Eventually

New improved module

Starting with Kotest 4.6, a new experimental module has been added which contains improved utilities for testing concurrent, asynchronous, or non-deterministic code. This module is kotest-framework-concurrency and is intended as a long term replacement for the previous module. The previous utilities are still available as part of the core framework.

Testing non-deterministic code can be hard. You might need to juggle threads, timeouts, race conditions, and the unpredictability of when events are happening.

For example, if you were testing that an asynchronous file write was completed successfully, you need to wait until the write operation has completed and flushed to disk.

Some common approaches to these problems are:

• Using callbacks which are invoked once the operation has completed. The callback can be then used to assert that the state of the system is as we expect. But not all operations provide callback functionality.

• Block the thread using Thread.sleep or suspend a function using delay, waiting for the operation to complete. The sleep threshold needs to be set high enough to be sure the operations will have completed on a fast or slow machine, and even when complete, the thread will stay blocked until the timeout has expired.

• Use a loop with a sleep and retry and a sleep and retry, but then you need to write boilerplate to track number of iterations, handle certain exceptions and fail on others, ensure the total time taken has not exceeded the max and so on.

• Use countdown latches and block threads until the latches are released by the non-determistic operation. This can work well if you are able to inject the latches in the appropriate places, but just like callbacks, it isn't always possible to have the code to be tested integrate with a latch.

As an alternative to the above solutions, kotest provides the eventually utility which solves the common use case of "I expect this code to pass after a short period of time".

Eventually does this by periodically invoking a given lambda until the timeout is eventually reached or too many iterations have passed. This is flexible and is perfect for testing nondeterministic code. Eventually can be customized in regardless to the types of exceptions to handle, how the lambda is considered a success or failure, with a listener, and so on.

API

There are two ways to use eventually. The first is simply providing a duration in either milliseconds (or using the Kotlin Duration type) followed by the code that should eventually pass without an exception being raised.

eventually(5000) { // duration in millis
  userRepository.getById(1).name shouldBe "bob"
}

The second is by providing a configuration block before the test code. This method should be used when you need to set more options than just the duration.

eventually({
  duration = 5000
  interval = 1000.fixed()
}) {
  userRepository.getById(1).name shouldBe "bob"
}

Configuration

Durations and Intervals

The duration is the total amount of time to keep trying to pass the test. The interval however allows us to specify how often the test should be attempted. So if we set duration to 5 seconds, and interval to 250 millis, then the test would be attempted at most 5000 / 250 = 20 times.

Initial Delay

Usually eventually starts executing the test block immediately, but we can add an initial delay before the first iteration using initialDelay, such as:

eventually({
  duration = 5000
  initialDelay = 1000
}) {
  userRepository.getById(1).name shouldBe "bob"
}

Retries

In addition to bounding the number of invocations by time, we can do so by iteration count. In the following example we retry the operation 10 times, or until 8 seconds has expired.

eventually({
  duration = 8000
  retries = 10
  suppressExceptions = setOf(UserNotFoundException::class)
}) {
  userRepository.getById(1).name shouldNotBe "bob"
}

Specifying the exceptions to trap

By default, eventually will ignore any AssertionError that is thrown inside the function (note, that means it won't catch Error). If you want to be more specific, you can tell eventually to ignore specific exceptions and any others will immediately fail the test.

For example, when testing that a user should exist in the database, a UserNotFoundException might be thrown if the user does not exist. We know that eventually that user will exist. But if an IOException is thrown, we don't want to keep retrying as this indicates a larger issue than simply timing.

We can do this by specifying that UserNotFoundException is an exception to suppress.

eventually({
  duration = 8000
  suppressExceptions = setOf(UserNotFoundException::class)
}) {
  userRepository.getById(1).name shouldNotBe "bob"
}

As an alternative to passing in a set of exceptions, we can provide a function which is invoked, passing in the throw exception. This function should return true if the exception should be handled, or false if the exception should bubble out.

eventually({
  duration = 8000
  suppressExceptionIf = { it is UserNotFoundException && it.username == "bob" }
}) {
  userRepository.getById(1).name shouldNotBe "bob"
}

Predicates

In addition to verifying a test case eventually runs without throwing an exception, we can also verify that the return value of the test is as expected - and if not, consider that iteration a failure and try again.

For example, here we continue to append "x" to a string until the result of the previous iteration is equal to "xxx".

var string = "x"

eventually({
  duration = 5.seconds()
  predicate = { it.result == "xxx" }
}) {
  string += "x"
  string
}

Listeners

We can attach a listener, which will be invoked on each iteration, with the state of that iteration. The state object contains the last exception, last value, iteration count and so on.

eventually({
  duration = 5.seconds()
  listener = { println("iteration ${it.times} returned ${it.result}") }
}) {
  string += "x"
  string
}

Sharing configuration

Sharing the configuration for eventually is a breeze with the EventuallyConfig data class. Suppose you have classified the operations in your system to "slow" and "fast" operations. Instead of remembering which timing values were for slow and fast we can set up some objects to share between tests and customize them per suite. This is also a perfect time to show off the listener capabilities of eventually which give you insight into the current value of the result of your producer and the state of iterations!

val slow = EventuallyConfig<ServerResponse>(
  duration = 5.minutes,
  interval = 25.milliseconds.fibonacci(),
  suppressExceptions = setOf(ServerException::class)
)

class FooTests : StringSpec({
  val logger = logger("FooTests")
  val fSlow = slow.copy(listener = { i, t -> logger.info("Current $i after {${t.times} attempts") })

  "server eventually provides a result for /foo" {
    eventually(fSlow) {
      fooApi()
    }
  }
})

class BarTests : StringSpec({
  val logger = logger("BarTests")
  val bFast = fast.copy(listener = { i, t -> logger.info("Current $i after {${t.times} attempts") })

  "server eventually provides a result for /bar" {
    eventually(bFast) {
      barApi()
    }
  }
})

Here we can see sharing of configuration can be useful to reduce duplicate code while allowing flexibility for things like custom logging per test suite for clear test logs.