Move install to separate step and give more leeway to benchmarks - #712
Conversation
The install step is executed per retry of the benchmark (up to 3 times) More leeway because of GitHub actions runners being slow and unreliable for testing performance.
| # ignore poetry install failures, these can be due to locking issues when | ||
| # 2 poetry installs are run at the same time. | ||
| poetry install || true; | ||
| poetry install |
There was a problem hiding this comment.
🟡 Medium - Removing the install failure suppression re-breaks concurrent sample-app startup in end-to-end CI
The shared install target now returns the raw poetry install exit status even though the repository still starts two targets from the same sample-app directory in parallel during end-to-end tests. In .github/workflows/end2end.yml, make runZenDisabled is backgrounded and then make run is started shortly after; both targets depend on install, so they can hit the same Poetry lock/contention that the removed || true was explicitly masking. When one of those installs exits non-zero, the corresponding app process never comes up and the matrix job fails or flakes across many sample apps.
| poetry install | |
| poetry install || true |
More info - Reply on this comment to give feedback or ignore the issue.
The install step is executed per retry of the benchmark (up to 3 times)
More leeway because of GitHub actions runners being slow and unreliable for testing performance.