Live observe session verified SSH access, captured baseline and Critical health/load findings (memory pressure, k3s API unavailable). RU checklist closed; engagement phase operating; schedule enabled; no privileged changes.
4.2 KiB
4.2 KiB
| agent | engagement_id | project | last_updated | session_count | confidentiality |
|---|---|---|---|---|---|
| host-operator | eng-coulomb-railiance01-ho-001 | coulomb-railiance01 | 2026-07-16 | 3 | client_owned |
Host Operator Memory — railiance01 pilot
Engagement Charter
- Duty: Keep railiance01 operational and secure; OS currency; load and workload review
- Cadence: daily health/load (business days); weekly OS/security pass
- Change windows: prefer low-traffic periods; reboot only with human approval
- Escalation: coulomb / railiance human operator (Bernd / on-call as designated)
- Out of scope: app features; other hosts; secret vending
- Access:
ssh railiance01as tegwick (host_observe verified 2026-07-16)
Project Context
Coulomb ecosystem production host railiance01 runs k3s and platform workloads
(including Forgejo/apps per railiance-hosts docs). This engagement is the first
forward-deployed host-operator pilot (KAIZEN-WP-0009).
Host Profiles
| hostname | role | typical load | services | last review |
|---|---|---|---|---|
| railiance01 | k3s production single-node | overload on 2 cores under current mix | k3s, gitea/forgejo, temporal, activity-core, state-hub edge, traefik, platform pods | 2026-07-16 |
OS & Patch State
- Ubuntu 24.04.3 LTS (noble); kernel 6.8.0-87-generic (needrestart KSTA=1 — kernel current)
- Many packages upgradable (security-relevant: bind9-*, curl, ca-certificates, dpkg, …)
unattended-upgrades.serviceflagged by needrestart for restart- No package upgrades applied this session (gated)
Security Posture
- UFW active: default deny in; allow OpenSSH 22, k3s API 6443/tcp Anywhere, flannel 8472/udp Anywhere
- Listeners include SSH, k3s components, localhost bridge ports (18000/18001), gitea stack processes
- Journal noise: reverse-forward port 18765 already in use (tunnel contention)
- Public 6443 exposure: policy review recommended (firewall_change gated)
- First snapshot: 2026-07-16 (see health report)
Load & Workload Envelope
| Field | Value (2026-07-16 sample) |
|---|---|
| Cores | 2 |
| Load 1/5/15 | ~12 / 11 / 16 (saturated) |
| RAM | 3.8 GiB; no swap; MemAvailable often <0.5 GiB |
| PSI memory full | avg60 ~24%, avg300 ~29% |
| Disk / | 54% of 96G |
| /var/log | ~5.7G (journal ~4.1G) |
| Heavy workloads | k3s server, gitea, temporal-server, activity-core worker/API, state-hub edge uvicorn, traefik, coredns |
| Saturation incident | 2026-07-16 — memory+load critical; k3s API ServiceUnavailable |
Envelope intent: treat load > 4 sustained or MemAvailable < 300Mi as Watch/Critical; escalate capacity.
Accumulated Findings
- Undersized RAM for workload density is the dominant risk
- k3s API instability under memory pressure
- Large journald footprint
- Pending OS security updates
What Worked
- Observe path via
ssh railiance01(tegwick + id_custodian_agent) - passwordless
sudo -nfor read-only needrestart/ufw/k3s inspect - Engagement vault + close-session for durable evidence
Watch Points
- Single-node production: privileged mistakes have full blast radius
- DinD / Actions runner privilege model (railiance-hosts ADR-004)
- Disk growth from images, logs, and backups
- Do not schedule heavy CI/agents until memory recovers
- Bridge reverse-forward port collisions (18765)
Recurring Findings
- Memory pressure / no swap · first seen 2026-07-16 · frequency 1
- Load ≫ cores · first seen 2026-07-16 · frequency 1
Cleared Issues
None yet (no remediation applied).
Open Threads
- Human decision: add RAM and/or temporary swap
- Human approval: journal vacuum; security package upgrades; UFW source restriction for 6443
- Re-check k3s API health after memory improves
- Align inventory hostname with kernel hostname / DNS naming
Session Log
- 2026-07-16 · railiance01 · standard_review · T09 wire-up smoke: prepare+close-session path verified (no host access) · ok
- 2026-07-16 · railiance01 · first live observe: Critical memory/load; k3s API unavailable; RU checklist complete · ok
- 2026-07-16 · railiance01 · deep_assessment · T10 ramp-up complete: Critical memory/load, k3s API unavailable, RU all done, phase operating · ok