Retry-ability

A core feature of Cypress that assists with testing dynamic web applications is retry-ability. Like a good transmission in a car, it usually works without you noticing it. But understanding how it works will help you write faster tests with fewer run-time surprises.

Commands vs assertions

There are two types of methods you can call in your Cypress tests: commands and assertions. For example, there are 6 commands and 2 assertions in the test below.

it('creates 2 items', () => {
  cy.visit('/') // command
  cy.focused() // command
    .should('have.class', 'new-todo') // assertion

  cy.get('.new-todo') // command
    .type('todo A{enter}') // command
    .type('todo B{enter}') // command

  cy.get('.todo-list li') // command
    .should('have.length', 2) // assertion
})

The Command Log shows both commands and assertions with passing assertions showing in green.

ommands and assertions

Let's look at the last command and assertion pair:

cy.get('.todo-list li') // command
  .should('have.length', 2) // assertion

Because nothing is synchronous in modern web applications, Cypress can't query all the DOM elements with the class todo-list and check if there are only two of them. There are many examples of why this would not work well.

  • What if the application has not updated the DOM by the time these commands run?
  • What if the application is waiting for its back end to respond before populating the DOM element?
  • What if the application does some intensive computation before showing the results in the DOM?

Thus the Cypress cy.get command has to be smarter and expect the application to potentially update. The cy.get() queries the application's DOM, finds the elements that match the selector, and then tries the assertion that follows it (in our case should('have.length', 2)) against the list of found elements.

  • ✅ If the assertion that follows the cy.get() command passes, then the command finishes successfully.
  • 🚨 If the assertion that follows the cy.get() command fails, then the cy.get() command will requery the application's DOM again. Then Cypress will try the assertion against the elements yielded from cy.get(). If the assertion still fails, cy.get() will try requerying the DOM again, and so on until the cy.get() command timeout is reached.

The retry-ability allows the tests to complete each command as soon as the assertion passes, without hard-coding waits. If your application takes a few milliseconds or even seconds to render each DOM element - no big deal, the test does not have to change at all. For example, let's introduce an artificial delay of 3 seconds when refreshing the application's UI below in an example TodoMVC model code:

app.TodoModel.prototype.addTodo = function (title) {
  this.todos = this.todos.concat({
    id: Utils.uuid(),
    title: title,
    completed: false,
  })

  // let's trigger the UI to render after 3 seconds
  setTimeout(() => {
    this.inform()
  }, 3000)
}

My test still passes! The last cy.get('.todo-list') and the assertion should('have.length', 2) are clearly showing the spinning indicators, meaning Cypress is requerying for them.

Retrying finding 2 items

Within a few milliseconds after the DOM updates, cy.get() finds two elements and the should('have.length', 2) assertion passes

Multiple assertions

A single command followed by multiple assertions retries each one of them -- in order. Thus when the first assertion passes, the command will be retried with first and second assertion. When the first and second assertion pass, then the command will be retried with the first, second, and third assertion, and so on.

For example, the following test has .should() and .and() assertions. .and() is an alias of the .should() command, so the second assertion is really a custom callback assertion in the form of the .should(cb) function with 2 expect assertions inside of it.

cy.get('.todo-list li') // command
  .should('have.length', 2) // assertion
  .and(($li) => {
    // 2 more assertions
    expect($li.get(0).textContent, 'first item').to.equal('todo a')
    expect($li.get(1).textContent, 'second item').to.equal('todo B')
  })

Because the second assertion expect($li.get(0).textContent, 'first item').to.equal('todo a') fails, the third assertion is never reached. The command fails after timing out, and the Command Log correctly shows that the first encountered assertion should('have.length', 2) passed, but the second assertion and the command itself failed.

Retrying multiple assertions

Not every command is retried

Cypress only retries commands that query the DOM: cy.get(), .find(), .contains(), etc. You can check if a particular command is retried by looking at the "Assertions" section in its API documentation. For example, "Assertions section" of .first() tells us that the command is retried until all assertions that follow it are passing.

  • .first() will automatically retry until the element(s) exist in the DOM
  • .first() will automatically retry until all chained assertions have passed

Why are some commands NOT retried?

Commands are not retried when they could potentially change the state of the application under test. For example, Cypress will not retry the .click() command, because it could change something in the application.

Built-in assertions

Often a Cypress command has built-in assertions that will cause the command to be retried. For example, the .eq() command will be retried even without any attached assertions until it finds an element with the given index in the previously yielded list of elements.

cy.get('.todo-list li') // command
  .should('have.length', 2) // assertion
  .eq(3) // command
Retrying built-in assertion

Some commands that cannot be retried still have built-in waiting. For example, as described in the "Assertions" section of .click(), the click() command waits to click until the element becomes actionable.

Cypress tries to act like a human user would using the browser.

  • Can a user click on the element?
  • Is the element invisible?
  • Is the element behind another element?
  • Does the element have the disabled attribute?

The .click() command will automatically wait until multiple built-in assertion checks like these pass, and then it will attempt to click just once.

Timeouts

By default each command that retries, does so for up to 4 seconds - the defaultCommandTimeout setting.

Increase time to retry

You can change this timeout for all commands. See Configuration: Overriding Options for examples of overriding this option.

For example, to set the default command timeout to 10 seconds via the command line:

cypress run --config defaultCommandTimeout=10000

We do not recommend changing the command timeout globally. Instead, pass the individual command's { timeout: ms } option to retry for a different period of time. For example:

// we've modified the timeout which affects default + added assertions
cy.get('.mobile-nav', { timeout: 10000 })
  .should('be.visible')
  .and('contain', 'Home')

Cypress will retry for up to 10 seconds to find a visible element of class mobile-nav with text containing "Home". For more examples, read the Timeouts section in the "Introduction to Cypress" guide.

Disable retry

Overriding the timeout to 0 will essentially disable retrying the command, since it will spend 0 milliseconds retrying.

// check synchronously that the element does not exist (no retry)
// for example just after a server-side render
cy.get('#ssr-error', { timeout: 0 }).should('not.exist')

Only the last command is retried

Here is a short test that demonstrates some flake.

it('adds two items', () => {
  cy.visit('/')

  cy.get('.new-todo').type('todo A{enter}')
  cy.get('.todo-list li').find('label').should('contain', 'todo A')

  cy.get('.new-todo').type('todo B{enter}')
  cy.get('.todo-list li').find('label').should('contain', 'todo B')
})

The test passes in Cypress without a hitch.

Test passes

But sometimes the test fails - not usually locally, no - it almost always fails on our continuous integration server. When the test fails, the recorded video and screenshots are NOT showing any obvious problems! Here is the failing test video:

Test fails

The problem looks weird - I can clearly see the label "todo B" present in the list, so why isn't Cypress finding it? What is going on?

Remember the delay we introduced into our application code that causes the test to time out? We added a 100ms delay before the UI rerenders itself.

app.TodoModel.prototype.addTodo = function (title) {
  this.todos = this.todos.concat({
    id: Utils.uuid(),
    title: title,
    completed: false,
  })

  setTimeout(() => {
    this.inform()
  }, 100)
}

This delay could be the source of our flaky tests when the application is running on our CI server. Here is how to see the source of the problem. In the Command Log, hover over each command to see which elements Cypress found at each step.

In the failing test, the first label was indeed found correctly:

First item label

Hover over the second "FIND label" command - something is wrong here. It found the first label, then kept requerying to find the text "todo B", but the first item always remains "todo A".

Second item label

Hmm, weird, why is Cypress only looking at the first item? Let's hover over the "GET .todo-list li" command to inspect what that command found. Ohh, interesting - there was only one item at that moment.

Second get li

During the test, the cy.get('.todo-list li') command quickly found the rendered <li> item - and that item was the first and only "todo A" item. Our application was waiting 100ms before appending the second item "todo B" to the list. By the time the second item was added, Cypress had already "moved on", working only with the first <li> element. It only searched for <label> inside the first <li> element, completely ignoring the newly created 2nd item.

To confirm this, let's remove the artificial delay to see what's happening in the passing test.

Two items

When the web application runs without the delay, it gets its items into the DOM before the Cypress command cy.get('.todo-list li') runs. After the cy.get() returns 2 items, the .find() command just has to find the right label. Great.

Now that we understand the real reason behind the flaky test, we need to think about why the default retry-ability has not helped us in this situation. Why hasn't Cypress found the 2 <li> elements after the second one was added?

For a variety of implementation reasons, Cypress commands only retry the last command before the assertion. In our test:

cy.get('.new-todo').type('todo B{enter}')
cy.get('.todo-list li') // queries immediately, finds 1 <li>
  .find('label') // retried, retried, retried with 1 <li>
  .should('contain', 'todo B') // never succeeds with only 1st <li>

Use retry-ability correctly

Luckily, once we understand how retry-ability works and how only the last command is used for assertion retries, we can fix this test for good.

Merging queries

The first solution we recommend is to avoid unnecessarily splitting commands that query elements. Instead of cy.get('.todo-list li').find('label') we can combine two separate queries into one - forcing the combined query to be retried.

it('adds two items', () => {
  cy.visit('/')

  cy.get('.new-todo').type('todo A{enter}')
  cy.get('.todo-list li label') // 1 query command
    .should('contain', 'todo A') // assertion

  cy.get('.new-todo').type('todo B{enter}')
  cy.get('.todo-list li label') // 1 query command
    .should('contain', 'todo B') // assertion
})

To show the retries, I increased the application's artificial delay to 500ms. The test now always passes because the entire selector is retried. It finds 2 list elements when the second "todo B" is added to the DOM.

Combined selector

Similarly, when working with deeply nested JavaScript properties using the .its() command, try not to split it across multiple calls. Instead, combine property names into a single call using the . separator:

// 🛑 not recommended
// only the last "its" will be retried
cy.window()
  .its('app') // runs once
  .its('model') // runs once
  .its('todos') // retried
  .should('have.length', 2)

// ✅ recommended
cy.window()
  .its('app.model.todos') // retried
  .should('have.length', 2)

See the Set flag to start tests blog for the full example.

Alternate commands and assertions

There is another way to fix our flaky test. Whenever you write a longer test, we recommend alternating commands with assertions. In this case, I will add an assertion after the cy.get() command, but before the .find() command.

it('adds two items', () => {
  cy.visit('/')

  cy.get('.new-todo').type('todo A{enter}')
  cy.get('.todo-list li') // command
    .should('have.length', 1) // assertion
    .find('label') // command
    .should('contain', 'todo A') // assertion

  cy.get('.new-todo').type('todo B{enter}')
  cy.get('.todo-list li') // command
    .should('have.length', 2) // assertion
    .find('label') // command
    .should('contain', 'todo B') // assertion
})
Passing test

The test passes, because the second cy.get('.todo-list li') is retried with its own assertion now .should('have.length', 2). Only after successfully finding two <li> elements, the command .find('label') and its assertion starts, and by now, the item with the correct "todo B" label has been correctly queried.

Use .should() with a callback

If you have to use commands that cannot be retried, but need to retry the entire chain, consider rewriting the commands into a single .should(callbackFn) chained off the very first retry-able command.

Below is an example where the number value is set after a delay:

<div class="random-number-example">
  Random number: <span id="random-number">🎁</span>
</div>
<script>
  const el = document.getElementById('random-number')
  setTimeout(() => {
    el.innerText = Math.floor(Math.random() * 10 + 1)
  }, 1500)
</script>
Random number

Incorrectly waiting for values

You may want to write a test like below, to test that the number is between 1 and 10, although this will not work as intended. The test yields the following values, noted in the comments, before failing.

// WRONG: this test will not work as intended
cy.get('#random-number') // <div>🎁</div>
  .invoke('text') // "🎁"
  .then(parseFloat) // NaN
  .should('be.gte', 1) // fails
  .and('be.lte', 10) // never evaluates

Unfortunately, the .then() command is not retried. Thus the test only runs the entire chain once before failing.

First attempt at writing the test

Correctly waiting for values

We need to retry getting the element, invoking the text() method, calling the parseFloat function and running the gte and lte assertions. We can achieve this using the .should(callbackFn).

cy.get('#random-number').should(($div) => {
  // all the code inside here will retry
  // until it passes or times out
  const n = parseFloat($div.text())

  expect(n).to.be.gte(1).and.be.lte(10)
})

The above test retries getting the element and invoking the text of the element to get the number. When the number is finally set in the application, then the gte and lte assertions pass and the test passes.

Random number using callback

Use aliases

When using cy.stub() or cy.spy() to test application's code, a good practice is to give it an alias and use the cy.get('@alias').should('...') assertion to retry.

For example, when confirming that the button component invokes the click prop testing with the @cypress/react plugin, the following test might or might not work:

Incorrectly checking if the stub was called

const Clicker = ({ click }) => (
  <div>
    <button onClick={click}>Click me</button>
  </div>
)

it('calls the click prop twice', () => {
  const onClick = cy.stub()
  // "mount" function comes from
  // https://github.com/cypress-io/cypress/tree/master/npm/react
  mount(<Clicker click={onClick} />)
  cy.get('button')
    .click()
    .click()
    .then(() => {
      // works in this case, but not recommended
      // because .then() does not retry
      expect(onClick).to.be.calledTwice
    })
})

The above example will fail if the component calls the click prop after a delay.

const Clicker = ({ click }) => (
  <div>
    <button onClick={() => setTimeout(click, 500)}>Click me</button>
  </div>
)
Expect fails the test without waiting for the delayed stub

The test finishes before the component calls the click prop twice, and without retrying the assertion expect(onClick).to.be.calledTwice.

Correctly waiting for the stub to be called

We recommend aliasing the stub using the .as command and using cy.get('@alias').should(...) assertions.

it('calls the click prop', () => {
  const onClick = cy.stub().as('clicker')
  // "mount" function comes from
  // https://github.com/cypress-io/cypress/tree/master/npm/react
  mount(<Clicker click={onClick} />)
  cy.get('button').click().click()

  // good practice 💡
  // auto-retry the stub until it was called twice
  cy.get('@clicker').should('have.been.calledTwice')
})
Retrying the assertions using a stub alias

Watch the short video below to see this example in action

See also