v0.1.0

Four mesh-daemon strategies and their shared discovery pipeline.

Added

• Cluster.Strategy.TailscaleLocal — peer discovery via
  tailscale status --json, working with Tailscale and with Headscale-based
  control servers, and needing no API key
• Cluster.Strategy.LolipopZTL — peer discovery via ztlctl status
• Cluster.Strategy.Netbird — peer discovery via netbird status --json
• Cluster.Strategy.WireGuard — peer discovery via wg show <if> dump for
  plain WireGuard meshes, selecting members by IPv4 CIDR and treating peers
  with a recent handshake as online
• cidr member selection (LibclusterMesh.CIDR) alongside hostname prefixes
• LibclusterMesh.Runner, a behaviour for running the mesh CLI, with a
  timeout-guarded port-based default that reports a missing binary or a hung
  daemon rather than raising, and signals a command outliving its timeout so
  that a wedged daemon cannot leave one behind on every poll
• Config validated at startup by LibclusterMesh.Config.new!/3

Notes

• Publishing is triggered by pushing a v* tag. The workflow refuses a tag
  whose version disagrees with mix.exs, or one with no section in this file,
  because a version on hex cannot be corrected or reused.
• A config key a strategy does not support raises ArgumentError at startup
  rather than being ignored. This is the failure that is hardest to diagnose
  otherwise: a mistyped or misplaced key means no peer ever matches, so the
  cluster never forms and nothing says why. socket is part of that — only
  Cluster.Strategy.LolipopZTL takes one.
• A daemon that answers with the right key and the wrong shape is an error,
  not an empty peer list. A strategy that keeps polling and finds nobody looks
  exactly like a mesh with nothing on it, so each parser guards its "no peers"
  clause on the key being absent rather than letting it catch everything —
  %{} matches every map in Elixir.

Installing

{:libcluster_mesh, "~> 0.1.0"}

Documentation: https://hexdocs.pm/libcluster_mesh/0.1.0