ci: tone down ci some more

[PaulRBerg]

We didn't tone CI down enough in #474. Some workflows are still taking a long time to run:

https://github.com/sablierhq/v2-core/actions/runs/4984025835/jobs/8921863702?pr=490

[andreivladbrg]

It looks like the time for the invariant tests has decreased significantly, but not so much for the fork and coverage. I believe this issue is caused by the foundry.

https://github.com/sablierhq/v2-core/actions/runs/4984576283

[PaulRBerg]

That's odd .. the need to communicate to the RPC might explain the fork tests, but the coverage job should have been faster because of the --no-match-path flag:

$ forge coverage --no-match-path "test/fork/**/*.sol" --report lcov

I'll investigate some more before merging this PR.

[PaulRBerg]

Following up with my investigation on why the coverage job is taking so long (>1 hour) still.

• It's not the unit tests (I ran them with --match-path)
• Not the fuzz tests (same as above)
• Not the fork tests (because --no-match-patch works)
• It's the invariant tests (I ran them with --match-path)

I proceeded to run the coverage command on my local machine, and noticed the same slow behavior. This has led me to conclude that there is a bug in Foundry, which I reported here:

foundry-rs/foundry#4952

In our PR, I updated the coverage job to include only unit and fuzz tests, like this:

$ forge coverage --match-path "test/{fuzz,unit}/**/*.sol" --report lcov

@andreivladbrg - let me know if this is good to merge now.

[PaulRBerg]

I've also investigated why the fork tests take so long (>1 hour). Here are my findings:

• They take 50% less time to run locally compared to CI, which suggests that the slow execution time in CI is in part due to using the standard GitHub runners, which have 2 cores on Linux
  • It takes ~30 minutes to run the fork tests locally, whereas in CI it's >1 hour
• The issue may be that we are getting rate limited by the RPC node provider (see my tweet about this). Some web socket connections may stall, temporarily, until the rate limit is lifted.

We might be able to speed up the CI run time by upgrading to larger GitHub runners with more cores, although this would require us to upgrade to a paid GitHub plan (e.g. Team). Similarly, we might get rate limited less often if we upgraded to a higher plan on Alchemy and/ or ChainNodes.

Anyway. In the meantime, I will simply lower the number of fuzz runs in the test-fork job. We get enough test coverage via the ci-deep workflow.

Update: it is also unclear how Foundry handles RPC retries. See foundry-rs/foundry#4958.

[PaulRBerg]

I will simply lower the number of fuzz runs in the test-fork job

Seems like this approach is working: the fork tests take ~15 minutes to run now:

https://github.com/sablierhq/v2-core/actions/runs/5002102460/jobs/8962236042?pr=492

Merging.