REPORT — attended release run 2026-10-24: the hub's tokens locked, the hub out of the open backup, the iperf3 line test, the watch released

Brief: „the hub's tokens locked, the hub out of the open backup, an iperf3 line test via ep0, then the upload watch released" (operator rulings 07:22 / 07:26, CONTEXT.md). Read first: 05 §16.2, 06 §4.5 / §5, 07, 08 §9, the spike audits/SPIKE-upload-watch-2026-10-23.md, the felhom-build-deploy skill. Evidence: audits/line-test-2026-10-24/.

Five lines

Hub tokens: sealed in hub 0.168.0 and still working (demo-hp is served the same tunnel and API tokens it runs; the template fetch works; the next ep0 copy holds 0 of 8 token values). The live file keeps old bytes in free space — your call: accepted, written down, with the rule „copy hub.db only with VACUUM INTO".
The hub is out of the open backup: the hub volume reads default: disabled; the folder holds no copy of it (your two deletions worked; nothing else is left). Tonight's job is the last proof (dated check 2026-10-25).
The line test: demo-hp measured 94.4 Mbit/s (Tester 1 94.3, demo-felhom 92.6) — the night backup said „at least 10".
Live where: hub 0.169.0 and controller 0.342.0 run on all three boxes (the global floor); agent 0.164.0 on all three (vouched); demo-felhom's qBittorrent is limited to 46 Mbit/s. Only these three boxes report to the hub.
You still do: nothing for this run. Two rows wait for a design (below).

What I asked you, and what you said

Deploy hub 0.168.0 → yes. Deploy hub 0.169.0 → yes. Global floor 0.342.0 → yes. Agent vouch → done.
The live file's old bytes → accept and write down (my first estimate of „2 minutes" was wrong; the disk writes ~1.3 MB/s, so a clean-up means an 8–20 minute hub stop).
No host has the iperf3 program → the agent speaks iperf3 itself (nothing installed on hosts).

Versions, digests, CI (by commit)

Repo	Commit	What	CI
felhom.eu	`00031150`	hub 0.168.0 manifest + rulings	2275 success
felhom.eu	`3363ffa8`	D40: the PVC label	2276 success
felhom-agent	`f7e1dcb`	the line test	2277 success
felhom-controller	`ed1011b`	the tested line in the watch	2278 success
felhom.eu	`1a788f33`	docs, register, ep0 unit, hub `line_test` model	2279 success
felhom-controller	`42cda07`	the start-line wording	2283 success
felhom-agent	`08ca678`	CHANGELOG released, REPORT	(docs)
felhom.eu	`01bf4e83`	hub 0.169.0 manifest	(see the final push)
felhom-controller	`bddc477`	CHANGELOG released, REPORT	(docs)

Hub 0.168.0 sha256:3bfdecaa… (from 5277d859), deployed 05:39Z. Hub 0.169.0 sha256:fb9b2489… (from 1a788f33), deployed 07:51Z; 30-minute watch: only info events, 0 errors.
Agent 0.164.0: binary 1b803d8b70f8afbe1546f6989f35b6b0e29c9ed63f6acb3843d3c04ef0d24b2b, bundle 1d7b54d39d0c1aff682ba0c0a7f91aea0f9407a10d2f5a8bbfdc858d32c13341, tag v0.164.0; signed agent_update to the three boxes, self-check 76/76; vouched (hub page + Artifact manifest set line).
Controller 0.342.0 sha256:d002d3b7… (from 42cda07); rc.2 on 9202; per-box floors → healthy on all three, box checks and https-only green; global floor 08:29:19Z, MinAgent 0.158.1, read back on the hub log and page.
ep0: felhom-iperf3.service (iperf3 3.18-2+deb13u2, already installed since 2026-07-03), one nftables line; no answer on the internet (:22 as the control), an answer in the tunnel. CVEs: four fixed in this build; two open „minor" server DoS issues bounded by tunnel-only, TCP-only and systemd caps (scripts/ep0-iperf3/README.md).

Rows

Before 194, after 191; opened 2 (R-1249, R-1250), closed 5 (R-975, R-1244, R-1245, R-1246, R-1247). R-1243 → VERIFY (dated check 2026-10-25); R-1241 keeps only the homelab half.

Observations

The start-up VACUUM after a legacy seal cannot finish on the Longhorn disk; its comment said 2.8 s — FILED: R-1249.
A qBittorrent that seeds all day keeps the uplink busy, so that box is never tested and gets no limit — FILED: R-1250.
The live hub file's old token bytes — NOT-A-FINDING: accepted by the operator 2026-10-24 (05 §16.2), every copy that leaves is clean.
Two pre-deploy hub snapshots in /data/snapshots held plain tokens — NOT-A-FINDING: they roll off (one already pruned at 06:06Z; the other at the 12:00Z snapshot); same volume, same reader as above.
The cross-repo wire gate refused the agent's new line_test field until the hub modelled it — fixed in the session (hub 0.169.0 sysfacts.LineTest, the System page cell).
My first ArgoCD sync used a wrong commit id — NOT-A-FINDING: it failed to fetch and changed nothing; redone.
9202 has no agent local API, so its controller cannot read a line test — NOT-A-FINDING: by design (scratch guest); the qBittorrent proof moved to demo-felhom, where the product path runs.
demo-hp's first watch minute could not resolve plex — NOT-A-FINDING: the known start-up race (2026-10-23).
The family's line on demo-felhom's qBittorrent page was not seen live — NOT-A-FINDING: the household's password is not CC's; the render is pinned by a test, and the text with the live value is in the evidence.
Old rc controller images (other sessions) sit on 9202 — NOT-A-FINDING: a scratch guest; mine (rc.2) is removed.
The hub's database is 530 MB, mostly report history (host_reports 305 MB, reports 128 MB) — NOT-A-FINDING here: it is why a VACUUM is slow, recorded in R-1249.

Teardown, three layers

Machine: 9202 — the throwaway qBittorrent removed through the product (0 containers, 0 volumes), back on the 0.342.0 release, my rc.2 image removed. demo-hp's host — the probe binary deleted, nothing running.
Host / DooPlex: the database copies I restored for the checks shredded; the 9202 password copy, the signed job envelopes and the release clone shredded / removed. ep0 — the one planned change stays (felhom-iperf3); the backup of the firewall file is /etc/nftables.conf.bak-20261024-iperf3.
Hub: the three per-box floors cleared; the global floor 0.342.0 stays (the release).

Instruction files, poster

No instruction file edited. System poster: a text-only change (ep0 also runs the tunnel-only line-test server) — the facts file and the matching poster text edited in the same commit.

6.4 KiB Raw Blame History Unescape Escape