bunker-admin 5331cdcc67 fix(approach-c): full E2E success on marcelle - byte-identical templates + core-only recreate
This session completed Approach C end-to-end on marcelle (status=COMPLETED,
mkdocs untouched, idempotent on re-run). Four fixes landed:

1. template-engine.ts: dropped nginx/conf.d/*.hbs (default, api, services)
   from renderAllTemplates AND renderAllTemplatesInMemory. The new
   prod-style docker-compose.yml.hbs does NOT mount conf.d/ into the
   nginx container ("Note: conf.d is NOT mounted (configs are generated
   at startup from templates)" — nginx confs are baked into the nginx
   Docker image). Writing them was a no-op orphan that showed up as 3
   "modified" lines in preview unnecessarily.
   Same reason removed nginx/nginx.conf from staticFiles.

2. templates/configs/{pangolin/resources.yml,prometheus/prometheus.yml,
   grafana/datasources/datasources.yml}.hbs: synced byte-identical to
   canonical changemaker.lite/configs/*. These ARE mounted into pangolin
   tunnel + prometheus + grafana respectively. Preview now reports
   "unchanged" for them on install.sh tenants.

3. templates/docker-compose.yml.hbs: dropped the CCP-tenant header
   comment, making the template now BYTE-IDENTICAL (58907 bytes) to
   canonical changemaker.lite/docker-compose.prod.yml. Even a 1-byte
   comment difference caused docker compose to compute new config hashes
   for every service, triggering full-stack recreates (including
   ccp-agent — the Phase 6 self-destruct trap from upgrade.sh).

4. upgrade.service.ts:runReleaseUpgrade — composeUp now restricted to
   core app services [api, admin, media-api, nginx] (same set as
   image-upgrade.sh). Unscoped composeUp would recreate ccp-agent
   mid-apply and orphan the runner. Until Approach C inherits the
   deferred-ccp-agent-restart pattern from upgrade.sh, this restriction
   keeps the apply path safe. Limitation: brand-new services in a
   release won't auto-deploy via Approach C alone — operator must
   follow with Approach A (full upgrade.sh) to pick them up.

E2E verification on marcelle:
  - Apply: status=COMPLETED, duration<10s.
  - mkdocs.yml md5 unchanged (38810d9df8b4258ad46a6739232cf88a).
  - mkdocs/docs file count unchanged (242).
  - docker-compose.yml now byte-identical to canonical (58907 bytes).
  - app + api public sites: 200 both.
  - Re-preview: ALL 10 files show "unchanged" — true idempotency.

Phase 6 acceptance gate met. Approach C now fully operational on the
install.sh fleet.

Bunker Admin
2026-05-23 11:00:38 -06:00

62 lines
1.4 KiB
Handlebars

global:
scrape_interval: 15s
evaluation_interval: 15s
external_labels:
monitor: 'changemaker-lite'
# Alertmanager configuration
alerting:
alertmanagers:
- static_configs:
- targets: ['alertmanager:9093']
# Load rules once and periodically evaluate them
rule_files:
- "alerts.yml"
# Scrape configurations
scrape_configs:
# V2 Unified API Metrics
- job_name: 'changemaker-v2-api'
static_configs:
- targets: ['changemaker-v2-api:4000']
metrics_path: '/api/metrics/internal'
scrape_interval: 10s
scrape_timeout: 5s
# N8N Metrics (if available)
- job_name: 'n8n'
static_configs:
- targets: ['n8n-changemaker:5678']
metrics_path: '/metrics'
scrape_interval: 30s
# Redis Metrics
- job_name: 'redis'
static_configs:
- targets: ['redis-exporter:9121']
scrape_interval: 15s
# cAdvisor - Docker container metrics
- job_name: 'cadvisor'
static_configs:
- targets: ['cadvisor:8080']
scrape_interval: 15s
# Node Exporter - System metrics
- job_name: 'node'
static_configs:
- targets: ['node-exporter:9100']
scrape_interval: 15s
# Prometheus self-monitoring
- job_name: 'prometheus'
static_configs:
- targets: ['localhost:9090']
# Alertmanager monitoring
- job_name: 'alertmanager'
static_configs:
- targets: ['alertmanager:9093']
scrape_interval: 30s