julianmrodri / hhreset-ci-issue

Hardhat Reset causes failure in CI (Github Actions)

Hardhat Reset issue

In this repo, any test that performs a hardhat_reset action ends in failure in CI (and when running local coverage) with the following error:

<--- Last few GCs --->
[2095:0x62c1840]    23100 ms: Scavenge (reduce) 2046.7 (2082.8) -> 2046.7 (2083.8) MB, 8.1 / 0.0 ms  (average mu = 0.298, current mu = 0.159) allocation failure 
[2095:0x62c1840]    23947 ms: Mark-sweep (reduce) 2047.6 (2083.8) -> 2047.5 (2084.5) MB, 786.2 / 0.0 ms  (+ 309.4 ms in 98 steps since start of marking, biggest step 14.6 ms, walltime since start of marking 1154 ms) (average mu = 0.221, current mu = 0.085
<--- JS stacktrace --->
FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory
....

This happens even if no contract is used at all and if no contract is deployed during the tests. It is important to remark that locally the test passes successfuly, however in github actions it breaks with the error shown above.

How to run tests locally - Successful test

1. Install dependencies

yarn 

2. Compile contracts

yarn compile

3. Run tests

yarn test

The test will succeed locally:

  Test Breaks with Reset
    Test
      ✔ Cannot reach in CI due to failure in reset

How to reproduce issue in CI - Failed test

The error occurs with any PR with any change to force the CI to run. The CI will fail with the error shown above, at the time the hardhat reset is performed.

<--- Last few GCs --->
[2095:0x62c1840]    23100 ms: Scavenge (reduce) 2046.7 (2082.8) -> 2046.7 (2083.8) MB, 8.1 / 0.0 ms  (average mu = 0.298, current mu = 0.159) allocation failure 
[2095:0x62c1840]    23947 ms: Mark-sweep (reduce) 2047.6 (2083.8) -> 2047.5 (2084.5) MB, 786.2 / 0.0 ms  (+ 309.4 ms in 98 steps since start of marking, biggest step 14.6 ms, walltime since start of marking 1154 ms) (average mu = 0.221, current mu = 0.085
<--- JS stacktrace --->
FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory

To reproduce simply look at the jobs for the open PR#16 (FAIL CI) and see how it fails in Github Actions.

Important Hint

If you remove contracts/fuzz/RTokenDiffTesting.sol from the repo then the test passes, which indicates this could be the issue. The interesting thing is that this contract is not used anywhere in the code but for some reason seems to break something in the HH network. You can see how it works in PR#14 (SUCCESSFUL CI) where the CI completed successfully.

Also happens in Coverage

This issue can be also replicated locally running npx hardhat coverage.

To replicate do this:

• Clone provided repo on Issue description
• Unstall dependencies and compile contracts
• On master run npx hardhat coverage

This would throw the FATAL ERROR. If you simply delete contracts/fuzz/RTokenDiff.sol and run again npx hardhat coverage works fine! Even though the contract is not used anywhere it breaks the coverage execution.