Viewer polish

Goal

Make every number in the viewer say when it is from and where it comes from, and make sure no screenshot shows the username.

Why

A gauge with no timestamp looks like “now” even if the reading is 7 minutes old (ssh failed and the last good one is shown). A curve with no window does not say whether it covers 15 minutes or 4 hours. And a counter that does not add up to the same as the cards below it makes you doubt everything else. It is the difference between data and metadata: a number is only useful if it travels with its time and its origin.

Follows 28 — Telemetry and 29 — Priorities.

Steps

Sections 1 to 3 are the steps: time clarity, three things that did not add up and path privacy.

1. Time clarity

• Each gauge carries ahora · hace N s (“now · N s ago”) underneath, computed from telemetria.at: the time of the measurement itself, not that of live.json. It turns amber at 60 s and red (“lectura vieja”, stale reading) at 3 min. The thresholds live in live.py (TELE_AMBER_S, TELE_RED_S) and travel in live.json (tele_umbrales), so the page and the tests use the same numbers.
• Each curve states its window and its step: últimos 1 h · 1 punto / 11 s. The step is measured as the median of the gaps between points in the ring, not written in by hand. Measured: 11 s (the 10 s cycle plus what the reading takes).
• Next to the selector: “the gauges are always now; the selector only changes the curves”.
• Processes: Procesos · top 12 · muestra de hace N s.
• On the node card there are two times, each with its content: Instantánea (discos, memoria, montajes) hace 4 min (the slow one, every 5 min) and the chip telemetría hace 9 s (the fast one).

(capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones)

2. Three things that did not add up

2.1 The header did not count the external jobs

Lab renders and adopted processes are listed by the planner, which runs after the header is built. live.count_externos() adds them at the end of the cycle and the header breaks them down:

(capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones)

2.2 Última señal hace — on an external job

A lab job has no heartbeat of its own. Its signal is the answer to GET /jobs, with the time of the poll. Progress does not come in /jobs: it is in each render’s log (logs/lab_jobs/<id>.log), on a line like 50/60 frames · … · ETA 42 s. collect.py reads it with --lab-logs; the folder is set in config/scheduler.toml. An adopted process takes the node’s process reading as its signal.

Real render launched for the test (POST /jobs/<id>/restart on the lab on the Mac):

4 min de render, 83 % (50/60 cuadros, faltan ~42 s)
última señal hace 3 s (GET /jobs)

(capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones)

The render was 60 frames with a line every 50, so there was only a single progress datum, 40 s before the end. A single-image render does not write that line and shows only the time.

2.3 Network and disk I/O on the PC; interface on the Mac

On the PC, Get-Counter and typeperf took the same time (~1.7 s: the sample is 1 s). typeperf was chosen: it does not start PowerShell and gives input and output separately.

PS> typeperf -sc 1 "\Network Interface(*)\Bytes Received/sec" "\Network Interface(*)\Bytes Sent/sec" "\PhysicalDisk(_Total)\Disk Bytes/sec"
"(PDH-CSV 4.0)","\\<equipo>\Network Interface(Intel[R] Ethernet Connection [2] I219-V)\Bytes Received/sec",...
"10/11/2026 07:12:31.415","74044.118943","0.000000","26028.697621","0.000000","524176.880953"

It runs in parallel with the CPU sample. The PC reading goes from ~0.25 s to ~1.35 s per 10 s cycle. The machine name from the CSV is not stored.

On the Mac, netstat -ibn per interface gives the deltas between readings, and networksetup -listallhardwareports says what each one is (en0 → Ethernet). The interface that took in the most bytes yields entra por en0 (Ethernet), the way to see that traffic goes over the cable (tutorial 29 §8). Disk comes from deltas of iostat -Id, which is cumulative and does not wait.

(capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones) (capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones)

Failure found along the way: parse_netstat read shifted columns on interfaces with no address (gif0, a VPN’s utun): they have one column fewer. With the VPN on, its traffic would have come out wrong. It now has a test.

3. Privacy: the cause without the username

The Attention table showed a cause with the full path /Users/<usuario>/.enjambre/catalogo/backup-plus-manifest.tsv. razones.home_paths() turns /Users/<u>/ into ~/ and C:\Users\<u>\ into ~\, also with JSON’s doubled backslashes. It is applied in four places:

• when writing a new cause (_clip);
• when showing the old ledger rows (build.alerts), because the ledger already written is not rewritten;
• in the page’s deid(), for the history and the details;
• when publishing (web.clean_why / clean_log).

On the web, that cause still comes out as exit 1 (detalle solo en local): ~/ still counts as a path for the leak check.

(capture of the real setup: not published; see docs/CAPTURAS.md for the demo-mode ones)

How to verify (4. Check)

PS> python -m pytest -q tests/test_visor_pulido.py
19 passed

The prompt’s fixtures: telemetria.at from 7 min ago → crit; an adopted process with no heartbeat takes the time of the process reading (lista de procesos); the exact ledger line with /Users/<u>/ comes out with ~/.

5. What is left

• The local viewer’s Projects tab shows the name of the Mac’s home folder as a project (sessions opened in ~). It is not a cause (outside §3) and the name is stored in history.db. Changing project_of would split the history into two groups. Proposal: rename it when displaying. The web copy already publishes projects by alias only.
• In the PC detail, the thermal gauge says (sin sudo), which is the Mac’s reason, not the PC’s. It predates this work; it was left alone.

What actually went wrong

• The header did not count lab renders or adopted processes (§2.1).
• A lab external job showed Última señal hace — because it has no heartbeat of its own (§2.2).
• parse_netstat read shifted columns on interfaces with no address (gif0, a VPN’s utun); it now has a test (§2.3).
• The Attention table showed the username in a full path (§3).

Self-check

1. Which field is the gauge’s “N s ago” computed from? Answer: from telemetria.at, the time of the measurement, not that of live.json.
2. Where do the signal and progress of a lab render come from? Answer: the signal is the answer to GET /jobs; the progress, from the log line in logs/lab_jobs/<id>.log.
3. Why is home_paths also applied when showing the old ledger rows? Answer: because the ledger already written is not rewritten.