Pause task-list polling when window is hidden

[pauldambra]

Problem

If you make cloud tasks somewhere else (e.g. PostHog Code mobile while on a walk) and then come back to the desktop, the new tasks don't appear in the list — not "slowly", not at all. Two things were going wrong:

• Polling was a hot 30s timer that never paused. It kept hitting the API even when the window was hidden.
• The on-return refetch was being silently swallowed. The global staleTime is 5 minutes and the global refetchOnWindowFocus is true, which means "refetch on focus if the query is stale". Returning within 5 minutes left the query "fresh", so no refetch fired — and 5 minutes is exactly the window where this hurts.

Changes

In apps/code/src/renderer/features/tasks/hooks/useTasks.ts, the three task-list queries (useTasks, useTaskSummaries, useSlackTasks) now:

• Gate refetchInterval on useRendererWindowFocusStore — polls every 3 minutes when the document is visible and the window has OS focus, otherwise stops.
• Set refetchOnWindowFocus: "always" so the return-to-desktop refetch fires regardless of staleness. This is the line that actually fixes the bug.

That's the whole change — one helper hook, one option per query.

How did you test this?

• Unit tests / typecheck on the touched file pass locally.
• I could not run the Electron app in my sandbox, so the on-return refetch hasn't been observed in a real window. Worth a 30-second smoke before un-drafting: blur the window, create a task from another client, return to the window, confirm the new task appears immediately (from the forced focus refetch) rather than waiting up to 3 minutes for the next poll.

Publish to changelog?

no

---

Created with PostHog Code

[greptile-apps]

_{Reviews (1): Last reviewed commit: "refactor(code): drop poll backoff — flat..." | Re-trigger Greptile}

[pauldambra]

Note

🤖 Automated comment by QA Swarm — not written by a human

Multi-perspective review: qa-team (specialists + generalists), paul-reviewer, xp-reviewer, security-audit.

Verdict: 💬 APPROVE WITH NITS

Clean, small, well-motivated change. The final-commit XP-style simplification (drop the tiered backoff, lean on refetchOnWindowFocus: "always") is a good design move. No security or correctness issues. The one finding with real weight is the 6x interval bump (30s → 3min) for users keeping the window focused — see inline comment on line 18. Most other notes are NITs around DRY (three identical option blocks across the three hooks) and naming.

Convergence (independent reviewers landed on the same things)

• Three places repeating the same refetchInterval + refetchOnWindowFocus: "always" pair — flagged by qa-team/generalist-a, paul, and xp (3 reviewers). See inline on useTasks.ts:83.
• 30s → 3min interval bump may make focused-and-watching feel slow — flagged by qa-team/data-integrity (MEDIUM) + paul (NIT). See inline on useTasks.ts:18.

Off-diff observations (no inline anchor — flagging here)

• apps/code/src/renderer/stores/rendererWindowFocusStore.ts:5 — qa-team/frontend: the JSDoc says "Used to pause inbox polling when the Electron window is in the background." This store now also drives task-list polling (and is imported elsewhere). Generalize, e.g. "Used to pause polling-style queries when the Electron window is in the background."
• apps/code/src/renderer/stores/rendererWindowFocusStore.ts — paul: TanStack Query already wires focus/blur/visibilitychange into its own focusManager over in queryClient.ts, and this store listens to the same three events for what feels like the same job. Worth checking whether focusManager.isFocused() (or just relying on TanStack's refetchIntervalInBackground default) could do the work without a second store. Not blocking — possibly already considered.
• Naming clash — xp: useFocusStore (session/workspace focus) and useRendererWindowFocusStore (OS window visibility) are close enough that a future reader will mis-import. Outside this PR's scope; flag as a follow-up rename candidate (e.g. useWindowVisibilityStore).

Reviewer summaries

Reviewer	Assessment
🔍 qa-team	APPROVE WITH NITS — one MEDIUM (interval bump trade-off), the rest LOW/NIT.
👤 paul	approve with comments — nice catch on the staleTime/refetchOnWindowFocus interaction; main poke is whether the new store duplicates TanStack's focusManager.
📐 xp	Good XP-shaped PR — the final-commit deletion of the backoff + tests was the right call. One real finding: collapse the three repeated option triples.
🛡 security-audit	No findings. Client-side polling cadence change only — no new endpoint, sink, auth surface, or input handling.

---

Automated by QA Swarm — not a human review