fix(e2e): extend hub-soft-probe skip to push + staging dispatch so a broken hub cannot red the post-merge check

YousifShkara commented

2026-07-01 10:15:20 +02:00

Owner

The merge of PR #306 fired the push: main trigger on e2e.yml. wait-for-deploy.mjs polls ${op}/v1/version on the API host (api.a8n.systems), which has been up throughout the outage, so the gate passed immediately and the Playwright suite ran against staging while the hub (a8n.systems) was still serving the mispushed Dioxus SPA. [setup] global.setup.ts timed out for 180s waiting for form[action="/login"] input[type="email"] on a page that has no such form; [auth-ui] login.spec.ts inherited the same failure and the merge check turned red.

BUNYIP-301 + BUNYIP-303 fixed the equivalent problem on pull_request by softening reachability and gating the suite step on steps.reachability.outputs.hub_live != 'false'. That gate did not fire on push because the reachability step's if: only listed pull_request and workflow_dispatch (production); on push the step never ran, its output was unset, and the gate's fallback treated an unset output as "hub OK" (which was the intended pre-BUNYIP-305 behaviour).

Drop the reachability step's if: entirely so it runs on every trigger. Move the soft/hard split into the E2E_HUB_SOFT env expression: 'false' (hard) only on production dispatch, 'true' (soft) everywhere else (push, pull_request, staging dispatch). The E2E suite gate needs no code change; its existing != 'false' conjunction now sees hub_live set on every non-production trigger and correctly skips the suite when the hub is soft-degraded.

Rationale for softening on push too: post-merge deploy of bunyip-web is a manual just app-restart on c-01. CI cannot force the restart and cannot know when the operator will run it. Skipping the suite with a loud log line ("hub /healthz probe reported not-live; ... operator must just app-restart bunyip-web on the target host to pick up the new image") surfaces the same signal as a 3-min timeout + red run, without the punitive UX. Production dispatch KEEPS the hard check: it is meant to page loudly on a prod outage, so E2E_HUB_SOFT resolves to 'false' for that event and the reachability step exits 1 before the suite runs.

health-check.mjs needs no change; it already writes hub_live=true|false to $GITHUB_OUTPUT when the reachability step runs, and now the step runs on every non-outage-hardgate path.

Comments on the reachability step + the E2E suite step + the "Note skipped suite (hub down)" step are all updated to spell out the new event coverage. The skipped-suite log line includes an operator-facing hint about just app-restart so the next reviewer knows which manual step unblocks a red post-merge check without having to search the runbook.

#BUNYIP-305

The merge of PR #306 fired the `push: main` trigger on `e2e.yml`. `wait-for-deploy.mjs` polls `${op}/v1/version` on the API host (`api.a8n.systems`), which has been up throughout the outage, so the gate passed immediately and the Playwright suite ran against staging while the hub (`a8n.systems`) was still serving the mispushed Dioxus SPA. `[setup] global.setup.ts` timed out for 180s waiting for `form[action="/login"] input[type="email"]` on a page that has no such form; `[auth-ui] login.spec.ts` inherited the same failure and the merge check turned red. BUNYIP-301 + BUNYIP-303 fixed the equivalent problem on `pull_request` by softening reachability and gating the suite step on `steps.reachability.outputs.hub_live != 'false'`. That gate did not fire on push because the reachability step's `if:` only listed `pull_request` and `workflow_dispatch (production)`; on push the step never ran, its output was unset, and the gate's fallback treated an unset output as "hub OK" (which was the intended pre-BUNYIP-305 behaviour). Drop the reachability step's `if:` entirely so it runs on every trigger. Move the soft/hard split into the `E2E_HUB_SOFT` env expression: `'false'` (hard) only on production dispatch, `'true'` (soft) everywhere else (push, pull_request, staging dispatch). The E2E suite gate needs no code change; its existing `!= 'false'` conjunction now sees `hub_live` set on every non-production trigger and correctly skips the suite when the hub is soft-degraded. Rationale for softening on push too: post-merge deploy of bunyip-web is a manual `just app-restart` on c-01. CI cannot force the restart and cannot know when the operator will run it. Skipping the suite with a loud log line ("hub /healthz probe reported not-live; ... operator must `just app-restart` bunyip-web on the target host to pick up the new image") surfaces the same signal as a 3-min timeout + red run, without the punitive UX. Production dispatch KEEPS the hard check: it is meant to page loudly on a prod outage, so `E2E_HUB_SOFT` resolves to `'false'` for that event and the reachability step exits 1 before the suite runs. `health-check.mjs` needs no change; it already writes `hub_live=true|false` to `$GITHUB_OUTPUT` when the reachability step runs, and now the step runs on every non-outage-hardgate path. Comments on the reachability step + the E2E suite step + the "Note skipped suite (hub down)" step are all updated to spell out the new event coverage. The skipped-suite log line includes an operator-facing hint about `just app-restart` so the next reviewer knows which manual step unblocks a red post-merge check without having to search the runbook. #BUNYIP-305

YousifShkara self-assigned this

2026-07-01 10:15:20 +02:00

YousifShkara added 1 commit

2026-07-01 10:15:20 +02:00

fix(e2e): extend hub-soft-probe skip to push + staging dispatch so a broken hub cannot red the post-merge check

Create release / Create release from merged PR (pull_request) Has been skipped

Details

E2E / Playwright against deployment (pull_request) Successful in 1m34s

Details

Check / fmt + clippy + build + tests (pull_request) Successful in 21m18s

Details

73b02971db

The merge of PR #306 fired the `push: main` trigger on `e2e.yml`. `wait-for-deploy.mjs` polls `${op}/v1/version` on the API host (`api.a8n.systems`), which has been up throughout the outage, so the gate passed immediately and the Playwright suite ran against staging while the hub (`a8n.systems`) was still serving the mispushed Dioxus SPA. `[setup] global.setup.ts` timed out for 180s waiting for `form[action="/login"] input[type="email"]` on a page that has no such form; `[auth-ui] login.spec.ts` inherited the same failure and the merge check turned red.

BUNYIP-301 + BUNYIP-303 fixed the equivalent problem on `pull_request` by softening reachability and gating the suite step on `steps.reachability.outputs.hub_live != 'false'`. That gate did not fire on push because the reachability step's `if:` only listed `pull_request` and `workflow_dispatch (production)`; on push the step never ran, its output was unset, and the gate's fallback treated an unset output as "hub OK" (which was the intended pre-BUNYIP-305 behaviour).

Drop the reachability step's `if:` entirely so it runs on every trigger. Move the soft/hard split into the `E2E_HUB_SOFT` env expression: `'false'` (hard) only on production dispatch, `'true'` (soft) everywhere else (push, pull_request, staging dispatch). The E2E suite gate needs no code change; its existing `!= 'false'` conjunction now sees `hub_live` set on every non-production trigger and correctly skips the suite when the hub is soft-degraded.

Rationale for softening on push too: post-merge deploy of bunyip-web is a manual `just app-restart` on c-01. CI cannot force the restart and cannot know when the operator will run it. Skipping the suite with a loud log line ("hub /healthz probe reported not-live; ... operator must `just app-restart` bunyip-web on the target host to pick up the new image") surfaces the same signal as a 3-min timeout + red run, without the punitive UX. Production dispatch KEEPS the hard check: it is meant to page loudly on a prod outage, so `E2E_HUB_SOFT` resolves to `'false'` for that event and the reachability step exits 1 before the suite runs.

`health-check.mjs` needs no change; it already writes `hub_live=true|false` to `$GITHUB_OUTPUT` when the reachability step runs, and now the step runs on every non-outage-hardgate path.

Comments on the reachability step + the E2E suite step + the "Note skipped suite (hub down)" step are all updated to spell out the new event coverage. The skipped-suite log line includes an operator-facing hint about `just app-restart` so the next reviewer knows which manual step unblocks a red post-merge check without having to search the runbook.

#BUNYIP-305

YousifShkara closed this pull request

2026-07-01 10:15:22 +02:00

YousifShkara deleted branch fix/BUNYIP-305-e2e-push-hub-skip

2026-07-01 10:15:26 +02:00

Create release / Create release from merged PR (pull_request) Has been skipped

Details

E2E / Playwright against deployment (pull_request) Successful in 1m34s

Required

Details

Check / fmt + clippy + build + tests (pull_request) Successful in 21m18s

Required

Details

Pull request closed

This pull request cannot be reopened because the branch was deleted.

Rows
Columns

fix(e2e): extend hub-soft-probe skip to push + staging dispatch so a broken hub cannot red the post-merge check #307

Pull request closed