Follow-up: sub-ms peerstore RTT, real-network health tests, and health GUI polish (#1453/#1455)

Description

Follow-up to #1453 / #1455 (opt-in connection health monitor with /ipfs/ping/1.0.0 probes, docs, tests, and demo GUI).

#1455 landed the core work. This issue tracks remaining quality gaps called out in review and unfinished polish for the real-network and GUI deliverables.

Motivation

• Ping RTT is stored as integer milliseconds (perform_ping_roundtrip / _ping). On fast localhost paths that truncates to 0, so peerstore LatencyEWMA can record 0.0 even after a successful probe.
• Real-network coverage still depends on host interfaces and can be skipped on loopback-only CI; we should harden the integration story so dropped paths and RTT updates are reliable where a real network is available.
• The health GUI is useful but still minimal (no screenshots/docs walkthrough beyond CLI flags); polish would help operators adopt the demo.

Proposed work

1. Sub-ms peerstore RTT

• Record peerstore latency with finer resolution than truncated integer ms (e.g. fractional milliseconds, or a small floor when converting to seconds), without breaking the public ping helper’s go-libp2p-style ms convention if that remains intentional.
• Add/adjust tests so CI does not depend on EWMA > 0 alone when sub-ms RTTs are possible.

2. Real-network health integration test

Goal (from #1453): integration test with real peers on a real network (not only 127.0.0.1): at least two hosts, health enabled, assert scores/RTT update and that a dropped path is observed.

Current state after #1455: non-loopback binds + skip when only loopback exists. Follow-up:

• Keep skip-on-loopback-only, but document when CI/runners will skip.
• Strengthen assertions around score/RTT update and closed-connection detection on a true non-loopback path.
• Consider optional CI labeling / manual job for environments with real NICs so the test is exercised regularly.

3. Health GUI polish (nice to have)

Goal (from #1453): small live view of peer table (score, latency, success, protected, unhealthy); TUI or local web page fed by get_network_health_summary / export_health_metrics (JSON or Prometheus).

Current state after #1455: --gui tui|web exists. Follow-up:

• Docs polish: short operator walkthrough and/or screenshots of TUI and web UI.
• Optional UX improvements (refresh controls, clearer unhealthy highlighting) without expanding scope into a full dashboard product.

Out of scope

• Reworking default-off health monitoring semantics.
• Cross-implementation (go/js) comparison docs.

References

• Issue: #1453
• PR: #1455
• Review note: integer-ms RTT → peerstore 0.0 on fast CI (AI review v1)

Are you planning to do it yourself in a pull request?

Yes (maintainer follow-up; contributions welcome)

Start with perform_ping_roundtrip and _ping for RTT conversion, then inspect get_network_health_summary and export_health_metrics plus the existing non-loopback integration coverage. Done means sub-ms latency is represented reliably, real-network tests cover score/RTT updates and dropped paths while documenting loopback skips, and the GUI walkthrough or listed UX polish is complete.