End-to-End Tests
Background
Gloo Edge is built to integrate with a user's environment. By enabling users to select their preferred tools for scheduling, persistence, and security, we must ensure that we have end-to-end tests to validate these scenarios.
We support the following types of end-to-end tests:
CI
Each test suite may run on different infrastructure. Refer to the README of each test suite for more information.
What if an end-to-end test fails on a Pull Request?
Tests must account for the eventually consistent nature of Gloo Edge. Writing deterministic end-to-end tests can be challenging, and at times a test will fail in our CI pipeline non-deterministically. We refer to these failures as flakes. We have found the most common flakes occur as a result of:
- Test pollution: another test in the same suite does not clean up resources
- Eventual consistency: the test does not wait long enough for resources to be applied and processed correctly
1. Identify that it is a flake
The best way to identify that a flake occurred is to run the test locally.
First, we recommend focusing the test to ensure that no other tests are causing an impact, and following the Ginkgo recommendations for managing flaky tests.
If running the test alone does not reproduce the error, it is likely the failure is caused by test pollution. Repeat the above process, but this time move the focus one level up, so that other tests which may be creating or deleting resources are also run, repeating as necessary until the whole suite is focussed.
2. Triage the flake
If a test failure is deemed to be a flake, we take the following steps:
- Determine if there is a GitHub issue tracking the existence of that test flake
- Investigate the flake, timeboxing to a reasonable amount of time (about an hour). Flakes impact the developer experience, and we want to resolve them as soon as they are identified
- If a solution can not be determined within the timebox, create a GitHub issue to track it
- If no issue exists, create one and include the
Type: CI Test Flake
label. If an issue already exists, add a comment with the logs for the failed run. We use comment frequency as a mechanism for determining frequency of flakes
- Retry the test (specific steps can be found in a README of each test suite) and comment on the Pull Request with a link to the GitHub issue tracking the flake
Debugging Tests
TODO: We should move these debug steps to the individual test suites.
Some of the gloo tests use a listener on 127.0.0.1 rather than 0.0.0.0 and will only run on linux (e.g. fault injection).
If you’re developing on a mac, ValidateBootstrap will not run properly because it uses the envoy binary for validation mode (which only runs on linux). See rbac_jwt_test.go for an example.