claw-store/src/main.rs

@@ -171,6 +171,20 @@ enum Cmd {
         /// Peer's RPC socket. Typically gossip_port + 1.
         #[arg(long)]
         rpc_addr: SocketAddr,
+        /// Phase 8c (2026-07-14): optional Tailscale RPC socket.
+        /// When set, `--rpc-addr` is tried first with a short
+        /// deadline (`--lan-probe-ms`); on failure or timeout we
+        /// fall through to this tailnet address. Roaming ops
+        /// (laptop on LTE, in-flight wifi) get connectivity
+        /// without hand-editing addresses per environment.
+        #[arg(long)]
+        tailscale_addr: Option<SocketAddr>,
+        /// LAN probe deadline in milliseconds. Only used when
+        /// `--tailscale-addr` is set. Default 200ms matches the
+        /// arch doc — long enough for a live LAN handshake, short
+        /// enough that roaming clients don't stall.
+        #[arg(long, default_value_t = 200)]
+        lan_probe_ms: u64,
         /// Directory holding this node's mTLS material
         /// (`ca.crt` + `node.crt` + `node.key` from `fleet-ca sign`).
         #[arg(long)]
@@ -247,6 +261,14 @@ enum Cmd {
         /// Peer's RPC socket. Typically gossip_port + 1.
         #[arg(long)]
         rpc_addr: SocketAddr,
+        /// Phase 8c (2026-07-14): optional Tailscale RPC socket
+        /// for LAN-first-with-fallback routing. See
+        /// `cluster-peer-status` for details.
+        #[arg(long)]
+        tailscale_addr: Option<SocketAddr>,
+        /// LAN probe deadline (ms) when `--tailscale-addr` is set.
+        #[arg(long, default_value_t = 200)]
+        lan_probe_ms: u64,
         /// Directory holding this node's mTLS material.
         #[arg(long)]
         tls_dir: PathBuf,
@@ -331,9 +353,11 @@ async fn main() -> Result<()> {
         Cmd::ClusterRepair {
             peer,
             rpc_addr,
+            tailscale_addr,
+            lan_probe_ms,
             tls_dir,
             dry_run,
-        } => cmd_cluster_repair(&cfg, &peer, rpc_addr, &tls_dir, dry_run).await?,
+        } => cmd_cluster_repair(&cfg, &peer, rpc_addr, tailscale_addr, lan_probe_ms, &tls_dir, dry_run).await?,
         Cmd::ClusterRefSweep {
             gitea_url,
             gitea_token,
@@ -346,8 +370,10 @@ async fn main() -> Result<()> {
         Cmd::ClusterPeerStatus {
             peer,
             rpc_addr,
+            tailscale_addr,
+            lan_probe_ms,
             tls_dir,
-        } => cmd_cluster_peer_status(&peer, rpc_addr, &tls_dir).await?,
+        } => cmd_cluster_peer_status(&peer, rpc_addr, tailscale_addr, lan_probe_ms, &tls_dir).await?,
         Cmd::FleetCaInit { .. } | Cmd::FleetCaSign { .. } | Cmd::FleetCaTailscaleSign { .. } => {
             // Handled by the config-independent short-circuit above.
             unreachable!("fleet-ca commands short-circuit before config load");
@@ -520,12 +546,14 @@ async fn cmd_cluster_repair(
     cfg: &Config,
     peer: &str,
     rpc_addr: SocketAddr,
+    tailscale_addr: Option<SocketAddr>,
+    lan_probe_ms: u64,
     tls_dir: &std::path::Path,
     dry_run: bool,
 ) -> Result<()> {
     use cluster::blob::BlobStore;
     use cluster::rpc::{call_get_chunk, call_has_chunk};
-    use cluster::transport::{NodeIdentity, QuicClient};
+    use cluster::transport::{ConnectRoute, NodeIdentity, QuicClient};
 
     let root = cfg
         .cluster
@@ -581,14 +609,24 @@ async fn cmd_cluster_repair(
         return Ok(());
     }
 
-    // Phase 2: connect to the peer.
+    // Phase 2: connect to the peer. Phase 8c: LAN-first with
+    // optional tailnet fallback when the operator supplied one.
     let identity = NodeIdentity::from_pem_dir(tls_dir)
         .with_context(|| format!("loading identity from {}", tls_dir.display()))?;
     let client = QuicClient::new("0.0.0.0:0".parse()?, identity)?;
-    let conn = client
-        .connect(rpc_addr, peer)
+    let (conn, route) = client
+        .connect_lan_first(
+            peer,
+            Some(rpc_addr),
+            tailscale_addr,
+            std::time::Duration::from_millis(lan_probe_ms),
+        )
         .await
-        .with_context(|| format!("connecting to {peer} at {rpc_addr}"))?;
+        .with_context(|| format!("connecting to {peer}"))?;
+    match route {
+        ConnectRoute::Lan(a) => println!("route:             LAN ({a})"),
+        ConnectRoute::Tailscale(a) => println!("route:             tailnet ({a})"),
+    }
 
     // Phase 3: repair. Fetcher probes HasChunk first (cheap) so a
     // peer that lacks the chunk is one round-trip, not a full pull
@@ -835,15 +873,31 @@ async fn cmd_cluster_ref_sweep(
 async fn cmd_cluster_peer_status(
     peer: &str,
     rpc_addr: SocketAddr,
+    tailscale_addr: Option<SocketAddr>,
+    lan_probe_ms: u64,
     tls_dir: &std::path::Path,
 ) -> Result<()> {
     use cluster::rpc::call_peer_status;
-    use cluster::transport::{NodeIdentity, QuicClient};
+    use cluster::transport::{ConnectRoute, NodeIdentity, QuicClient};
 
     let identity = NodeIdentity::from_pem_dir(tls_dir)
         .with_context(|| format!("loading identity from {}", tls_dir.display()))?;
     let client = QuicClient::new("0.0.0.0:0".parse()?, identity)?;
-    let conn = client.connect(rpc_addr, peer).await?;
+    // Phase 8c: LAN-first with optional tailnet fallback. When the
+    // caller didn't pass --tailscale-addr the behavior is
+    // byte-identical to the pre-8c path (single-addr dial).
+    let (conn, route) = client
+        .connect_lan_first(
+            peer,
+            Some(rpc_addr),
+            tailscale_addr,
+            std::time::Duration::from_millis(lan_probe_ms),
+        )
+        .await?;
+    match route {
+        ConnectRoute::Lan(a) => println!("route: LAN ({a})"),
+        ConnectRoute::Tailscale(a) => println!("route: tailnet ({a})"),
+    }
     let status = call_peer_status(&conn).await?;
 
     println!(