Say which kind of empty the Live page is

[blessdyb]

Shipping as v0.5.14. Reported from a desktop: "I see nothing in live even I refreshed firefox".

What was actually wrong

Nothing, in the window, the daemon or the probes. Payload capture had run out — its eight-hour session had ended — so the daemon was reading nothing. The Live page said:

Nothing read in this window
That is not the same as nothing happening. Coverage says what could not be read.

which reads as your machine was quiet when it means Flowlight stopped reading. Confirmed on the machine:

$ flowlightd budget
Payload capture has run out and is no longer reading anything. Renew it to start again.

$ journalctl -u flowlightd | grep 'run out'
12:09:36 payload capture has run out and has stopped. Connections are still attributed.

The daemon knew, logged it, and would say it on demand. The window never asked. The handshake it does ask carries storing, which is built once at startup and hard-coded true, so it cannot express something that becomes true later.

What changed

• The window asks whether payloads are being read on every tick, on whichever page is open — the same two seconds as everything else. (It was throttled to ten seconds at first; a live view that takes ten seconds to admit it stopped being live is this same bug, quieter.)
• A banner says so and carries a Renew button, because somebody reading "renew it" wants to renew it. It is a second banner so it cannot fight with the one the handshake owns.
• The empty page names which of the three silences it is: switched off, run out, or genuinely quiet.

Verified on the machine first

A controlled expiry, rather than a guess:

live session        → captured
after expiry        → not captured
renewed, no restart → captured      (within the 2s budget poll)

That last line also retired a bug I thought I had: --renew works without a restart. My first test made the request one second after renewing and lost the race to the poll.

Reading::of is the mapping, extracted so it can be tested without a display. The tests were run against the old behaviour — one page for every silence — and two of them failed there before being trusted here.

Worth knowing

The session is wall-clock. The machine this came from had jumped forward ~6.6 hours (NTP inactive, RTC disagreeing), which ended an eight-hour session early and also parked every stored request outside the "Last hour" view.

🤖 Generated with Claude Code