refs https://github.com/TryGhost/Toolbox/issues/592
- this commit extracts the regression tests into a separate workflow
- this means they run in parallel and reduce the time we have to wait
for DB tests in general
- also fixes a test that was reliant on being run after the E2E
tests (!)
refs https://github.com/TryGhost/Toolbox/issues/528
- we're moving towards making Node 18 our recommended version, so that
involves ensuring all of CI is running Node 18
- we still have some Node 16 matrix runs to ensure compatibility
refs https://ghost.slack.com/archives/C02G9E68C/p1685446707169999
- `actions/setup-node` has a bug where it doesn't parse paths where the path is pretty-printed with colors, so we have to explicitly disable colors
- Added whitespace change inside signup-form to trigger new CI run
- this will allow us to see which set of tests are consuming the most
amount of time in CI
- in order to split apart the commands, I've had to override the
coverage thresholds for integration+regression tests in order to keep
c8 happy
- also sprinkled some more labels into the workflows to make things
clearer to read
fixes https://github.com/TryGhost/Toolbox/issues/583
- we currently upload the codecov for unit tests and e2e/admin tests
separately
- unfortunately this means that when one submits before the other, the
codecov % will be incorrect
- this leaves a big red cross on the commit until the other coverage is
uploaded
- this should fix that by upload the coverage in one step once CI is
happy
- I switched over `build` to run in production mode because we now do
`lerna run build` during the release process, but this was bundling
development assets
- this adds `build:dev` and switches the browser test to use that, so
they should use the development assets moving forwards
- Updates the prepare script in the top level to run prepare on packages, so
that packages can be built when running `yarn`
- Updates the build script in ghost/core to run build on packages, so that
packages are built before being monobundled
- Updates monobundle to be a dependency and use the new TryGhost repo, which
includes some minor fixes and improvements, such as supporting devDeps
- Updates the GitHub workflows to run the build command in the top level
directory rather than ghost/core so that other packages are built, too.
- by default, yarn will try and fetch dependencies from the network, and
fallback to local cache if the network is down
- `--prefer-offline` switches that around so we can take advantage of
local caches before defaulting to the network
- hopefully this speeds up dependency install steps in CI
fixes https://github.com/TryGhost/Toolbox/issues/532
- we should protect against failures entering `main` which could be
avoided by running a quick unit test beforehand
- this reintroduces Lerna as it supports parallelisation and `--since`,
to run linting and unit tests on packages that have changed since
upstream
- by using the same group (the workflow name), we should be able to
limit the concurrency to 1 execution, which might help with some
flaky tests we keep seeing
- now the vast majority of our flaky tests are gone, we don't need the
bigger machines
- however, browser tests seem to be a little slow, so we can try giving
them more power than the defaults
- this should also help with resource contention when we have multiple
jobs running at the same time, as we have double the numbers of
runners on the free machines
refs https://ghost.slack.com/archives/C02G9E68C/p1677753889082979
- Firefox tests have historically been flaky in CI and a real
distraction when developing
- Firefox is, unfortunately :(, not a widely used browser at only 2.9%
of global market share
- we've not had any Firefox-specific bugs that were detected in CI for
a very long time, so it doesn't add anything anyway
- given SQLite3 is only supported for development, we don't really care about
running tests on Node versions which aren't the recommended version
- this saves 2 jobs per CI run, which helps improve the health of CI in
general
- turns out our concurrency on these 8 core machines is only 10 jobs, so
everything is running really slowly
- by opening up to `linux`, we allow executions on 4, 8 and 16 core
machines with a total concurrency of 30
- we're seeing low availability for the 16 core machines and they might
be constrained as everyone jumps to the highest spec
- in theory, we don't need super fast multi-core machines to run tests,
so I'll try with 8 core ones
