20260914 - Point the ADS-B truth feed at the node's own tar1090

[Purple10101]

What

Changes the shipped truth.adsb.tar1090 default from sfo1.retnode.com to localhost:8078, and adds a migration so the change actually reaches deployed nodes.

Why

The default named one machine that every node read its truth from, and it has stopped answering. sfo1.retnode.com returns HTTP 530 from every node checked and from off-estate. Every node that never had this set by hand has been labelling nothing.

Four nodes, three different values, one working:

Node	truth.adsb.tar1090	State
jonathan-node-1	localhost:8078	working, 240 labelled events across 25 aircraft
owl-ded9	sfo1.retnode.com (inherited)	530, blind
owl-debb	sfo1.retnode.com (in user.yml)	530, blind
fairforest B	192.168.1.143:30005	BEAST port, parse error every cycle

jonathan-node-1 was fixed by hand in its own user.yml some time after 2026-07-16 and the value never came back into the repo. Every broken node already had a working tar1090 container on localhost:8078 that nothing was reading.

Why localhost:8078 is the right default

It is node-invariant, which is what makes it defaultable at all. Every node runs its own tar1090 container on that port, fed by adsb.lol and by a local receiver wherever tar1090.adsb_source names one. The genuinely per-site part stays in adsb_source, so attaching a receiver is three fields in one place and never involves the truth feed.

That separation is also the fix for how fairforest B went wrong: the receiver's BEAST address was put in the truth field. A raw feed port is not an HTTP endpoint, so fetchADSB() logged Parse Error: Expected HTTP/, RTSP/ or ICE/ every cycle for 15 hours and every detection's adsb came back null.

Why the migration is needed

Changing default.yml alone reaches no deployed node. First boot seeds user.yml with a whole copy of the defaults, so every node persists whatever shipped the day it booted and that overlay shadows the new value.

migrate_adsb_truth_server() follows migrate_doppler_span() and migrate_tracker_forward() exactly: drop the key when it still holds the legacy host, so the shipped default applies. Safe on the same reasoning, and more clearly so than either, since nothing is lost by moving a node off a host that answers 530.

Any other value is a real choice and survives untouched, including a receiver address like fairforest B's, which is wrong in a way this cannot safely judge. forced.yml still wins, and user.yml on disk is not rewritten, so the migration stays load-bearing rather than being a one-shot fixup.

Verification

As lint.yml runs it:

• ruff check . clean
• pytest config-merger/test/ 47 passed
• vulture clean

8 new tests. Disabling the migration fails 2 of them; the other 6 are guards asserting values are kept (a deliberate host, a receiver address, the legacy host with a port appended, forced.yml precedence, and user.yml left unrewritten).

Verified on hardware

fairforest B was corrected by hand ahead of this change and the pipeline was traced end to end: local receiver → tar1090 (messages 0 → 4009), tar1090 → blah2-api (parse errors stopped), blah2-api → adsb2dd (aircraft converted to bistatic delay and Doppler). Association was still unproven at 04:00 local only because no aircraft had yet been inside both the delay and ±200 Hz Doppler windows. jonathan-node-1 proves that final link on identical software.

🤖 Generated with Claude Code