Skip to content

CI: Known limitations of the Vercel preview smoke test #7261

@talissoncosta

Description

@talissoncosta

Context

PR #7192 adds a Vercel preview smoke test to the frontend PR workflow. The current approach parses preview URLs from the Vercel bot PR comment and curls them. This works reliably but has a few known limitations worth tracking.

Known limitations

  1. Grep pattern is brittlehttps://flagsmith-frontend[^ )]* depends on Vercel's comment format and project naming. If either changes, the grep silently finds nothing and the job warns but doesn't test anything.

  2. First-ever PR push — if the Vercel bot hasn't commented yet (brand new PR, first deployment still building), the poll waits up to 5 minutes. Usually enough, but not guaranteed if Vercel is slow.

  3. No content validation — curl checks HTTP status only. A broken deployment serving Vercel's error page still returns 200. Could add a body check (e.g., grep for a known string in the HTML response).

Possible improvements

  • Pin the grep pattern to the Vercel comment's markdown structure (e.g., match the table row format) instead of raw URL matching
  • Add a simple body check: curl ... | grep -q 'Flagsmith' to verify the app actually loaded
  • Consider using the Vercel API directly (api.vercel.com/v6/deployments) with a Vercel token for more reliable URL resolution — though this adds a secret dependency

References

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions