Add backoff period before subscribing to persistent peers (#491)

waku/v2/node/peer_manager/peer_manager.nim

@@ -76,6 +76,10 @@ proc dialPeer(pm: PeerManager, peerId: PeerID,
 proc loadFromStorage(pm: PeerManager) =
   # Load peers from storage, if available
   proc onData(peerId: PeerID, storedInfo: StoredInfo, connectedness: Connectedness) =
+    if peerId == pm.switch.peerInfo.peerId:
+      # Do not manage self
+      return
+
     pm.peerStore.addressBook.set(peerId, storedInfo.addrs)
     pm.peerStore.protoBook.set(peerId, storedInfo.protos)
     pm.peerStore.keyBook.set(peerId, storedInfo.publicKey)
@@ -149,9 +153,17 @@ proc hasPeer*(pm: PeerManager, peerInfo: PeerInfo, proto: string): bool =
 
   pm.peerStore.get(peerInfo.peerId).protos.contains(proto)
 
+proc hasPeers*(pm: PeerManager, proto: string): bool =
+  # Returns `true` if manager has any peers for the specified protocol
+  pm.peers.anyIt(it.protos.contains(proto))
+
 proc addPeer*(pm: PeerManager, peerInfo: PeerInfo, proto: string) =
   # Adds peer to manager for the specified protocol
 
+  if peerInfo.peerId == pm.switch.peerInfo.peerId:
+    # Do not attempt to manage our unmanageable self
+    return
+  
   debug "Adding peer to manager", peerId = peerInfo.peerId, addr = peerInfo.addrs[0], proto = proto
   
   # ...known addresses

waku/v2/node/wakunode2.nim

@@ -419,10 +419,31 @@ proc mountRelay*(node: WakuNode, topics: seq[string] = newSeq[string](), rlnRela
 
   info "mounting relay"
 
-  node.subscribe(defaultTopic, none(TopicHandler))
+  if node.peerManager.hasPeers(WakuRelayCodec):
+    trace "Found previous WakuRelay peers. Reconnecting."
+    # Reconnect to previous relay peers
+    waitFor node.peerManager.reconnectPeers(WakuRelayCodec)
+    
+    ## GossipSub specifies a backoff period after disconnecting and unsubscribing before attempting
+    ## to re-graft peer on previous topics. We have to respect this period before starting WakuRelay.
+    trace "Backing off before grafting after reconnecting to WakuRelay peers", backoff=wakuRelay.parameters.pruneBackoff
 
-  for topic in topics:
-    node.subscribe(topic, none(TopicHandler))
+    proc subscribeFuture() {.async.} =
+      # Subscribe after the backoff period
+      await sleepAsync(wakuRelay.parameters.pruneBackoff)
+
+      node.subscribe(defaultTopic, none(TopicHandler))
+
+      for topic in topics:
+        node.subscribe(topic, none(TopicHandler))
+    
+    discard subscribeFuture() # Dispatch future, but do not await.
+  else:
+    # Subscribe immediately
+    node.subscribe(defaultTopic, none(TopicHandler))
+
+    for topic in topics:
+      node.subscribe(topic, none(TopicHandler))
 
   if rlnRelayEnabled:
     # TODO pass rln relay inputs to this proc, right now it uses default values that are set in the mountRlnRelay proc
@@ -438,9 +459,6 @@ proc mountRelay*(node: WakuNode, topics: seq[string] = newSeq[string](), rlnRela
     waitFor node.wakuRelay.start()
 
     info "relay mounted and started successfully"
-  
-  # Reconnect to previous relay peers
-  waitFor node.peerManager.reconnectPeers(WakuRelayCodec)
 
 ## Helpers
 proc dialPeer*(n: WakuNode, address: string) {.async.} =

waku/v2/protocol/waku_relay.nim

@@ -41,6 +41,11 @@ method initPubSub*(w: WakuRelay) =
   w.verifySignature = false
   w.sign = false
 
+  # Here we can fine-tune GossipSub params for our purposes
+  w.parameters = GossipSubParams.init()
+  # Setting pruneBackoff allows us to restart nodes and trigger a re-subscribe within reasonable time.
+  w.parameters.pruneBackoff = 30.seconds
+
   procCall GossipSub(w).initPubSub()
 
   w.init()