flake: e2e externalAuth

[spikecurtis]

==> Starting test external auth device
==> Finished test external auth device: failed
==> Starting test external auth web
==> Finished test external auth web: skipped
==> Starting test successful external auth from workspace
==> Finished test successful external auth from workspace: skipped
Data from pprof has been saved to test-results/debug-pprof-goroutine-external-auth-device.txt
  pw:api => selectors.setTestIdAttribute started +8ms
==> Output
  pw:api <= selectors.setTestIdAttribute succeeded +3ms
==> Errors
Error: listen EADDRINUSE: address already in use 0.0.0.0:50516
Data from pprof has been saved to test-results/debug-pprof-goroutine-successful-external-auth-from-workspace.txt
==> Output
Data from pprof has been saved to test-results/debug-pprof-goroutine-external-auth-web.txt
==> Output
[stderr] 
[stderr] 
[stderr] CSRF token not found

seen here: https://github.com/coder/coder/runs/32976358533

Looks like these use hardcoded ports, and so they flake if the port isn't available.

[mtojek]

I might be wrong but we enabled parallel e2e tests, right @matifali?

It could be a reason why these are failing. We can refactor them, but it could be harder than a simple port change (starting agent with Prometheus endpoint).

[matifali]

I think we do not run e2e tests in parallel., What I changed was only the runner type (moved from 16-core to a 4-core runner).

[dannykopping]

Provisionally closed by coder/coder#15537

[johnstcn]

This cropped up again, opened #356 to track.