a linear model prediction
#* @post /predict
function(req) {
  # req$body is already parsed from JSON
  input_data <- as.data.frame(req$body)

  # Run prediction with a pre-loaded model
  predictions <- predict(trained_model, newdata = input_data)

  list(
    predictions = as.numeric(predictions),
    n           = nrow(input_data)
  )
}

JSON-Serialisierer

# Default: automatic JSON serialization
#* @get /data
function() {
  list(values = 1:5, labels = c('a', 'b', 'c', 'd', 'e'))
}

# Explicit JSON serializer with options
#* @serializer json list(na = 'null', auto_unbox = TRUE)
#* @get /data_explicit
function() {
  list(value = 42, missing = NA)
}
# With auto_unbox=TRUE: {"value":42} not {"value":[42]}

HTTP-Verben – PUT, DELETE, PATCH

# plumber.R — CRUD-style endpoints

#* Update a model configuration
#* @put /config/<model_id>
function(model_id, req) {
  config <- req$body
  save_config(model_id, config)
  list(updated = model_id, config = config)
}

#* Remove cached results
#* @delete /cache/<key>
function(key) {
  cache_env <- globalenv()$cache
  rm(list = key, envir = cache_env)
  list(deleted = key)
}

Pfadparameter

# plumber.R

#* Get stats for a specific dataset
#* @param dataset_id:str The dataset identifier
#* @get /datasets/<dataset_id>/stats
function(dataset_id) {
  if (!dataset_id %in% available_datasets()) {
    stop(paste('Dataset not found:', dataset_id))
  }
  ds <- load_dataset(dataset_id)
  list(
    id    = dataset_id,
    rows  = nrow(ds),
    cols  = ncol(ds),
    names = names(ds)
  )
}

Fehlerbehandlung

# plumber.R

#* Divide two numbers safely
#* @get /divide
function(a, b, res) {
  a <- suppressWarnings(as.numeric(a))
  b <- suppressWarnings(as.numeric(b))

  if (is.na(a) || is.na(b)) {
    res$status <- 400  # Bad Request
    return(list(error = 'Both a and b must be numeric'))
  }
  if (b == 0) {
    res$status <- 422  # Unprocessable Entity
    return(list(error = 'Division by zero is not allowed'))
  }
  list(result = a / b)
}

Automatisch generierte Swagger-Dokumentation

# plumber.R with Swagger metadata

#* @apiTitle My ML Prediction API
#* @apiDescription Serves predictions from trained R models
#* @apiVersion 1.0.0

#* @tag model
#* @get /health
function() list(status = 'healthy')

#* Predict house price
#* @tag prediction
#* @param sqft:dbl Square footage
#* @param bedrooms:int Number of bedrooms
#* @get /predict
function(sqft = 1000, bedrooms = 3) {
  pred <- predict(price_model, data.frame(sqft = as.numeric(sqft),
                                          bedrooms = as.integer(bedrooms)))
  list(predicted_price = round(as.numeric(pred), 2))
}

Kurzer Test

Zusammenfassung: plumber und REST

# Complete minimal plumber API
library(plumber)

#* @apiTitle Simple Prediction API

#* Health check
#* @get /health
function() list(status = 'ok')

#* Predict mpg from weight
#* @param wt:dbl Car weight (1000 lbs)
#* @get /predict
function(wt = 3.0) {
  pred <- predict(lm(mpg ~ wt, data = mtcars),
                  newdata = data.frame(wt = as.numeric(wt)))
  list(wt = as.numeric(wt), predicted_mpg = round(pred, 2))
}

# Run:
# api <- pr('plumber.R')
# pr_run(api, port = 8000)