How much flakiness do you tolerate in end to end tests?
How much flakiness do you tolerate in end to end tests?
End to end and smoke tests give a really valuable angle on what the app is doing and can warn you about failures before they happen. However, because they're working with a live app and a live database over a live network, they can introduce a lot of flakiness. Beyond just changes to the app, different data in the environment or other issues can cause a smoke test failure.
How do you handle the inherent flakiness of testing against a live app?
When do you run smokes? On every phoenix branch? Pre-prod? Prod only?
Who fixes the issues that the smokes find?
You're viewing a single thread.
I think/hope that the wording you used was a mistake.
End to end tests do not introduce flakiness, but uncover it.
Whenever we discover flakiness, we try to fix it immediately. When there is no time for the fix (which is more than often the case) we create a ticket that vanishes in the backlog.
For a long time the company I currently work at didn't have end to end tests save unit tests for a lot of their code.
Through a push of newcomers we finally managed to add end to end tests to many more parts of the code. However, these are still not properly documented. Some end to end tests overlap and some only cover a small part of one larger functionality. That is why we often find bugs that were introduced by us, because we had no end to end tests covering those parts.
We used to run end end tests only every night on the whole product. They usually take an hour or more to complete. This takes too long to run them before each merge. However, we have them organized enough such that for sub-product A we can run the sub-product A end to end tests only before each merge where we assume that we did only touch code affecting sub-product A. In case the code changes affected some other parts of the product, the nightly tests help us out. We are doing this in my team for a long while now. But we just recently started to establish this procedure in the other teams of the company, too.
My experience with E2E testing is that the tools and methods necessary to test a complex app are flaky. Waits, checks for text or selectors and custom form field navigation all need careful balancing to make the test effective. On top of this, there is frequently a sequentiality to E2E tests that causes these failures to multiply in frequency, as you're at the mercy of not just the worst test, but the product of every test in sequence.
I agree that the tests cause less flakiness in the app itself, but I have found smokes inherently flaky in a way that unit and integration tests are not.
Okay I must admit that I do not have much experience with smoke and integration tests. We run end to end tests only and skip running the other two types entirely. They would be covered by the end to end tests anyways.
Perhaps I am lucky in that our software doesn't require us to use many waits at all. Most things are synchronous and those that are not mostly have API endpoints where the status of the process an be safely queried, i.e. a
wait(1000)and hope for the best is not necessary, but ratherdo wait(1000) until isFinished().And yes, for us it is also a mess of errors popping up when one step in a pipeline fails, where many tests rely on this single step. I don't know whether there is a way to approach this issue neatly. This is surely a chance in the market to be taken.
I'm a fan of randomizing the test order. That helps catch ordering issues early.
Also, it's usually valuable to have E2E tests all be as completely independent as possible so it's impossible for one to affect another. Have each one spin up the whole system, even though it takes longer. Use more parallelism, use dozens of VMs each running a fraction of the tests rather than trying to get the sequential time down.
Wherever possible, this is a good idea. The campsite rule - tests don't touch data they didn't bring with them - helps as well.
However, many end to end tests exist as a pipeline, especially for entities that are core to the business function of the app. Cramming all sequentiality into single tests will give you all the problems described, but in a giant single test that you need to fish out the result for.
The problem with randomising the test order is that it compromises the reproducibility of results. If there are ordering issues, then your tests will sometimes fail and sometimes pass, but will developers look at that and think "ah there must be an ordering issue" or will they think "damn these flaky tests, guess I'd better rerun the pipeline"?
End-to-end tests are basically non-deterministic state machines. Flakiness can come from any point in the test: bad tests, bad state management, conflicting tests, network hiccups, etc.
Your goal is to reduce every single point of that flakiness. Just make sure you keep track of it. Sometimes flakiness in tests is really pointing at flakiness in the product itself.
Some things that can help reduce that flakiness:
- Dedicated network
- No external dependencies
- Polling instead of static waits/sleeps
Polling is certainly useful, but at some point introducing reliability degrades effectiveness. I certainly want to know if the app is unreachable over the open internet, and I absolutely need to know if a partner's API is down.
You run E2E test before each merge. So, you don't merge very often?
How about running an integration test before each merge instead of a full fledged E2E and mocking out external dependencies (other services) during the test, then do E2E testing on a schedule like nightly?
I prefer it this way, because mocking out external dependencies cut out network instability and bugginess from dependencies. So, we can merge faster. Agree that test scenarios are overlapping, and if your E2E is very stable then it is probably not worth it, but unfortunately it's not so stable in my environment.
Luckily, our e2e tests are pretty stable. And unfortunately we are not given the time to write integration tests as you describe. The good thing would be that with these mocks we were then also be able to load test single services instead of the whole product.
We merge multiple times a day and run only those e2e tests we think are relevant. Of course, this is not optimal and it is not too rare that one of the teams merges a regression, where one team or more talented at that than the others.
You see, we have issues and we realize we have them. Our management just thinks these are not important enough to spend time on writing integration tests. I think money and developer time are two of the reasons, but the lack of feature documentation, the lack of experts for parts of the codebase (some already left for another employer), and the amount of spaghetti code and infrastructure we have are other important reasons.
Reading the 3rd paragraph and I see myself 😄. Glad that you and the team managed to add another layer of testing successfully.