mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-08 22:13:59 +01:00
fb96537e28
## Thinking Path > - Paperclip lets people manage agent work through the board. > - Deploys can interrupt reads and writes while people use the board. > - Component tests cannot prove the browser keeps task content and saved drafts through a reload. > - The read and cache changes need a regression suite against the built app. > - This pull request drives those journeys through temporary API failures. > - CI discovers the new suite through the existing e2e shard plan. ## Linked Issues or Issue Description **What happened?** Outage recovery lacked one browser suite covering retained content, write recovery, cached reloads, access errors and sign-out clearing together. **Expected behavior** Keep loaded task content visible through a 20-second API outage. Show one connection banner and its recovery state. Save comments once and retain document edits. Restore approved reads after a reload. Keep real error responses readable and clear query records on sign-out. **Steps to reproduce** Open a task or settings. Return 503 with `tenant_app_unavailable` from browser API requests for 20 seconds. Type a comment, edit a document, or reload. Restore the API and compare the visible result with the saved server state. Repeat with real 403/404 statuses and sign-out. Stacked on Refs #18825, which depends on Refs #18822. Related durable-write foundation: Refs #17960. ## What Changed - Add `disconnect-resilience.spec.ts` with ten browser cases for comments, a queued comment across reload, document autosave, cached settings/sidebar/task reads, legacy Inbox reloads, 403 and 404 responses, sign-out clearing, restored-draft conflict detection, and first-socket outage control. - Assert one connection banner, readable recovery, no transient error boxes or raw codes, retained task text, one saved comment, and the saved document after reload. - Use the real local server and isolated embedded PostgreSQL. Intercept browser API traffic for outages. Control the logout/session response for the sign-out test. Create no agents or model runs. - Register realtime interception before navigation. Track both ends of current sockets, close them at outage start, and keep reconnects unavailable through the HTTP outage deadline. Preserve normal close forwarding and close tracked pairs during teardown. The regression captures the first browser socket without waiting for history or open, then checks that the outage closes it. - Prepare complete server-derived experimental settings and workspace preferences before the legacy Inbox navigation. Its route callbacks serve per-test snapshots across reloads, avoiding reads from disposed route.fetch responses. - Document the direct test command. The existing CI e2e shards already discover this file exactly once; no workflow or package script is added. ## Verification - Current head `228c1b68d30` is merge-ready: all 54 check runs completed (52 successful, one skipped, one neutral), the legacy Snyk status is successful, and both required gates pass. The single failed-jobs debug rerun in run `41752978283` passed on attempt 2, including Runner package job `126328280650` and `verify` job `126332119116`. Greptile Apex reviewed this exact head at 5/5 with zero unresolved threads; the branch has no merge conflicts. The three-PR stack is awaiting operator merge approval. - Original attempt 1 of current-head CI run `41752978283` had 46 successful jobs, one skipped and two failed: Runner package tests and the downstream `verify` gate. GitHub reports lost communication with the Runner host; the test step was cancelled, its log returns HTTP 404, and no Runner diagnostic artifact is available. The underlying test outcome and disconnection cause are unknown. One failed-jobs retry with debug logging was accepted after verifying the unchanged current head and passed on attempt 2. No source change was made; earlier passing CI and local tests do not certify this incomplete Runner result. Current-head CI now passes; the earlier disconnection cause remains unknown. - Preceding head `6531caccf62` reached merge-ready: all 54 check runs completed (52 successful, one skipped, one neutral), the legacy Snyk status was successful, and both required gates passed in run `41749512329`. Greptile reviewed that exact head at 5/5 with zero unresolved threads. Current head `228c1b68d30` carries later account-preference and bulk-selection fixes and now has its own passing current-head CI. - All 23 built-UI outage/write/loading cases pass in 4.5 minutes on exact current head `228c1b68d30`: ten outage, six durable-write and seven loading cases. The run used bundled Chromium, a fresh local-trusted server and isolated embedded PostgreSQL on macOS arm64/Node 25.6.1. Owned temporary server and database stopped. - Current integrated head `228c1b68d30` contains cache revision `fe4f682c879` and read head `09d1db29dea`. The exact application source passes all 1,198 focused tests across twenty-eight files, UI typecheck and both guards. Seventeen new cases cover restored-account settings, owned crown cleanup and pending bulk selections; sixteen fail the previous source and one preserves denial compatibility. All three browser specs pass standalone strict checking. All 23 current-head browser cases pass in 4.5 minutes. - Application revision `9c1a10f25d9` passed all 1,137 focused tests across twenty-one files, UI typecheck and both guards. Exact preceding application/spec head `4fec693ec94` passed all 23 built-UI outage/write/loading cases in 4.6 minutes and standalone strict checking of all three specs. It used bundled Chromium, fresh isolated PostgreSQL and local-trusted auth on macOS arm64/Node 25.6.1. Owned test processes stopped. - Preceding integrated head `870aa709657`, containing cache `cbb891dd37d`, passes all 23 built-UI outage/write/loading cases in 4.6 minutes: ten outage cases, six durable-write cases and seven loading cases. It used bundled Chromium, the built UI, a fresh local-trusted server and isolated embedded PostgreSQL on macOS arm64/Node 25.6.1. Owned test processes stopped. - The earlier head `763fe431b74` passes the affected legacy Inbox outage/reload case (34.3 seconds; 50.8 seconds including startup) and strict three-spec checking after the response-snapshot fix. The preceding head `973a12d77f4` passed all 23 built-UI outage/write/loading cases in 4.6 minutes: ten outage cases, six durable-write cases and seven loading cases. Its first-socket regression fails the previous helper because that socket stays open. The run used bundled Chromium, the built UI, fresh isolated PostgreSQL and local-trusted auth on macOS arm64/Node 25.6.1. That earlier isolated callback revision was verified with its affected case; the later application/spec head `4fec693ec94` also passes the full suite. Owned test servers and databases stopped. - The current base gives known skills/app errors priority over sibling loading. All 60 affected tests and both application UI typechecks passed. Its twelve new regressions failed the preceding source. All 47 recovery-worker tests and seven related Runner fixture variants also passed during the historical CI audit; those focused checks do not claim a full current server/Runner suite. - All 22 outage/write/loading cases passed in 4.7 minutes on preceding integrated head `b7deaaee5da`, including direct cache revision `97c0bc3966f`. The run used the built UI, bundled Chromium and a fresh local-trusted server with isolated embedded PostgreSQL on macOS arm64/Node 25.6.1. It covered all nine outage, six durable-write and seven loading cases. The owned temporary server and database stopped. - All 1,038 focused tests passed on that direct cache revision, with UI typecheck and guards. Six regressions failed the previous source: online retries could trigger a hidden-tab worker reload, and sign-out failed without crypto.randomUUID. Both modes and stamp states now pass. Standalone strict checking passes for all three browser specs. - The restored cached-A/draft-B/server-C test asserts zero replay PATCHes and retains the conflict. The queued-comment case holds fresh session verification after general API recovery and asserts zero POSTs until release. Denial, logout and cached-navigation cases require saved records. Sign-out seeds several company records to exercise native Storage key reordering. Healthy navigation waits for conversation-history readiness. - Earlier runs exposed real cached-header, storage cleanup and replay defects in the base. Those failures and corrected fixture assumptions remain in the discussion. A prior run passed 21/22 while real sockets remained available during simulated HTTP failures. A diagnostic trace established the contradiction. The first repair gated new sockets, but review identified the first pre-existing socket still escaping. This revision intercepts before navigation and closes existing pairs too. The new regression fails the preceding helper, and all 23 current cases pass without weaker assertions. Shared-memory exhaustion prevented two earlier diagnostic startups; only unused segments from completed owned test databases were removed. - All six separate cross-organization link/upload variants passed on an earlier integrated base. Standalone strict checking of that unrelated spec reports five existing errors in its shared fixture; this PR does not claim that check passes. - CI on older head `53d979e384c` failed the legacy Inbox case with `apiResponse.json: Response has been disposed` inside a route.fetch callback. Its other 25 cases passed and one skipped. The seven other browser shards succeeded; the aggregate only reports that child failure. The new snapshot callbacks remove those fetched-response lifetimes. Original CI evidence remains recorded in the discussion. - Greptile Apex reviewed exact current head `228c1b68d30` at 5/5 with no actionable findings and zero unresolved threads (summary updated 2026-11-04 at 01:26:14 UTC). The branch has no merge conflicts. Current-head CI now passes; the earlier disconnection cause remains unknown. Controlled local outages do not prove hosted authentication or a real Cloud restart/deploy. Commands run: ```sh pnpm exec playwright test --config tests/e2e/playwright.config.ts tests/e2e/disconnect-resilience.spec.ts --grep 'a legacy Inbox reload retains' --reporter=list pnpm exec playwright test --config tests/e2e/playwright.config.ts tests/e2e/disconnect-resilience.spec.ts tests/e2e/disconnect-write-recovery.spec.ts tests/e2e/task-loading-frame.spec.ts --reporter=list pnpm exec tsc --ignoreConfig --noEmit --strict --target ES2022 --module NodeNext --moduleResolution NodeNext --skipLibCheck tests/e2e/disconnect-resilience.spec.ts tests/e2e/disconnect-write-recovery.spec.ts tests/e2e/task-loading-frame.spec.ts ``` ## Risks The suite adds about three minutes to one e2e shard and depends on current accessible UI controls. It uses controlled outages and logout responses. A passing local run must not be treated as real deployment acceptance. There are no production behavior, schema, dependency, or API changes in this PR. ## Model Used OpenAI `gpt-6.1-sol` through Codex, with xhigh reasoning effort, tool use, and code execution. The context window size is not exposed. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for related PRs and linked them above - [x] I have described the issue following the bug report template - [x] I have not referenced internal/instance-local Paperclip issues or links - [x] My branch name describes the change and contains no internal ticket id - [x] All 23 built-UI cases pass on current head `228c1b68d30` - [x] All three browser specs pass standalone strict typechecking on the current head - [x] The current application source passes all 1,198 focused tests, UI typecheck and both guards - [x] I have added tests and updated relevant documentation - [x] I have considered and documented risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: devinfoley <139239+devinfoley@users.noreply.github.com> Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Bender (Fable) <bender@agents.paperclip.ing> Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>