Skip to content

armgit / avibe

Public Imported from GitHub
0

Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.

2.8 KB
Raw Blame

Incremental Regression Updates

Goal

Reduce repeated writes and contention on the local Incus VM disk without moving storage, changing resource limits, resetting product state, or restarting the host Avibe service.

Design

  • Use the existing base image's rsync through the Incus exec transport. Sync changed source files and delete stale source, preserving excluded dependency and runtime trees. Preserve executable modes and symlinks, but create files as the service user rather than importing host ownership. Use rsync's normal size/mtime quick check, not a full destination-content checksum pass.
  • Reuse the Show Runtime archive only when its receipt matches the resolved upstream commit, target Node/platform and npm versions, build recipe, and archive checksum. Build the exact resolved commit; atomically publish the archive and publish its receipt last. Keep one artifact, not an unbounded cache. Always run the existing runtime preparation/validation path.
  • Reuse the runner's file-lock mechanism for a shared heavy-I/O slot across environments managed by the same primary checkout and daemon. Acquire it before instance creation or stopping the target service, and hold it through update/recovery. Base-image builds use the same slot. Other services remain online; updates using older runners do not participate until updated.

Invariants and Verification

  • Repeating an unchanged sync preserves file inodes and timestamps. A changed source tree converges including removals, modes, links, and directory/file transitions; excluded trees are neither copied nor descended into.
  • Reuse requires both matching build inputs and matching artifact bytes. Failed builds do not mark an artifact as current. Upstream changes between resolution and fetch cannot silently select a different revision.
  • Different environments sharing the I/O slot cannot run heavy update phases concurrently. Queued updates keep their existing service running. Exceptions release locks after recovery; dry runs do not acquire locks.
  • Run focused unit/real-filesystem tests, then local Incus repeated updates and health checks with existing state and pairing preserved. Verify source sync statistics and Show Runtime reuse output; do not invent performance claims from tests alone.

Status

  • Inspect update lifecycle, existing exclusions, fingerprints, and locks.
  • Implement the three optimizations and focused coverage (378 focused tests; real macOS-to-Incus fixture: unchanged sync transfers zero file bytes, a single edit transfers one file, Linux unchanged inode/mtime/ctime and ownership stay intact, and obsolete source is removed).
  • Pass Codex review and expected CI through the managed delivery loop.
  • Merge and verify local Incus update/reuse and Models health.