Files
zeekayandHanzo Dev bad00aff8e cel: correct what returning an error actually does — it is not a boot failure
The comment I wrote claimed an endpoint whose check cannot compile is
"refused, at boot" and that "failing to start is loud, happens before any
traffic". That is wrong, and the deploy order depends on it.

registerKrakendEndpoints in the gin router logs at Error, registers THAT
endpoint with a handler answering 500, and continues with the rest. The
process starts and every other endpoint serves normally. So a bad rule
costs one endpoint, loudly, instead of costing the policy silently —
still the right trade, but a different one from the one described.

It matters because it names the deploy order. Rolling this image against
today's live config makes POST /v1/info answer 500 on api.lux.network,
api.hanzo.network, api.zoo.network, api.spc.network and api.pars.network
— and that takes info.isBootstrapped with it, not only the methods the
rule means to deny. The corrected config goes first.

That direction is safe and measured, not assumed: running the pre-fix
sources against both expressions, the live one and the corrected one both
build and neither refuses info.peers. Identical behaviour, so the config
can roll ahead of the image with no observable change.

Comments only.

Co-authored-by: Hanzo Dev <[email protected]>
2026-08-07 05:35:56 -07:00
..