beacon_node: graceful shutdown (#1033)

* beacon_node: graceful shutdown

* separate BeaconNodeStatus and BeaconNode instances

beacon_chain/beacon_node.nim

@@ -36,6 +36,13 @@ type
   RpcServer* = RpcHttpServer
   KeyPair = eth2_network.KeyPair
 
+  # "state" is already taken by BeaconState
+  BeaconNodeStatus* = enum
+    Starting, Running, Stopping
+
+# this needs to be global, so it can be set in the Ctrl+C signal handler
+var status = BeaconNodeStatus.Starting
+
 template init(T: type RpcHttpServer, ip: IpAddress, port: Port): T =
   newRpcHttpServer([initTAddress(ip, port)])
 
@@ -722,7 +729,16 @@ proc installAttestationHandlers(node: BeaconNode) =
 
   waitFor allFutures(attestationSubscriptions)
 
+proc stop*(node: BeaconNode) =
+  status = BeaconNodeStatus.Stopping
+  info "Graceful shutdown"
+  waitFor node.network.stop()
+
 proc run*(node: BeaconNode) =
+  if status == BeaconNodeStatus.Starting:
+    # it might have been set to "Stopping" with Ctrl+C
+    status = BeaconNodeStatus.Running
+
     if node.rpcServer != nil:
       node.rpcServer.installRpcHandlers(node)
       node.rpcServer.start()
@@ -759,7 +775,15 @@ proc run*(node: BeaconNode) =
 
     node.syncLoop = runSyncLoop(node)
 
-  runForever()
+  # main event loop
+  while status == BeaconNodeStatus.Running:
+    try:
+      poll()
+    except CatchableError as e:
+      debug "Exception in poll()", exc = e.name, err = e.msg
+
+  # time to say goodbye
+  node.stop()
 
 var gPidFile: string
 proc createPidFile(filename: string) =
@@ -955,15 +979,6 @@ programMain:
     stderr.write "Invalid value for --log-level. " & err.msg
     quit 1
 
-  ## Ctrl+C handling
-  proc controlCHandler() {.noconv.} =
-    when defined(windows):
-      # workaround for https://github.com/nim-lang/Nim/issues/4057
-      setupForeignThreadGc()
-    debug "Shutting down after having received SIGINT"
-    quit(QuitFailure)
-  setControlCHook(controlCHandler)
-
   case config.cmd
   of createTestnet:
     var deposits: seq[Deposit]
@@ -1043,6 +1058,16 @@ programMain:
     createPidFile(config.dataDir.string / "beacon_node.pid")
 
     var node = waitFor BeaconNode.init(config)
+
+    ## Ctrl+C handling
+    proc controlCHandler() {.noconv.} =
+      when defined(windows):
+        # workaround for https://github.com/nim-lang/Nim/issues/4057
+        setupForeignThreadGc()
+      info "Shutting down after having received SIGINT"
+      status = BeaconNodeStatus.Stopping
+    setControlCHook(controlCHandler)
+
     when hasPrompt:
       initPrompt(node)
 

beacon_chain/eth2_network.nim

@@ -718,6 +718,13 @@ proc start*(node: Eth2Node) {.async.} =
   node.discoveryLoop = node.runDiscoveryLoop()
   traceAsyncErrors node.discoveryLoop
 
+proc stop*(node: Eth2Node) {.async.} =
+  # ignore errors in futures, since we're shutting down
+  await allFutures(@[
+    node.discovery.closeWait(),
+    node.switch.stop(),
+    ])
+
 proc init*(T: type Peer, network: Eth2Node, info: PeerInfo): Peer =
   new result
   result.info = info

beacon_chain/mainchain_monitor.nim

@@ -442,6 +442,7 @@ proc getGenesis*(m: MainchainMonitor): Future[BeaconStateRef] {.async.} =
   if m.genesisState != nil:
     return m.genesisState
   else:
+    result = new BeaconStateRef # make the compiler happy
     raiseAssert "Unreachable code"
 
 method getBlockByHash*(p: Web3DataProviderRef, hash: BlockHash): Future[BlockObject] =

tests/simulation/start.sh

@@ -110,7 +110,7 @@ if [[ "$USE_TMUX" != "no" ]]; then
   $TMUX select-window -t "${TMUX_SESSION_NAME}:sim"
 fi
 
-$MAKE -j3 NIMFLAGS="$CUSTOM_NIMFLAGS $DEFS" LOG_LEVEL="${LOG_LEVEL:-DEBUG}" beacon_node process_dashboard deposit_contract
+$MAKE -j3 --no-print-directory NIMFLAGS="$CUSTOM_NIMFLAGS $DEFS" LOG_LEVEL="${LOG_LEVEL:-DEBUG}" beacon_node process_dashboard deposit_contract
 
 if [ ! -f "${LAST_VALIDATOR}" ]; then
   if [ "$WEB3_ARG" != "" ]; then