[statebyte]

📋 Summary

Adds the daemon's local database: an SQLite index of installed apps (<data_dir>/index.db) that mirrors everything in every app's .asc/ and stays in sync with the files. asc ls, ListApps/GetStatus and the other app listings read from it instead of re-reading every meta.json and running git rev-parse per app, and the CLI checks the daemon with a light GET /v1/health instead of a full app listing.

🔗 Related task / issue

• DMN-160, DMN-161, DMN-163, and the batched-observation part of DMN-162 — design: daemon-app-index
• Docs site: docs(cli): document the daemon's local database and asc db asc-documentation#2 · roadmap and platform side: AdminServiceCloud/asc-platform#132

🧩 Type of change

• ✨ Feature (feat)
• 🐛 Bug fix (fix)
• 📝 Documentation (docs)
• ♻️ Refactor (refactor)
• 🧪 Tests (test)
• 🔧 Chore / tooling (chore)

🗂️ Affected component

App management, API (REST), CLI, Scheduler

📝 What changed

• src/daemon/apps/index/ — AppIndex over rusqlite (bundled, WAL, busy_timeout, file created 0600 before SQLite writes to it):
  • schema v1: apps (meta split into columns + meta_json, commit, .git stamp, last observed state) and app_files (content of every top-level .asc/ file with an inode:size:mtime_ns:ctime_ns stamp); not mirrored: op.lock, *.tmp, symlinks, sub-directories; files over 1 MiB keep no content;
  • sync_app reads the files inside BEGIN IMMEDIATE, so concurrent writers (two threads, or the CLI and the daemon) converge on the file;
  • verify — stat-only comparison with the files; rebuild — full scan (same rules as the old listing, including layout::migrate);
  • no database / newer or broken schema / changed apps_dir → rebuild; an unusable file is moved aside as index.db.broken-<ts>;
  • process-wide registry: every AppStore of the same root shares one index; one database file serves one apps root per process; ASC_NO_INDEX=1 keeps the old scan.
• Sync with the files: write-through from AppStore::save/remove, SettingValues::save and the compose override writer; inotify watcher (Linux, non-recursive watches on the apps root, app dirs, .asc/, repository/.git, 200 ms debounce, overflow → verify); verify every minute in the daemon and before reads in a process that does not maintain the index (the CLI without a daemon).
• Reads: AppStore::list, name resolution in get_authorized (the match is then loaded from its file), rename, reconcile, prune protection, ownership maps, declared ports and the backup scheduler (AppStore::settings from the mirrored settings.json); commits from the index instead of git rev-parse.
• Batched observation (AppManager::observe): one Engine container list for all docker/compose apps, one systemctl show -p ActiveState for all systemd units (falls back to is-active per unit), pid check for process apps.
• API: GET /v1/health; root-only GET /v1/db, POST /v1/db/verify, POST /v1/db/rebuild. The CLI probes /v1/health and falls back to /v1/status on older daemons (404).
• CLI: asc db status|verify|rebuild (EN + RU messages). Named db, not index: commands/index on the docs site is the command overview page.
• Docs: docs/english|russian/app-management.md — new "🗃️ Local database" section.

✅ Checklist

• The change starts from a task, and the relevant doc in docs/ is updated (docs first, then code — see AGENTS.md).
• task check passes locally (clippy with -D warnings, cargo fmt, tests) — run in WSL: 731 unit tests + all integration tests.
• New logic has unit tests (index, scan, schema, watcher, systemd batch parser) and integration tests (/v1/health, /v1/db, root-only refusal over the socket).
• Commits follow Conventional Commits.
• Does not break builds for x86_64 / ARM64 / ARMv7 — rusqlite bundled compiles SQLite with the target C toolchain cross already provides for ring; the cross builds themselves were not run locally (release CI does).
• User-facing CLI strings go through the translation system (EN + RU keys); debug/log messages are English-only.
• proto/ changes are backward compatible — no proto changes.

🔎 How to test

task test
sudo asc ls
sudo asc db status
echo '{"MODE":"dev"}' | sudo tee /asc/apps/<id>/.asc/settings.json
sudo asc db verify

Measured on 50 process apps with git checkouts (debug build, CLI without a daemon): asc ls ~15 ms with the index vs ~80 ms with ASC_NO_INDEX=1. In a container running asc serve: index.db, -wal, -shm are 0600, an app directory written by hand shows up in asc ls within a second (watcher), asc db verify picks up a hand-edited settings.json and a removed app.

📌 Notes for reviewers

• The files stay the source of truth; the database is always rebuildable and deleting index.db is safe at any time.
• Still open in DMN-162: Docker Engine events feeding observed_state, a shared Docker client per process, asc ls --live.
• Pre-existing, not touched here: CLI requests over the socket fail for app names containing spaces (invalid uri character) — paths are not percent-encoded in src/daemon/client.rs.
• No version bump: that happens at release time per AGENTS.md.

🤖 Generated with Claude Code