Map and note: Odin folded back to 17871, coordinator retired, four defects fixed on the way
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
parent
b8ef6c62ae
commit
1602225fe4
2 changed files with 37 additions and 7 deletions
|
|
@ -177,3 +177,33 @@ next Muninn redeploy, which is why it is not done here.
|
|||
zero odin-daemon errors since the release, and Idunn has logged nothing
|
||||
since the 21:58:17Z sweep. The old kill window was 2.5–16 minutes. Odin is
|
||||
up under the incarnation-keyed projection.
|
||||
|
||||
## Fold-back to 17871, 2026-09-10 22:06–22:45Z
|
||||
|
||||
Odin was admitted but unreachable: its stable endpoint was `10.77.0.1:17971`,
|
||||
a staging port chosen while the old Node coordinator held `17871`, and every
|
||||
consumer, runbook and firewall rule still named `17871`. Operator ruled: fold
|
||||
back. Done end to end, with four defects found and fixed on the way, each
|
||||
committed separately:
|
||||
|
||||
- The coordinator container `odin-coordinator-1` is stopped with
|
||||
`restart=no`; `odin.service` now starts only `hermodr`. Binding
|
||||
`stable_endpoint` is `rudp://10.77.0.1:17871`; backups of the binding and
|
||||
the unit are under `/tmp/projsnap/` on the host.
|
||||
- Idunn `f16fb46`: the route driver owns one ufw allow per routed stable
|
||||
endpoint (needs `/etc/ufw` in the unit's `ReadWritePaths`, `103c557`).
|
||||
- Idunn `e81c73e`: Warming admitted the candidate's correlation twice with one
|
||||
stale record and failed the second time. Once now.
|
||||
- Idunn `692ba1f`: an aged-out Odin correlation is absence to wait on, not a
|
||||
fault. This is what had failed every Heimdall deployment.
|
||||
- Odin `b8ef6c6`: the daemon is PID 1 of its private PID namespace and ignored
|
||||
SIGTERM, so every stop was a 90 s SIGKILL and the candidate's 60 s bootstrap
|
||||
wait expired during fencing. Odin catches SIGTERM now; the wait is 300 s.
|
||||
- Admitted at 22:44:21Z as `up-e728909f` / `tx-0e8fa81e`, release
|
||||
`sha256-4d6e959a…`, listening on `10.77.0.1:17871` → `127.0.0.1:17972`. The
|
||||
driver's stale `17971` allow was deleted by hand; the operator's original
|
||||
`17871` rule remains beside the driver's. Continuity restored the previous
|
||||
generation within 11 s after each failed attempt; Odin was never down for
|
||||
more than about a minute.
|
||||
- Not Odin's: `10.77.0.4` is blocked sending UDP to `10.77.0.1:17870` (Idunn
|
||||
health). Handed to the Muninn session to name.
|
||||
|
|
|
|||
|
|
@ -5,7 +5,7 @@
|
|||
# commands that do so are in AGENTS.md. Where this disagrees with yggdrasil,
|
||||
# yggdrasil is right and this file is a bug.
|
||||
|
||||
verified_at: 2026-09-10T22:06Z
|
||||
verified_at: 2026-09-10T22:45Z
|
||||
verified_against: yggdrasil (systemctl, idunn status)
|
||||
|
||||
deployment_authority:
|
||||
|
|
@ -15,24 +15,24 @@ deployment_authority:
|
|||
source_selection: ref-head on refs/heads/main
|
||||
minimum_revision: 08965a44e135b393a130d62b075dc5036d2df901
|
||||
idunn_minimum: 1a54db1 # incarnation-keyed projection (903e1c8) plus the queue and give-up fixes
|
||||
idunn_live: 1899df2 # 2026-09-10T21:58:17Z; supervise withdraws one stale projected incarnation per tick
|
||||
idunn_live: 692ba1f # route driver owns the stable endpoint allow (f16fb46); warming fixes (e81c73e, 692ba1f)
|
||||
|
||||
runtime:
|
||||
admitted: true
|
||||
admitted_by: up-838e26d8-e997-496c-8356-6074b4aa5461 # tx-dde64dc5, 2026-09-10 21:51:02Z
|
||||
release: sha256-067ab605c82828cd2b1b1feac0d431cc217e076007c28a5dbd2902ae294878e3 # Odin main 27ca3b3
|
||||
incarnation: odin@sha256-e98b0459f230345da7eb85144eac806608bb0e6606d19d5c1f53e1cb6dbaf28a
|
||||
admitted_by: up-e728909f-abee-49e4-9cf3-0f88b24cdea5 # tx-0e8fa81e, 2026-09-10 22:44:21Z
|
||||
release: sha256-4d6e959a655f0e7158d8a520587be10f10dbf7782abdc6092edd4c32fc18e701 # Odin main b8ef6c6
|
||||
incarnation: odin@sha256-903af5f8691a… # full digest in the projection
|
||||
# The live unit is idunn-odin-<activation-hash>.service and that hash changes
|
||||
# on every restart. It is not this release hash. 318 failed unit residues
|
||||
# were loaded on 2026-09-10T19:23Z, all with distinct hashes.
|
||||
state_root: /var/lib/gamecult/odin
|
||||
projection: /var/lib/gamecult/idunn-projection/topology.cc
|
||||
route: /etc/nginx/idunn-stream-routes/odin.conf # rudp://10.77.0.1:17971
|
||||
route: /etc/nginx/idunn-stream-routes/odin.conf # rudp://10.77.0.1:17871 -> 127.0.0.1:17972; the Node coordinator that held 17871 is retired
|
||||
|
||||
# Before Odin 27ca3b3 and Idunn 1899df2, Odin exited 1 after 2.5-16 min per
|
||||
# unit because the queued deployment overwrote the target-keyed Expected slot.
|
||||
# That path no longer exists; see notes/odin-incarnation-keyed-projection-2026-09-10.md.
|
||||
healthy: true # sixteen-minute hold passed 22:06Z: same PID since 21:50:48Z, zero rejections, zero daemon errors
|
||||
healthy: true # admitted 22:44:21Z, zero rejections, mesh reachable on 17871
|
||||
condition: admitted
|
||||
|
||||
# A unit named odin.service exists in /etc/systemd/system and is `active` with
|
||||
|
|
|
|||
Loading…
Reference in a new issue