AUT-1462: Playwright harness to reproduce the intermittent stg redirect loop#20
AUT-1462: Playwright harness to reproduce the intermittent stg redirect loop#20shoffman-smartling wants to merge 1 commit into
Conversation
…direct loop Adds test/playwright-repro/, a self-contained tool that provisions a throwaway, per-client-scoped stg Keycloak client + test user (dry-run by default, guarded against touching anything it didn't create), then drives real Chrome through this repo's own keycloak-connect middleware (via a minimal fixture app) to try to trigger the KC26 session-cap redirect loop on demand instead of by luck. Along the way, confirmed and worked around several environment specifics (public vs internal Keycloak hostnames, realm-scoped vs master admin accounts, the stg login theme's two-step form), and surfaced a real, separate defect in grant-manager.js's refresh_token handling for public clients, tracked as AUT-1468. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
| @@ -0,0 +1,73 @@ | |||
| 'use strict'; | |||
|
|
|||
| // Minimal Express app built on THIS repo's own keycloak-connect code (not | |||
There was a problem hiding this comment.
It's a fascinating idea to isolate and test the problem. However, we have two additional layers that could introduce additional business logic (smartling-express/security and ti-next/tms-dashboard-app).
I believe it makes sense to use the smartling-express/security library to create this app.
There was a problem hiding this comment.
Adding additional layers on the framework, eventually getting to full e2e dashboard/CAT tool operations is an end goal, but starting at the lowest level and adding levels sequentially allows us to isolate issues between the levels at this critical phase.
| entirely in this code, so testing it in isolation is sufficient to reproduce | ||
| and diagnose it. | ||
|
|
||
| **Does NOT (yet):** touch `tms-dashboard-app`, `ti-next`, `wa`, or any other |
There was a problem hiding this comment.
Thank you for this scaffold project. I'm sure it will save us time to create the second test tomorrow (or on Fri).
To team: Guys, you are welcome to challenge the LLM to create it while DmitryM is working on a new dirty hack.
2 participants
Summary
test/playwright-repro/, a self-contained Playwright harness that provisions a throwaway, per-client-scoped stg Keycloak client + test user and drives real Chrome through this repo's ownkeycloak-connectmiddleware (via a minimal fixture app) to try to trigger the intermittent KC26 session-cap redirect loop on demand.provision.jsis dry-run by default, scoped to stg only, and guarded so it can never create/modify/delete anything it doesn't already own (see README "Safety model").grant-manager.js's refresh_token handling for public clients: AUT-1468.test/playwright-repro/README.md.ROADMAP-dashboard.mdsketches what it'd take to extend this to the realtms-dashboard-app.Current status
0/6 iterations reproduced the loop against current
master(1 single-tab + 5 two-tab runs) - clean in every case, no errors. Per the README's own guidance, this isn't trustworthy as a "no repro" result yet until the harness is validated against the known pre-fix bug (checking out the commit beforecbef5ad) - that's the natural next step.Test plan
cd test/playwright-repro && npm install, copy.env.exampleto.envwith your own stg admin credentials (must be a Smartling-realm admin, not master - see README)npm run provision:planthennpm run provision:applynode repro.js --iterations=1 --tabs=1 --headedto watch one runcbef5ad) to confirm the harness actually catches the known bug before trusting further "no repro" results🤖 Generated with Claude Code