Diagnostics subscribes to live process samples but reads timeline history only once. On a fresh monitor, the timeline stays empty while live counters update; clicking refresh immediately shows the retained samples.
Refresh the existing history query every five seconds while it is observed, and dispose it immediately when the view closes so polling stops. This is a focused stale-data bug fix under the small, obvious bug exception: it updates the existing timeline from the same RPC and leaves sampling, retention, contracts, and controls unchanged.
Fixes #30520.
Verification:
dd69b3166561): history is read once instead of refreshing. It passes with the fix and checks that polling stops after unmount.vp run check:changed passed: formatting, lint, both knip checks, affected typechecks (including web, desktop, and mobile), and all 34 tests in server.test.ts.Evidence is attached to an evidence-only prerelease in my fork, outside the source history:
Before recording · After recording: timeline updates without clicking refresh
The after recording keeps the Resource timeline visible throughout; monitor/server sample counts increase automatically (127 to 151 during the capture).
Implemented with GPT-6.1 Sol through the Codex harness in T3 Code.
Added 1 commit
Verdict: Needs Human Approval
Macroscope's review found this PR requires human approval — Diagnostics history will automatically refresh every five seconds while observed, with polling stopped when the query is disposed. This changes the default refresh behavior and request cadence, which requires human acceptance.
You can add or adjust custom eligibility rules. Learn more.
No actionable comments were generated in the recent review.
The resource telemetry history query now refreshes every five seconds and uses a zero idle TTL. A test verifies periodic reads while the history atom is mounted and confirms reads stop after unmount.
Changes: Resource history updates
| Layer / File(s) | Summary |
|---|---|
History polling and lifecycle validation packages/client-runtime/src/state/server.ts, packages/client-runtime/src/state/server.test.ts | The history query now polls every five seconds and uses a zero idle TTL. The test verifies timer-triggered reads and confirms that reads stop after unmount. |
Estimated code review effort: 2 (Simple) | ~10 minutes
Change: Bug fix · Severity of issue fixed: Medium
Possibly related PRs
Suggested labels: bugfix, size:S
Suggested reviewers: t3dotgg, juliusmarminge, macroscopeapp
c7e11The timeline refreshes while Diagnostics is open. No merge-blocking risk was identified.
| Check name | Status | Explanation |
|---|---|---|
| Title check | Passed | The title "fix(web): refresh resource timeline while diagnostics is open" matches the main change. The PR adds five-second polling to the existing resource timeline history query while the Diagnostics… |
| Linked Issues check | Passed | Issue #30520 requires the Diagnostics resource timeline to update without manual refresh. server.ts adds a 5-second refresh interval to the existing resourceTelemetryHistory query and sets `idleTt… |
| Out of Scope Changes check | Passed | The change is limited to the existing resource telemetry history query and its regression test. The implementation does not change sampling, retention data, RPC contracts, or controls. The test direct… |
| Description check | Passed | The description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the bug, the implementation, the small obvious bug exception and linked issue, focuse… |
Explanation
The title "fix(web): refresh resource timeline while diagnostics is open" matches the main change. The PR adds five-second polling to the existing resource timeline history query while the Diagnostics view is open, and stops polling on unmount. The title follows the conventional commit format and is specific. It names the web scope, though the change also applies to the desktop app through the shared Diagnostics entry point. That is a minor scope note, not a mismatch.
Explanation
Issue #30520 requires the Diagnostics resource timeline to update without manual refresh. server.ts adds a 5-second refresh interval to the existing resourceTelemetryHistory query and sets idleTtlMs: 0, so the query stops being retained when unmounted. server.test.ts verifies the initial read, the timed refresh, the updated sample count, and no reads after unmount.
Explanation
The change is limited to the existing resource telemetry history query and its regression test. The implementation does not change sampling, retention data, RPC contracts, or controls. The test directly verifies the requested polling lifecycle, so no unrelated change is demonstrated.
Explanation
The description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the bug, the implementation, the small obvious bug exception and linked issue, focused regression coverage, manual verification, untested connection modes, evidence, and the agent model and harness.
3687a5ec-80c6-4aca-9dcd-8aa304fd02abReviewing files that changed from the base of the PR and between dd69b316656119b4613b13f3c076f45d4dbda2ce and c7e11a61f3cb0cf509ce6e65a2b43a514c7c1a4a.
packages/client-runtime/src/state/server.test.tspackages/client-runtime/src/state/server.tsIncluded review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.
Note
Written by gpt-6.1-sol on behalf of Maria
verified as-is at c7e11a61f3cb0cf509ce6e65a2b43a514c7c1a4a on a disposable linux vm. used repository Playwright with /usr/bin/google-chrome, one background dev server at a time, isolated empty development databases, and the real native resource monitor. created the same project pointing at a real git checkout through the app and completed Codex gpt-6.1-sol turns through the web composer on both revisions. no mocks, fixtures, replay provider, or source changes.
github/main (d51bc58f5e05123a416a3737ac4bce42d1d76b7c), opened settings > maintenance > diagnostics, installed the missing native monitor, and clicked retry monitor. live counters updated, but the resource timeline stayed empty for 52 seconds. manual refresh immediately displayed 53 server/monitor samples and retained provider processes.windowMs=300000, bucketMs=15000; manual refresh sent another history request; returning to 15m restored its parameters. leaving diagnostics for general settings produced zero further history requests over 34 seconds. reopening resumed polling and displayed fresh history, including the real Codex process after a successful follow-up turn.| 1 | main: vp i; background vp run dev --no-seed |
| 2 | server=16733 web=8693; reproduced stale timeline; stopped captured process group |
| 3 | head: vp i; background vp run dev --home-dir /home/user/t3code/.t3 --no-seed |
| 4 | server=13773 web=5733; automatic history updates and lifecycle checks passed |
| 5 | blacksmith: vp test run packages/client-runtime/src/state/server.test.ts |
| 6 | 1 file passed, 34 tests passed |
same viewport, project, prompt, 15m window, and fresh-monitor recovery scenario; resource values are measured independently on each run. inspected captures and verified uploaded video playback and seeking.
main: healthy monitor but empty resource timeline
pr: resource timeline updates automatically to 21 samples
pr: empty timeline fills automatically and sample counts increase
packaged electron, native ios/android, macos/windows host signals, and relay/tunnel or multiple-environment connections were not exercised on this linux vm. mobile has no consumer of this history query. other provider adapters were not exercised; this change uses the existing environment history RPC and does not change provider protocols. CI is green; the local verification ran the focused existing test file through blacksmith rather than rerunning the full CI suite.