## Summary
The nav bar's packet / node / observer counts were refreshed three ways
at once: a 15 s timer, every batch of live WebSocket packets
(`debouncedOnWS`, 250 ms), and the `/stats` client cache being cleared
every 5 s while packets arrive. None of them checked whether the tab was
visible. So every open tab, including hidden ones, asked `/api/stats`
every few seconds for as long as it stayed open.
On one production instance's nginx log for a single day, `/api/stats`
was **135,654 of 187,411** CoreScope API requests (72%), about 20,000 a
day for each browser that kept a tab open. Each response is cheap thanks
to the 10 s server-side cache (#1910), so this is about request volume,
bandwidth and log noise, not server CPU.
This PR adds `createVisibleThrottle()` to `app.js` and routes all three
triggers through one throttled refresh:
- at most one `/api/stats` request per 15 s from a visible tab (the old
timer's own period);
- none from a hidden tab;
- one immediate refresh when a hidden tab becomes visible again.
No existing issue covers this.
## Performance
Measured with headless Chromium, counting `/api/stats` requests from a
real tab: 60 s visible, then 60 s with the page hidden
(`document.hidden` and a `visibilitychange` event), then back to
visible.
| `/api/stats` requests | production (current release) | staging,
`master` | staging, this PR |
|---|---|---|---|
| visible tab, per minute | 9 | 4 | 4 (at +10, +25, +40, +55 s) |
| hidden tab, per minute | 6 | 5 | **0** |
| on return to visible | n/a | n/a | 1, immediately |
The same check confirmed the nav bar still renders its counts (`… pkts ·
… nodes · … obs`).
## Tests
- Red commit `2c81a3e4`: `tests/unit/test-nav-stats-throttle.js` loads
the real `public/app.js` in a `vm` context. `createVisibleThrottle`
lands as a pass-through stub, so the test fails on its first assertion
(40 refreshes where 1 is expected).
- Green commit `f0ba92ab`: the throttle, and the nav stats timer, WS
hook and a new `visibilitychange` listener calling one throttled
trigger. All 6 cases pass: one refresh per burst, again after the
interval, at most 241 an hour, none while hidden, one on return, and a
source check that `updateNavStats` is wrapped and not called by the
timer directly.
- Added to `test-all.sh` in sorted order; `test-test-inventory.js`
passes.
- Frontend suites: 194 of 195 pass. The one failure,
`test-issue-1956-release-routing.js`, needs CI's `GITHUB_REF_NAME` and
fails the same way on `master` locally.
## Review notes
- `home.js` keeps its own one-shot `/stats` load on the home page; only
the nav bar's polling changed.
- A tab opened in the background loads its counts when it is first
shown.
- The 15 s interval is a constant at the call site. Could be exposed in
the customizer later (AGENTS.md rule 8); not added here.
- No Playwright E2E added: the change is timing behaviour, covered by
the unit test against the real file and by the browser measurement
above. Happy to add an E2E if wanted.
## SonarLint
SonarQube Community rules (the SonarLint engine) were run on the files
this PR touches, and only issues on lines it adds are listed.
- **Fixed** in `9eb94ea1`: `node:` specifiers (S7772).
- **Kept on purpose:**
- S7773 wants `Number.isFinite`/`Number.parseInt` for the `isFinite` and
`parseInt` handed to the `vm` sandbox, but `app.js` calls the global
functions, so those are the ones it needs.
- S1523 flags `vm.runInContext`, which is how the frontend unit tests
load the repo's own `public/app.js`.
## Checked against the repo rules
- Rebased onto `master` (`673442cc`). Each red commit was re-run and
still fails on its assertions; each green commit passes (TDD.md).
- `make test`: all 16 modules pass. Frontend suites: all pass except
`test-issue-1956-release-routing.js`, which needs CI's `GITHUB_REF_NAME`
and fails the same way on `master` locally (#2160 addresses it).
- No private data: test fixtures and examples use the codes the existing
tests use (SJC/SFO/LAX, `#be`-style scopes).
- No new `map[string]interface{}`, no hard-coded colours. `git add` was
explicit, and there is one logical change per commit.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_01KaR4xu8gVykQBkH8yG8m6H
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
