No description
- Go 84.8%
- Scala 10.4%
- Nix 4.5%
- Shell 0.3%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
Børge Lundsaunet
016b6b11cd
The public secondary never caught up with the portal's zone. Two independent breaks, both found against Knot 3.5.10's source: 1. The NOTIFY was not a NOTIFY. It went out with the query opcode and qtype NOTIFY (31), so Knot routed it as an ordinary query and answered NODATA — the notify path was never reached. It is now sent the RFC 1996 way: opcode NOTIFY, a single SOA question named after the zone. 2. The TCP server handled one query per connection, but Knot runs a whole refresh on one connection: the SOA check first, then the zone transfer behind it. The portal closed the socket after the SOA answer, so every transfer died with "connection reset" and only the 300 s poll backstop remained — which hit the same wall. Connections now serve queries until the client closes them. (The secondary also needed automatic-acl in its knot.conf to admit the notify at all; that is a fleet change, landing separately.) Proven against a real knotd: with the fixes, an admin-triggered zone change lands in the secondary's zone file within the same second — notify accepted, refresh, zone file updated. 🤖 Generated with [ECA](https://eca.dev) (jokull/rotom) Co-Authored-By: eca-agent <git@eca.dev> |
||
| .forgejo/workflows | ||
| agent | ||
| android | ||
| channel | ||
| cmd | ||
| docs | ||
| front/core | ||
| nix | ||
| portal | ||
| proto | ||
| sniff | ||
| splice | ||
| .envrc | ||
| .gitignore | ||
| flake.lock | ||
| flake.nix | ||
| go.mod | ||
| go.sum | ||
| purity_test.go | ||
| SKILLS.md | ||