[backport/1.4] Improve frontend: Deduplicate system-information requests

[joaoantoniocardoso]

This is a backport of #4027 into 1.4.

---

Supersedes #4034 (recreated from fork joaoantoniocardoso/BlueOS-docker instead of same-repo head).

[github-actions]

Automated PR Review

0. Summary

• Verdict: MINOR SUGGESTIONS ✏️

Backport of #4027 to 1.4-dev. Introduces a reference-counted subscription API on the system-information store so that the Cpu/Disk widgets, SystemCondition, and HealthTrayMenu no longer each spin their own setInterval to poll overlapping endpoints. All subscribed types are now fetched by a single OneMoreTime task at 2 s cadence, and the task is paused when the last subscriber unmounts. HealthTrayMenu also gains explicit subscriptions for temperature+disk, so it works even when the System page hasn't been visited. The diff is small, focused, and mechanically consistent with master.

1. Correctness & Implementation Bugs

• 1.1 [minor] core/frontend/src/store/system-information.ts:275-281 — subscribeSystemInformation reaches into OneMoreTime private state (isPaused, isRunning, timeoutId) via as any to disambiguate "never started" / "paused" / "already scheduled". This works today but is fragile: renaming any of those private fields in one-more-time.ts will silently regress the subscribe logic (starts get skipped or duplicated) with no compile-time signal. Consider adding a small public accessor on OneMoreTime (e.g. isActive(): boolean, or making the start() method idempotent w.r.t. a pending timeoutId) so the store doesn't need as any.
• 1.2 [minor] core/frontend/src/store/system-information.ts:79-81 — fetchSystemNetworkTask delay changed from 1000 → 2000 ms. This is a user-visible behavior change (network graph updates half as fast) that isn't mentioned in the PR title/description and isn't logically part of "deduplicate system-information requests". Either call it out in the description or split into its own commit.

4. Performance

• 4.1 [minor] core/frontend/src/widgets/Disk.vue:30 and core/frontend/src/widgets/Cpu.vue:96 — the standalone poll intervals were 30 s (Disk) and 5 s (Cpu); after this change both widgets subscribe to the shared 2 s task. When only one of these widgets is mounted (e.g. a dashboard with just the Disk widget), disk endpoint request frequency increases 15× and CPU 2.5×. The deduplication win still stands when multiple consumers are mounted, but the isolated-widget case gets more chatty than before. Confirm this is intentional.

6. Code Quality & Style

• 6.1 [minor] core/frontend/src/store/system-information.ts:268-282 — subscribeSystemInformation mixes two responsibilities (counter bookkeeping + task lifecycle) with an eslint-disable + as any escape hatch in the middle. Extracting the "make sure the task is active" tail into a private helper (ensureTaskRunning()) would isolate the any and make intent clearer.
• 6.2 [nit] core/frontend/src/store/system-information.ts:44-49 — a one-line comment explaining why system_information_subscribers is a module-scope map (rather than class state) would help future readers, since the rest of the store keeps its data on the class. Same pattern as the pre-existing notifier above it, so mentioning "kept outside Vuex reactivity" would suffice.
• 6.3 [nit] core/frontend/src/store/system-information.ts:290 — has_subscribers uses snake_case while the surrounding local (fetches) and task names are camelCase. Root config has camelcase: 'off', so ESLint won't flag it, but consistency with the neighboring code would be hasSubscribers.

8. Documentation

• 8.1 [nit] The three new public @Actions (fetchSubscribedSystemInformation, subscribeSystemInformation, unsubscribeSystemInformation) are the store's new public API. A short comment on each describing the reference-counted lifecycle (subscribe = ref++, unsubscribe = ref--, task auto-starts/stops) would save the next reader a trip through OneMoreTime.

Generated by PR Review Bot. This is advisory, a human reviewer must still approve.