Fluffy state network now enabled by default and improve status logs (#2640)

* Enable state network by default. Create status log loop for state and beacon networks. Create status log loop for portal node. Implement stop functions.

fluffy/conf.nim

@@ -89,7 +89,7 @@ type
 
     portalSubnetworks* {.
       desc: "Select which networks (Portal sub-protocols) to enable",
-      defaultValue: {PortalSubnetwork.history},
+      defaultValue: {PortalSubnetwork.history, PortalSubnetwork.state},
       name: "portal-subnetworks"
     .}: set[PortalSubnetwork]
 

fluffy/network/beacon/beacon_light_client.nim

@@ -178,10 +178,13 @@ proc new*(
   )
 
 proc start*(lightClient: LightClient) =
-  notice "Starting beacon light client",
-    trusted_block_root = lightClient.trustedBlockRoot
+  info "Starting beacon light client", trusted_block_root = lightClient.trustedBlockRoot
   lightClient.manager.start()
 
+proc stop*(lightClient: LightClient) =
+  info "Stopping beacon light client"
+  discard lightClient.manager.stop()
+
 proc resetToFinalizedHeader*(
     lightClient: LightClient,
     header: ForkedLightClientHeader,

fluffy/network/beacon/beacon_network.nim

@@ -31,6 +31,7 @@ type BeaconNetwork* = ref object
   forkDigests*: ForkDigests
   trustedBlockRoot: Opt[Eth2Digest]
   processContentLoop: Future[void]
+  statusLogLoop: Future[void]
 
 func toContentIdHandler(contentKey: ContentKeyByteList): results.Opt[ContentId] =
   ok(toContentId(contentKey))
@@ -364,13 +365,29 @@ proc processContentLoop(n: BeaconNetwork) {.async: (raises: []).} =
   except CancelledError:
     trace "processContentLoop canceled"
 
+proc statusLogLoop(n: BeaconNetwork) {.async: (raises: []).} =
+  try:
+    while true:
+      info "Beacon network status",
+        routingTableNodes = n.portalProtocol.routingTable.len()
+
+      await sleepAsync(60.seconds)
+  except CancelledError:
+    trace "statusLogLoop canceled"
+
 proc start*(n: BeaconNetwork) =
   info "Starting Portal beacon chain network"
+
   n.portalProtocol.start()
   n.processContentLoop = processContentLoop(n)
 
 proc stop*(n: BeaconNetwork) =
+  info "Stopping Portal beacon chain network"
+
   n.portalProtocol.stop()
 
   if not n.processContentLoop.isNil:
     n.processContentLoop.cancelSoon()
+
+  if not n.statusLogLoop.isNil():
+    n.statusLogLoop.cancelSoon()

fluffy/network/history/history_network.nim

@@ -704,17 +704,7 @@ proc processContentLoop(n: HistoryNetwork) {.async: (raises: []).} =
 proc statusLogLoop(n: HistoryNetwork) {.async: (raises: []).} =
   try:
     while true:
-      # This is the data radius percentage compared to full storage. This will
-      # drop a lot when using the logbase2 scale, namely `/ 2` per 1 logaritmic
-      # radius drop.
-      # TODO: Get some float precision calculus?
-      let radiusPercentage =
-        n.portalProtocol.dataRadius() div (UInt256.high() div u256(100))
-
       info "History network status",
-        radiusPercentage = radiusPercentage.toString(10) & "%",
-        radius = n.portalProtocol.dataRadius().toHex(),
-        dbSize = $(n.contentDB.size() div 1000) & "kb",
         routingTableNodes = n.portalProtocol.routingTable.len()
 
       await sleepAsync(60.seconds)
@@ -725,6 +715,7 @@ proc start*(n: HistoryNetwork) =
   info "Starting Portal execution history network",
     protocolId = n.portalProtocol.protocolId,
     accumulatorRoot = hash_tree_root(n.accumulator)
+
   n.portalProtocol.start()
 
   n.processContentLoop = processContentLoop(n)
@@ -732,10 +723,12 @@ proc start*(n: HistoryNetwork) =
   pruneDeprecatedAccumulatorRecords(n.accumulator, n.contentDB)
 
 proc stop*(n: HistoryNetwork) =
+  info "Stopping Portal execution history network"
+
   n.portalProtocol.stop()
 
   if not n.processContentLoop.isNil:
     n.processContentLoop.cancelSoon()
 
-  if not n.processContentLoop.isNil:
+  if not n.statusLogLoop.isNil:
     n.statusLogLoop.cancelSoon()

fluffy/network/state/state_gossip.nim

@@ -128,7 +128,6 @@ proc gossipOffer*(
   debug "Offered content gossipped successfully with peers", keyBytes, peers
 
 # Currently only used for testing to gossip an entire account trie proof
-# This may also be useful for the state network bridge
 proc recursiveGossipOffer*(
     p: PortalProtocol,
     srcNodeId: Opt[NodeId],

fluffy/network/state/state_network.nim

@@ -31,6 +31,7 @@ type StateNetwork* = ref object
   contentDB*: ContentDB
   contentQueue*: AsyncQueue[(Opt[NodeId], ContentKeysList, seq[seq[byte]])]
   processContentLoop: Future[void]
+  statusLogLoop: Future[void]
   historyNetwork: Opt[HistoryNetwork]
   validateStateIsCanonical: bool
 
@@ -221,15 +222,32 @@ proc processContentLoop(n: StateNetwork) {.async: (raises: []).} =
   except CancelledError:
     trace "processContentLoop canceled"
 
+proc statusLogLoop(n: StateNetwork) {.async: (raises: []).} =
+  try:
+    while true:
+      info "State network status",
+        routingTableNodes = n.portalProtocol.routingTable.len()
+
+      await sleepAsync(60.seconds)
+  except CancelledError:
+    trace "statusLogLoop canceled"
+
 proc start*(n: StateNetwork) =
   info "Starting Portal execution state network",
     protocolId = n.portalProtocol.protocolId
+
   n.portalProtocol.start()
 
   n.processContentLoop = processContentLoop(n)
+  n.statusLogLoop = statusLogLoop(n)
 
 proc stop*(n: StateNetwork) =
+  info "Stopping Portal execution state network"
+
   n.portalProtocol.stop()
 
-  if not n.processContentLoop.isNil:
+  if not n.processContentLoop.isNil():
     n.processContentLoop.cancelSoon()
+
+  if not n.statusLogLoop.isNil():
+    n.statusLogLoop.cancelSoon()

fluffy/portal_node.nim

@@ -9,6 +9,7 @@
 
 import
   results,
+  chronos,
   eth/p2p/discoveryv5/protocol,
   beacon_chain/spec/forks,
   ./network_metadata,
@@ -39,6 +40,7 @@ type
     historyNetwork*: Opt[HistoryNetwork]
     stateNetwork*: Opt[StateNetwork]
     beaconLightClient*: Opt[LightClient]
+    statusLogLoop: Future[void]
 
 # Beacon light client application callbacks triggered when new finalized header
 # or optimistic header is available.
@@ -179,7 +181,27 @@ proc new*(
     beaconLightClient: beaconLightClient,
   )
 
+proc statusLogLoop(n: PortalNode) {.async: (raises: []).} =
+  try:
+    while true:
+      # This is the data radius percentage compared to full storage. This will
+      # drop a lot when using the logbase2 scale, namely `/ 2` per 1 logaritmic
+      # radius drop.
+      # TODO: Get some float precision calculus?
+      let radiusPercentage = n.contentDB.dataRadius div (UInt256.high() div u256(100))
+
+      info "Portal node status",
+        radiusPercentage = radiusPercentage.toString(10) & "%",
+        radius = n.contentDB.dataRadius.toHex(),
+        dbSize = $(n.contentDB.size() div 1000) & "kb"
+
+      await sleepAsync(60.seconds)
+  except CancelledError:
+    trace "statusLogLoop canceled"
+
 proc start*(n: PortalNode) =
+  debug "Starting Portal node"
+
   if n.beaconNetwork.isSome():
     n.beaconNetwork.value.start()
   if n.historyNetwork.isSome():
@@ -189,3 +211,21 @@ proc start*(n: PortalNode) =
 
   if n.beaconLightClient.isSome():
     n.beaconLightClient.value.start()
+
+  n.statusLogLoop = statusLogLoop(n)
+
+proc stop*(n: PortalNode) =
+  debug "Stopping Portal node"
+
+  if n.beaconNetwork.isSome():
+    n.beaconNetwork.value.stop()
+  if n.historyNetwork.isSome():
+    n.historyNetwork.value.stop()
+  if n.stateNetwork.isSome():
+    n.stateNetwork.value.stop()
+
+  if n.beaconLightClient.isSome():
+    n.beaconLightClient.value.stop()
+
+  if not n.statusLogLoop.isNil:
+    n.statusLogLoop.cancelSoon()

fluffy/tools/portal_bridge/portal_bridge_state.nim

@@ -298,7 +298,9 @@ proc runBackfillGossipBlockOffersLoop(
     for k, v in offersMap:
       try:
         let numPeers = await portalClient.portal_stateGossip(k.to0xHex(), v.to0xHex())
-        if numPeers == 0:
+        if numPeers > 0:
+          debug "Offer successfully gossipped to peers: ", numPeers, workerId
+        elif numPeers == 0:
           warn "Offer gossipped to no peers", workerId
           retryGossip = true
           break