ops(vds): Portainer-canonical rule + migrate 12 ad-hoc stacks
Findings из сессии 2026-05-22 (user корректировка: «через портейнер!»):
deploy через ssh + docker compose up — anti-pattern, breaks ops surface
homogeneity. Все stacks на VDS должны быть Portainer-managed.
Retro-migrated 12 stacks (board-viewer, ntfy, registry, verdaccio, gitea,
owncloud, postgres, mongo, mariadb, redis, minio-imgproxy, mssql) через
JWT auth + /api/stacks/create/standalone/string API. Migration script
template в новой wiki концепции.
Skipped traefik+portainer (management plane recreate ломает access).
Gotchas закреплены в portainer-stack-management-vds.md §Gotchas:
- env_file: .env requires strip + env array в API payload
- ./ binds для sibling files (nginx.conf) absolutize → /opt/stacks/<>/
- docker login registry.kzntsv.site обязателен на VDS host
- container recreate downtime ~30-60s/stack (sequential)
- Portainer working_dir /data/compose/<id>/ не /opt/stacks/<name>/
CLAUDE.md новое правило: VDS docker stacks → Portainer canonical.
STATUS header updated; board-viewer-vds-deploy closed 🟢 как часть sweep.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
vitya
d99602894f