Merge pull request 'feat(companion): 60s mesh probe window + session pre-warm from the VPN service' (#111) from feat/mesh-probe-window-prewarm into main

Android/app/build.gradle.kts

@@ -11,8 +11,8 @@ android {
         applicationId = "com.archipelago.app"
         minSdk = 26
         targetSdk = 35
-        versionCode = 24
+        versionCode = 25
-        versionName = "0.5.4"
+        versionName = "0.5.5"
 
         vectorDrawables {
             useSupportLibrary = true

Android/app/src/main/java/com/archipelago/app/fips/ArchyVpnService.kt

@@ -10,10 +10,15 @@ import android.os.Build
 import android.util.Log
 import com.archipelago.app.MainActivity
 import com.archipelago.app.R
+import com.archipelago.app.data.ServerPreferences
 import kotlinx.coroutines.CoroutineScope
 import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.Job
 import kotlinx.coroutines.SupervisorJob
 import kotlinx.coroutines.cancel
+import kotlinx.coroutines.delay
+import kotlinx.coroutines.flow.first
+import kotlinx.coroutines.isActive
 import kotlinx.coroutines.launch
 
 /**
@@ -27,6 +32,7 @@ import kotlinx.coroutines.launch
 class ArchyVpnService : VpnService() {
 
     private val scope = CoroutineScope(SupervisorJob() + Dispatchers.IO)
+    private var warmerJob: Job? = null
 
     override fun onStartCommand(intent: Intent?, flags: Int, startId: Int): Int {
         if (intent?.action == ACTION_STOP) {
@@ -89,10 +95,57 @@ class ArchyVpnService : VpnService() {
         val fd = pfd.detachFd()
         val result = FipsNative.start(identity.secret, peersJson, fd)
         Log.i(TAG, "mesh start: $result")
-        if (result.contains("\"error\"")) shutdown()
+        if (result.contains("\"error\"")) {
+            shutdown()
+        } else {
+            startSessionWarmer()
+        }
+    }
+
+    /**
+     * Pre-warm + keep-warm mesh sessions to every known node ULA.
+     *
+     * Discovery + first session through the public tree can take 15s+
+     * (HANDOFF-2026-07-23 node diagnosis) — paying that cost here, the
+     * moment the tunnel is up, means the connect probe and WebView hit an
+     * established session instead of timing out on a cold one. The periodic
+     * touch afterwards keeps the session from idling out. Failed connects
+     * are expected and cheap; the attempt itself is what drives discovery.
+     */
+    private fun startSessionWarmer() {
+        warmerJob?.cancel()
+        warmerJob = scope.launch {
+            val prefs = ServerPreferences(this@ArchyVpnService)
+            var round = 0
+            while (isActive && FipsNative.isRunning()) {
+                val ulas = try {
+                    prefs.savedServers.first().mapNotNull { it.meshIp.ifBlank { null } }.distinct()
+                } catch (_: Exception) {
+                    emptyList()
+                }
+                for (ula in ulas) {
+                    try {
+                        java.net.Socket().use { s ->
+                            s.connect(
+                                java.net.InetSocketAddress(java.net.InetAddress.getByName(ula), 80),
+                                20_000,
+                            )
+                        }
+                    } catch (_: Exception) {
+                        // Cold path / node away — the connect attempt still
+                        // drove session establishment; try again next round.
+                    }
+                }
+                round++
+                // Aggressive for the first ~minute (session bring-up), then a
+                // slow keep-warm tick that costs nearly nothing.
+                delay(if (round < 12) 5_000 else 60_000)
+            }
+        }
     }
 
     private fun shutdown() {
+        warmerJob?.cancel()
         FipsNative.stop()
         stopForeground(STOP_FOREGROUND_REMOVE)
         stopSelf()

Android/app/src/main/java/com/archipelago/app/ui/screens/ServerConnectScreen.kt

@@ -185,13 +185,17 @@ fun ServerConnectScreen(
                     useHttps = false,
                     port = "",
                 )
-                // The tunnel + mesh route need a moment on cold start — and on
+                // Mesh discovery + first session can take 15s+ through the
-                // a first-ever pairing the VPN consent dialog is on screen at
+                // public tree (HANDOFF-2026-07-23 node diagnosis), and on a
-                // the same time, so give the user time to tap it.
+                // first-ever pairing the VPN consent dialog is on screen at
-                for (attempt in 0 until 6) {
+                // the same time — so probe patiently inside a 60s budget with
-                    if (attempt > 0) delay(2000)
+                // per-attempt timeouts wide enough to ride out TCP
-                    reachable = testConnection(meshServer)
+                // retransmit backoff. The VPN service pre-warms the session
-                    if (reachable) break
+                // in parallel (ArchyVpnService.startSessionWarmer).
+                val deadline = System.currentTimeMillis() + 60_000
+                while (!reachable && System.currentTimeMillis() < deadline) {
+                    reachable = testConnection(meshServer, timeoutMs = 15_000)
+                    if (!reachable) delay(3000)
                 }
             }
             isConnecting = false
@@ -673,8 +677,10 @@ private fun sanitizeAddress(input: String): String {
         .trimEnd('/')
 }
 
-/** Test RPC connectivity. Accepts self-signed certs for local LAN servers. */
+/** Test RPC connectivity. Accepts self-signed certs for local LAN servers.
-private suspend fun testConnection(server: ServerEntry): Boolean {
+ *  [timeoutMs] is per-phase (connect / read) — mesh probes need far more
+ *  patience than LAN ones (first session through the tree can take 15s+). */
+private suspend fun testConnection(server: ServerEntry, timeoutMs: Int = 5000): Boolean {
     return withContext(Dispatchers.IO) {
         try {
             val url = URL("${server.toUrl()}/rpc/v1")
@@ -694,8 +700,8 @@ private suspend fun testConnection(server: ServerEntry): Boolean {
             }
 
             connection.requestMethod = "POST"
-            connection.connectTimeout = 5000
+            connection.connectTimeout = timeoutMs
-            connection.readTimeout = 5000
+            connection.readTimeout = timeoutMs
             connection.setRequestProperty("Content-Type", "application/json")
             connection.doOutput = true
             val body = """{"method":"server.echo","params":{"message":"ping"}}"""

docs/HANDOFF-2026-07-23-companion-apk-deploy.md

@@ -126,3 +126,19 @@ Chain of findings, each verified live:
   broader public mesh.
 - **Upstream:** report the direct-peer session-path asymmetry to
   jmcorgan/fips (0.4.1).
+
+## App-side recommendations implemented (23:30–23:50, Mac agent — 0.5.5/vc25)
+
+- Connect probe: mesh ULA now probed inside a **60s budget** with 15s
+  per-phase timeouts (rides out TCP retransmit backoff), replacing the old
+  ~8s window.
+- **Session pre-warm**: the VPN service starts probing every saved node ULA
+  the moment the tunnel is up (5s cadence for the first minute, then a 60s
+  keep-warm tick) — discovery/handshake cost is paid in the background, and
+  the session never idles out while the mesh is connected.
+- Extra evidence for the infra decision: from the phone on 5G, a 25-packet
+  ping burst to the node's ULA (25s continuous traffic, post-nginx-fix)
+  still got **zero replies** — discovery through the public tree isn't just
+  slow, it's failing outright. The private-tree option (detach fleet from
+  the public v0l mesh, root at vps2) looks like the real fix; app-side
+  patience only helps once discovery succeeds at all.