Skip to content

Latest commit

 

History

History
129 lines (119 loc) · 7.93 KB

features.md

File metadata and controls

129 lines (119 loc) · 7.93 KB

Features

What Kiri supports today, through the kiri Docker context.

Platforms

Apple silicon and Intel Macs (macOS 14+), Windows 10 and 11 (x86-64 and Arm, through WSL 2), and Linux (.deb and .rpm, tested on Ubuntu and Fedora).

What works

  • Containers: run (--rm, -d, -it, -e, -u, -w, --name, --restart, --entrypoint), exec, logs, attach, stop, kill, restart, rm, wait, rename, pause/unpause, update, stats, top, diff, export, commit, cp (both ways, running or not), container prune; records survive a daemon restart; filters on every list
  • Limits and security: --memory, --cpus, --cpu-shares, --pids-limit, --shm-size; Docker's default capabilities and a default seccomp filter; --cap-add/--cap-drop, --privileged, --read-only, --security-opt seccomp=unconfined and no-new-privileges. Even privileged and unconfined containers keep the host boundary: no vsock sockets (the way to the Mac) and no io_uring
  • Networking: a default bridge, user networks with DNS by name and alias, network connect/disconnect, --network host|none, -p/-P (TCP and UDP) on all interfaces (as Docker does; -p 127.0.0.1:8080:80 keeps a port local), --add-host, host.docker.internal, --network container:<name> (Compose network_mode: service:x; as with Docker, if that container restarts, the ones sharing its network keep the old, now empty namespace until they are restarted too), IPAM (--subnet, --gateway, --ip-range, --aux-address), --ip, --mac-address, --dns, --dns-search, --dns-option, --domainname, --link, --ipv6 (Compose enable_ipv6: an IPv6 subnet beside the IPv4 one, from --subnet or a /64 of Kiri's own under fd00::/8; the default bridge stays IPv4-only, as Docker's does), and http://<name>.kiri.local once you turn it on (the app asks on first open; Settings → Domains changes it), or http://<name>.kiri.localhost with no setup. Compose services answer at <service>.<project>.kiri.local, and a dev.kiri.domain=api.myapp label picks a container's own names; HTTPS too (https://<name>.kiri.local), with Kiri's own certificate authority in ~/.kiri/tls, whose name constraints allow only *.kiri.local and *.kiri.localhost (Settings → Domains → HTTPS asks macOS to trust it)
  • Storage: named and anonymous volumes, bind mounts of Mac paths (virtiofs), --tmpfs, --mount, volume commands
  • Images: pull (with docker login credentials), push, tag, save, load, history, search, image prune, system df; amd64 images run through Rosetta, --platform honoured
  • Builds: docker build / buildx build through BuildKit running in a Kiri container (buildx's docker-container driver), stopped when idle; on Apple silicon, amd64 build steps run through Rosetta too
  • Compose: up, down, ps, logs, exec, multiple networks, events
  • USB devices (beta): the USB tab hands a device (a serial adapter, a Zigbee stick, a development board) to your containers on Mac, Windows and Linux, and shows where it is inside Linux with the --device line to copy. On a Mac, a device that drops off the bus while attached (the board resets, or the Mac sleeps) is attached again when it comes back
  • Mac integration: container files at ~/Kiri in Finder (read-only NFS, always current), SSH agent forwarding at /run/host-services/ssh-auth.sock
  • Memory: a balloon gives unused memory back while running, and the VM sleeps after 60 s idle (the daemon stays at ~15 MB; waking takes ~0.15 s)
  • The window: containers, images, volumes and networks, with Compose projects grouped (start, stop, restart or remove a whole project); each container has its own page with details, live logs (find, level filter), a command line (and "Open in Terminal" for a full shell) and its files (preview, Save to Downloads, Show in Finder); Settings for fonts, text size, the VM's CPUs and memory (~/.kiri/vm.json, used from the VM's next start; KIRI_CPUS and KIRI_MEMORY_MIB override it) and local names
  • Menu bar: Kiri stays in the menu bar when its window closes, with the VM's state, running containers (click one to open it) and Start/Stop
  • Kubernetes: kiri k8s start — a single-node k3s cluster in one container, ready in ~20 s, as the kiri kubectl context
  • Linux machines: kiri machine create dev --distro ubuntu and kiri machine shell dev — your user, sudo, and /Users at the same path
  • The app: cargo xtask app builds Kiri.app with the docker CLI, Compose and buildx inside; opening it runs kiri setup (docker context, LaunchAgent, build builder) and the window; kiri update installs new releases after checking their SHA-256 and signature
  • Linux hosts: the same kiri with no VM — the agent runs on the host's kernel (cargo xtask linux, contrib/linux/kiri.service); arm64 and x86-64; tests/linux-host.sh checks it. Packaged as .deb and .rpm (packaging/linux): kiri with the docker CLI, Compose and buildx, or kiri-lite for the engine alone. Membership of the kiri group is root-equivalent, exactly like the docker group
  • Windows: the x86-64 Linux build in a WSL2 distro, and kiri-win.exe, which serves the API on npipe:////./pipe/kiri (the kiri docker context) and carries it into WSL over one wsl.exe … kiri bridge, so it needs no change to WSL's or Windows' network settings (cargo xtask msi builds target/Kiri-<version>-x86_64.msi, a per-user installer around target/kiri-windows-x86_64; kiri.ps1 in that folder installs the same without an MSI). Not TCP: WSL's localhost relay drops half-closes, and docker run loses its output

Kubernetes LoadBalancer services are published on localhost, and kiri machine create NAME --systemd boots systemd as the machine's init.

Releases

The app and DMGs are signed with a Developer ID and notarized. Releases and the update feeds are at https://download.runkiri.com, and every file and feed names its architecture: Kiri-<version>-arm64.dmg, Kiri-<version>-arm64.tar.gz and latest-mac-arm64.json for Apple silicon (-x86_64 for Intel Macs), Kiri-<version>-x86_64.msi / -arm64.msi with latest-windows-x86_64.json / latest-windows-arm64.json for Windows. Each app reads only its own feed, and each platform is published on its own. Two DMGs: Kiri-<version>-arm64.dmg bundles the docker CLI, Compose and buildx; Kiri-Lite-<version>-arm64.dmg (cargo xtask app-lite) leaves them out and downloads the same pinned versions on first launch (crates/kiri/src/cli_pins.rs, checked by SHA-256) unless the Mac already has a docker CLI.

Known issues

  • Mac: an attached USB device leaves your containers while the Mac sleeps. macOS takes USB devices off the bus when it sleeps, so a device such as /dev/ttyACM0 is gone from containers until the Mac wakes. Kiri then attaches it again by itself. Kiri doesn't keep the Mac awake for a device, since that would drain a laptop's battery. For a long job that runs unattended (flashing boards, logging a serial port), keep the Mac awake while it runs: caffeinate -s in a terminal (on power), or turn on "Prevent automatic sleeping when the display is off" in System Settings (Battery › Options on a laptop, Energy on a desktop Mac).

  • docker build --add-host name:host-gateway with BuildKit fails with host-gateway is not supported by the docker-container driver. buildx refuses host-gateway for every builder except Docker's built-in one, and Kiri's builder is a docker-container builder, so the build stops before Kiri sees it. Docker's own docker-container builders do the same. Until Kiri has a built-in builder (planned), use the address itself (--add-host name:172.18.0.1 on Windows and Linux, the default bridge's gateway; 198.19.248.254 on a Mac) or DOCKER_BUILDKIT=0, where Kiri maps host-gateway itself.