fix: start_node() after stop_node() re-opens the listener (#4007)

logos_delivery/waku/node/waku_node.nim

@@ -136,6 +136,7 @@ type
     wakuMix*: WakuMix
     wakuKademlia*: WakuKademlia
     ports*: BoundPorts
+    relayReconnectFut*: Future[void]
 
   SubscriptionManager* = ref object of RootObj
     node*: WakuNode
@@ -620,8 +621,9 @@ proc start*(node: WakuNode) {.async.} =
   ## NOTE: This will dispatch gossipsub start to the WakuRelay.start method override
   await node.switch.start()
 
-  # After switch.start, run custom Logos Delivery relay start logic
-  await node.reconnectRelayPeers()
+  # Reconnect to known relay peers in the background; it waits a prune backoff
+  # and must not block startup.
+  node.relayReconnectFut = node.reconnectRelayPeers()
 
   node.started = true
 
@@ -650,6 +652,10 @@ proc start*(node: WakuNode) {.async.} =
 proc stop*(node: WakuNode) {.async.} =
   ## By stopping the switch we are stopping all the underlying mounted protocols
 
+  # Cancel the background relay reconnection (may still be in its backoff wait).
+  if not node.relayReconnectFut.isNil():
+    await node.relayReconnectFut.cancelAndWait()
+
   await node.subscriptionManager.stop()
 
   node.stopProvidersAndListeners()

logos_delivery/waku/waku.nim

@@ -563,8 +563,10 @@ proc stop*(waku: Waku): Future[Result[void, string]] {.async: (raises: []).} =
     if not waku.healthMonitor.isNil():
       await waku.healthMonitor.stopHealthMonitor()
 
-    ## Clear RequestConnectionStatus provider
+    ## Clear all providers registered in start() so a later start() can re-set them.
     RequestConnectionStatus.clearProvider(waku.brokerCtx)
+    RequestProtocolHealth.clearProvider(waku.brokerCtx)
+    RequestHealthReport.clearProvider(waku.brokerCtx)
 
     if not waku.restServer.isNil():
       await waku.restServer.stop()

tests/node/test_wakunode_restart.nim

@@ -0,0 +1,42 @@
+{.used.}
+
+import std/options
+import testutils/unittests, chronos, chronicles
+import libp2p/switch
+
+import logos_delivery/waku/[waku_node, waku_core, node/peer_manager]
+import ../testlib/[wakucore, wakunode, testasync]
+
+suite "WakuNode - restart (#3979)":
+  asyncTest "start -> stop -> start re-opens the listener promptly":
+    ## A restart must not block on the relay-reconnect backoff.
+    let
+      node1 =
+        newTestWakuNode(generateSecp256k1Key(), parseIpAddress("0.0.0.0"), Port(0))
+      node2 =
+        newTestWakuNode(generateSecp256k1Key(), parseIpAddress("0.0.0.0"), Port(0))
+
+    (await node1.mountRelay()).isOkOr:
+      raiseAssert "mountRelay node1: " & error
+    (await node2.mountRelay()).isOkOr:
+      raiseAssert "mountRelay node2: " & error
+
+    await allFutures(node1.start(), node2.start())
+
+    # node1 learns node2 as a relay peer, so a restart triggers reconnectRelayPeers.
+    await node1.connectToNodes(@[node2.peerInfo.toRemotePeerInfo()])
+
+    await node1.stop()
+
+    # The restart must complete promptly and yield a usable, listening node.
+    let startFut = node1.start()
+    let restarted = await startFut.withTimeout(20.seconds)
+    if not restarted:
+      await startFut.cancelAndWait()
+
+    check:
+      restarted
+      node1.started
+      node1.switch.peerInfo.listenAddrs.len > 0
+
+    await allFutures(node1.stop(), node2.stop())