Merge remote-tracking branch 'gitea-ai/fix/reticulum-daemon-process-group'

core/Cargo.lock

@@ -129,6 +129,7 @@ dependencies = [
  "hyper-ws-listener",
  "iroh",
  "iroh-blobs",
+ "libc",
  "mainline",
  "mdns-sd",
  "nostr-sdk",

core/archipelago/Cargo.toml

@@ -22,6 +22,7 @@ iroh-swarm = ["dep:iroh", "dep:iroh-blobs"]
 [dependencies]
 # Core dependencies
 tokio = { version = "1", features = ["full"] }
+libc = "0.2"   # process-group signalling for the supervised reticulum daemon
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 anyhow = "1.0"

core/archipelago/src/mesh/reticulum.rs

@@ -104,6 +104,11 @@ fn daemon_command(
             .arg(x);
     }
     cmd.kill_on_drop(true)
+        // Run the daemon as its own process-group leader. The packaged binary is
+        // a PyInstaller one-file bootloader that forks the real Python process;
+        // making it a group leader lets Drop signal the WHOLE group so the
+        // forked child can't be orphaned and keep holding the serial port.
+        .process_group(0)
         .stdin(std::process::Stdio::null())
         .stdout(std::process::Stdio::piped())
         .stderr(std::process::Stdio::piped());
@@ -945,9 +950,21 @@ fn contains_detect_resp(buf: &[u8]) -> bool {
 
 impl Drop for ReticulumLink {
     fn drop(&mut self) {
-        // Best-effort: ask the daemon to shut down cleanly (frees the serial
+        // The packaged daemon is a PyInstaller one-file bootloader that forks the
-        // port promptly); `kill_on_drop` on the Command is the hard backstop
+        // real Python process into the same (leader) group we spawned it in.
-        // if the daemon doesn't exit in time.
+        // SIGKILLing only the bootloader (what `start_kill`/`kill_on_drop` do)
+        // orphans that child, which keeps holding the serial port — the root
+        // cause of daemons piling up across reconnects and jamming the RNode.
+        // Signal the whole process group instead: SIGTERM is caught by the
+        // daemon's handler (clean RNode + socket release), and SIGKILL is the
+        // hard backstop so a wedged daemon can never survive.
+        if let Some(pid) = self.child.id() {
+            let pgid = -(pid as i32);
+            unsafe {
+                libc::kill(pgid, libc::SIGTERM);
+                libc::kill(pgid, libc::SIGKILL);
+            }
+        }
         let _ = self.child.start_kill();
         let _ = std::fs::remove_file(&self.socket_path);
     }