How do I use webfakes in my package?

First, you need to add webfakes to the DESCRIPTION file of your package. Use the Suggests field, as webfakes is only needed for testing:


Then, unless the URL to the web service is an argument of your package functions, you might need to tweak your package code slightly to make sure every call to a real web service can be targeted at another URL instead (of a fake app). See next subsection.

Last but not least, you need to decide if you want a single web app for all your test cases. The alternative is to use different apps for some or all test files. Occasionally you may want to use a special app for a single test case. Each app runs in a new subprocess, and it takes typically about 100-400ms to start.

See the sections later on on writing tests with a single app or multiple apps.

How do I make my app connect to webfakes when the tests are running?

In the typical scenario, you want your package to connect to the test app only when running the tests. If the URL to the web service is not an argument of the functions, one way to achieve this is to allow specifying the web server URL(s) via environment variables. E.g. when writing a GitHub API client, your package can check use GITHUB_URL environment variable.


service_url <- function() {
    Sys.getenv("GITHUB_URL", "")

# rest of the package code

foobar <- function() {

When this is not set, the package connects to the proper GitHub API. When testing, you can point it to your test app.

new_app_process() helps you setting up temporary environment variables. These are active while the process is running, and they are removed or reset in $stop(). For example:

In $local_env() environment variables, webfakes replaces {url} with the actual app URL. This is needed by default, because the web server process starts up only later, so the URL is not known yet.

http <- webfakes::local_app_process(webfakes::httpbin_app(), start = TRUE)
#> Setting deferred event(s) on global environment.
#>   * Execute (and clear) with `withr::deferred_run()`.
#>   * Clear (without executing) with `withr::deferred_clear()`.
http$local_env(list(GITHUB_API = "{url}"))
#> [1] ""
#> [1] ""

How can I write my own app?

You create a new app with new_app(). This returns an object with methods to add middleware and API endpoints to it. For example, a simple app that returns the current time in JSON would look like this:

time <- webfakes::new_app()
time$get("/time", function(req, res) {
  res$send_json(list(time = format(Sys.time())), auto_unbox = TRUE)

Now you can start this app on a random port using web$listen(). Alternatively, you can start it in a subprocess with new_app_process().

web <- webfakes::new_app_process(time)
#> [1] ""

Use web$url() to query the URL of the app. For example:

url <- web$url("/time")
#> $time
#> [1] "2021-04-05 11:12:50"

web$stop() stops the app and the subprocess as well:

#> [1] "not running"

local_app_process() is similar to new_app_process(), but it stops the server process at the end of the calling block. This means that the process is automatically cleaned up at the end of a test_that() block or at the end of the test file.

You can create your app at the beginning of your test file. Or, if you want to use the same app in multiple test files, use a testthat helper file. Sometimes it useful if your users can create and use your test app, for example to create reproducible examples. You can include a (possibly internal) function in your package, that creates the app.

See ?new_app(), ?new_app_process() and ?local_app_process for more details.

How do I use httpbin_app() (or another app) with testthat?

You can use testthat’s setup files. You start the app in a setup file and also register a teardown expression for it. local_app_process() can do both in one go. Your tests/testthat/setup-http.R may look like this:

http <- webfakes::local_app_process(
  .local_envir = testthat::teardown_env()

(Before testthat 3.0.0, you had to write the teardown expression in a tests/testthat/teardown-http.R file. That still works, but a single setup file is considered to be better practice, see this testthat vignette.)

In the test cases you can query the http app process to get the URLs you need to connect to:

test_that("fails on 404", {
  url <- http$url("/status/404")
  response <- httr::GET(url)
    class = "http_404"
#> Test passed 😸

When writing your tests interactively, you may create a http app process in the global environment, for convenience. You can source() your setup-http.R file for this. Alternatively, you can start the app process in a helper file. See “How do I start the app when writing the tests?” just below.

How do I start the app when writing the tests?

It is convenient to start the webfakes server process(es) when working on the tests interactively, e.g. when using devtools::load_all(). With local_app_process() in the testthat setup*.R file this is not automatic, because devtools::load_all() does not run these files. So you would need to source the setup*.R files manually, which is error prone.

One solution is to create server processes in the testthat helper*.R files. load_all() executes the helper files by default. So instead of using a setup file, you can simply do this in the helper-http.R file:

If the app process is created in the helper file, then it is ready use after load_all(), and (by default) the actual process will be started at the first $url() or $get_port() call. You can also start it manually with $start().

Processes created in helper files are not cleaned up automatically at the end of the test suite, unless you clean them up by registering a $stop() call in a setup file, like this:

withr::defer(httpbin$stop(), testthat::teardown_env())

In practice this is not necessary, because R CMD check runs the tests in a separate process, and when that finishes, the webfakes processes are cleaned up as well.

When running devtools::test(), testthat::test_local() or another testthat function to run (part of) the test suite in the current session, the helper*.R files are (re)loaded first. This will terminate the currently running app processes, if any, and create new app process objects. Should the test suite auto-start some of the test processes from helper*.R, these will not be cleaned up at the end of the test suite, but only at the next load_all() or test() call, or at the end of the R session. This lets you run your test code interactively, either via test() or manually, without thinking too much about the webfakes processes.

Can I have an app for a single testthat test file?

To run a web app for a single test file, start it with new_app_process() at the beginning of the file, and register its cleanup using withr::defer(). Even simpler, use local_app_process() which is the same as new_app_process() but it automatically stops the web server process, at the end of the test file:

app <- webfakes::new_app()
app$get("/hello/:user", function(req, res) {
  res$send(paste0("Hello ", req$params$user, "!"))
web <- webfakes::local_app_process(app)

Then in the test cases, use web$url() to get the URL to connect to.

test_that("can use hello API", {
  url <- web$url("/hello/Gabor")
  expect_equal(httr::content(httr::GET(url)), "Hello Gabor!")
#> No encoding supplied: defaulting to UTF-8.
#> Test passed 😸

Can I use an app for a single testthat test?

Sure. For this you need to create the app process within the testthat::test_that() test case. local_app_process() automatically cleans it up at the end of the block. It goes like this:

test_that("query works", {
  app <- webfakes::new_app()
  app$get("/hello", function(req, res) res$send("hello there"))
  web <- webfakes::local_app_process(app)

  echo <- httr::content(httr::GET(web$url("/hello")))
  expect_equal(echo, "hello there")
#> No encoding supplied: defaulting to UTF-8.
#> Test passed 🎉

How do I test a sequence of requests?

To test a sequence of requests, the app needs state information that is kept between requests. app$locals is an environment that belongs to the app, and it can be used to record information and then retrieve it in future requests. You could store anything in app$locals, something simple like a counter variable, something fancier like a sqlite database. You can add something to app$locals via methods or directly after creating the app.

store <- webfakes::new_app()
store$locals$packages <- list("webfakes")
#> [1] "packages"
#> [[1]]
#> [1] "webfakes"

E.g. here is an end point that fails three times, then succeeds once, fails again three times, etc.

Note that the counter created by the code below starts at 0, not 1.

flaky <- webfakes::new_app()
flaky$get("/unstable", function(req, res) {
  if (identical(res$app$locals$counter, 3L)) {
    res$app$locals$counter <- NULL
    res$send_json(object = list(result = "ok"))
  } else {
    res$app$locals$counter <- c(res$app$locals$counter, 0L)[[1]] + 1L

Let’s run this app in another process and connect to it:

pr <- webfakes::new_app_process(flaky)
url <- pr$url("/unstable")
httr::RETRY("GET", url, times = 4)
#> Request failed [401]. Retrying in 1 seconds...
#> Request failed [401]. Retrying in 3.7 seconds...
#> Request failed [401]. Retrying in 3 seconds...
#> Response []
#>   Date: 2021-04-05 11:13
#>   Status: 200
#>   Content-Type: application/json
#>   Size: 17 B

Another example where we send information to an app and then retrieve it. On a POST request we store the name query parameter in app$locals$packages, which can be queried with a GET request.

store <- webfakes::new_app()
# Initial "data" for the app
store$locals$packages <- list("webfakes")
# Get method
store$get("/packages", function(req, res) {
  res$send_json(res$app$locals$packages, auto_unbox = TRUE)
# Post method, store information from the query
store$post("/packages", function(req, res) {
  res$app$locals$packages <- c(res$app$locals$packages, req$query$name)
  res$send_json(res$app$locals$packages, auto_unbox = TRUE)

Now we start the app in a subprocess, and run a GET query against it.

web <- webfakes::local_app_process(store, start = TRUE)
# Get current information
get_packages <- function() {
        path = "packages"
#> [[1]]
#> [1] "webfakes"

Let’s POST some new information.

post_package <- function(name) {
    path = "packages",
    query = list(name = name)
#> Response []
#>   Date: 2021-04-05 11:13
#>   Status: 200
#>   Content-Type: application/json
#>   Size: 18 B
# Get current information
#> [[1]]
#> [1] "webfakes"
#> [[2]]
#> [1] "vcr"

#> Response []
#>   Date: 2021-04-05 11:13
#>   Status: 200
#>   Content-Type: application/json
#>   Size: 29 B
# Get current information
#> [[1]]
#> [1] "webfakes"
#> [[2]]
#> [1] "vcr"
#> [[3]]
#> [1] "httptest"

Stop the app process:


How can I debug an app?

To debug an app, it is best to run it in the main R process, i.e. not via new_app_process(). You can add breakpoints, or browser() calls to your handler functions, and then invoke your app from another process. You might find the curl command line tool to send HTTP requests to the app, or you can just use another R process. Here is an example. We will simply print the incoming request object to the screen now. For a real debugging session you probably want to place a browser() command there.

app <- webfakes::new_app()
app$get("/debug", function(req, res) {
  res$send("Got your back")

Now start the app on port 3000:

app$listen(port = 3000)
#> Running webfakes web app on port 3000

Connect to the app from another R or curl process:

curl -v
#> *   Trying
#> * TCP_NODELAY set
#> * Connected to ( port 3000 (#0)
#> > GET /debug HTTP/1.1
#> > Host:
#> > User-Agent: curl/7.54.0
#> > Accept: */*
#> >
#> < HTTP/1.1 200 OK
#> < Content-Type: text/plain
#> < Content-Length: 13
#> <
#> * Connection #0 to host left intact
#> Got your back

Your main R session will print the incoming request:

#> <webfakes_request>
#> method:
#>   get
#> url:
#> client:
#> query:
#> headers:
#>   Host:
#>   User-Agent: curl/7.54.0
#>   Accept: */*
#> fields and methods:
#>   app                    # the webfakes_app the request belongs to
#>   headers                # HTTP request headers
#>   hostname               # server hostname, the Host header
#>   method                 # HTTP method of request (lowercase)
#>   path                   # server path
#>   protocol               # http or https
#>   query_string           # raw query string without '?'
#>   query                  # named list of query parameters
#>   remote_addr            # IP address of the client
#>   url                    # full URL of the request
#>   get_header(field)      # get a request header
#>  # see ?webfakes_request for details

Press CTRL+C or ESC to interrupt the app in the main session.

Can I test asynchronous or parallel HTTP requests?

R is single threaded and a webfakes app runs an R interpreter, so it cannot process multiple requests at the same time. The web server itself runs in a separate thread, and it can also process each request in a separate thread, but at any time only one request can use the R interpreter.

This is important, because sometimes test requests may take longer to process. For example the /delay/:secs end point of httpbin_app() wait for the specified number of seconds before responding, to simulate a slow web server. If this wait is implemented via the standard Sys.sleep() R function, then no other requests can be processed until the sleep is over. To avoid this, webfakes can put the waiting request on hold, return from the R interpreter, and respond to other incoming requests. Indeed, the /delay/ end point is implemented using this feature.

However, the request thread of the web server is still busy while on hold, so to take advantage of this, you need to allow multiple threads. The num_threads argument of the $listen() method of webfakes_app lets you specify the number of request threads the web server will use. Similarly, the num_threads argument of local_app_process() lets you modify the number of threads.

When testing asynchronous or parallel code, that might invoke multiple, possibly delayed requests, it is best to increase the number of threads. The code below calls the same API request concurrently, three times. Each request takes 1 second to answer, but if the web server has more than three threads, together they’ll still take about 1 second.

web <- webfakes::local_app_process(
  opts = webfakes::server_opts(num_threads = 3)
test_that("", {
  url <- web$url("/delay/1")
  p <- curl::new_pool()
  handles <- replicate(3, curl::new_handle(url = url))
  resps <- list()
  for (handle in handles) {
      done = function(x) message("one is done"),
      fail = stop,
      pool = p
  st <- system.time(curl::multi_run(timeout = 5, pool = p))
  expect_true(st[["elapsed"]] < 3.0)
#> one is done
#> one is done
#> one is done
#>    user  system elapsed 
#>   0.002   0.001   1.680 
#> Test passed 🥇

How to make sure that my code works with the real API?

Indeed, if you use webfakes for your test cases, then they never touch the real web server. As you might suspect, this is not ideal, especially when you do not control the server. The web service might change their API, and your test cases will fail to warn you.

One practical solution is to write (at least some) flexible tests, that can run against a local fake webserver, or a real one, and you have a quick switch to change their behavior. I have found that environment variables work great for this.

E.g. if the FAKE_HTTP_TESTS environment variable is not set, the tests run with the real web server, otherwise they use a fake one. Another solution, that works best is the HTTP requests are in the downstream package code, is to introduce one environment variable for each API you need to connect to. These might be set to the real API servers, or to the fake ones.

Once you have some tests that can use both kinds or servers, you can set up your continuous integration (CI) framework, to run the tests agains the real server (say) once a day. This special CI run makes sure that your code works well with the real API. You can run all the other tests, locally and in the CI, against the fake local web server.

See the question on how webfakes helps you setting environment variables that point to your local server.

How do I simulate a slow internet connection?

You need to use the throttle server option when you start your web app. This means that you can run the very same app with different connection speed. This is how it goes:

slow <- new_app_process(
  opts = server_opts(throttle = 100)
resp <- curl::curl_fetch_memory(slow$url("/bytes/200"))
#>      redirect    namelookup       connect   pretransfer starttransfer 
#>      0.000000      0.000134      0.000326      0.000370      0.010574 
#>         total 
#>      2.016041

throttle gives the number of bytes per second, so downloading 200 random bytes from the fake app will take about 2 seconds.