chore: Separation of node health and initialization state from rln_relay

apps/wakunode2/wakunode2.nim

@@ -18,7 +18,8 @@ import
   ../../waku/common/logging,
   ../../waku/factory/external_config,
   ../../waku/factory/networks_config,
-  ../../waku/factory/app
+  ../../waku/factory/app,
+  ../../waku/node/health_monitor
 
 logScope:
   topics = "wakunode main"
@@ -88,54 +89,74 @@ when isMainModule:
     doInspectRlnDb(conf)
   of noCommand:
     case conf.clusterId
-      # cluster-id=0
-      of 0:
-        let clusterZeroConf = ClusterConf.ClusterZeroConf()
-        conf.pubsubTopics = clusterZeroConf.pubsubTopics
-        # TODO: Write some template to "merge" the configs
-      # cluster-id=1 (aka The Waku Network)
-      of 1:
-        let twnClusterConf = ClusterConf.TheWakuNetworkConf()
-        if len(conf.shards) != 0:
-          conf.pubsubTopics = conf.shards.mapIt(twnClusterConf.pubsubTopics[it.uint16])
-        else:
-          conf.pubsubTopics = twnClusterConf.pubsubTopics
-
-        # Override configuration
-        conf.maxMessageSize = twnClusterConf.maxMessageSize
-        conf.clusterId = twnClusterConf.clusterId
-        conf.rlnRelay = twnClusterConf.rlnRelay
-        conf.rlnRelayEthContractAddress = twnClusterConf.rlnRelayEthContractAddress
-        conf.rlnRelayDynamic = twnClusterConf.rlnRelayDynamic
-        conf.rlnRelayBandwidthThreshold = twnClusterConf.rlnRelayBandwidthThreshold
-        conf.discv5Discovery = twnClusterConf.discv5Discovery
-        conf.discv5BootstrapNodes =
-          conf.discv5BootstrapNodes & twnClusterConf.discv5BootstrapNodes
-        conf.rlnEpochSizeSec = twnClusterConf.rlnEpochSizeSec
-        conf.rlnRelayUserMessageLimit = twnClusterConf.rlnRelayUserMessageLimit
+    # cluster-id=0
+    of 0:
+      let clusterZeroConf = ClusterConf.ClusterZeroConf()
+      conf.pubsubTopics = clusterZeroConf.pubsubTopics
+      # TODO: Write some template to "merge" the configs
+    # cluster-id=1 (aka The Waku Network)
+    of 1:
+      let twnClusterConf = ClusterConf.TheWakuNetworkConf()
+      if len(conf.shards) != 0:
+        conf.pubsubTopics = conf.shards.mapIt(twnClusterConf.pubsubTopics[it.uint16])
       else:
-        discard
+        conf.pubsubTopics = twnClusterConf.pubsubTopics
+
+      # Override configuration
+      conf.maxMessageSize = twnClusterConf.maxMessageSize
+      conf.clusterId = twnClusterConf.clusterId
+      conf.rlnRelay = twnClusterConf.rlnRelay
+      conf.rlnRelayEthContractAddress = twnClusterConf.rlnRelayEthContractAddress
+      conf.rlnRelayDynamic = twnClusterConf.rlnRelayDynamic
+      conf.rlnRelayBandwidthThreshold = twnClusterConf.rlnRelayBandwidthThreshold
+      conf.discv5Discovery = twnClusterConf.discv5Discovery
+      conf.discv5BootstrapNodes =
+        conf.discv5BootstrapNodes & twnClusterConf.discv5BootstrapNodes
+      conf.rlnEpochSizeSec = twnClusterConf.rlnEpochSizeSec
+      conf.rlnRelayUserMessageLimit = twnClusterConf.rlnRelayUserMessageLimit
+    else:
+      discard
 
     info "Running nwaku node", version = app.git_version
     logConfig(conf)
 
+    # NOTE: {.threadvar.} is used to make the global variable GC safe for the closure uses it
+    # It will always be called from main thread anyway.
+    # Ref: https://nim-lang.org/docs/manual.html#threads-gc-safety
+    var nodeHealthMonitor {.threadvar.}: WakuNodeHealthMonitor
+    nodeHealthMonitor = WakuNodeHealthMonitor()
+    nodeHealthMonitor.setOverallHealth(HealthStatus.INITIALIZING)
+
+    let restServerRes = startRestServerEsentials(nodeHealthMonitor, conf)
+    if restServerRes.isErr():
+      error "Starting REST server failed.", error = $restServerRes.error()
+      quit(QuitFailure)
+
     var wakunode2 = App.init(conf).valueOr:
       error "App initialization failed", error = error
       quit(QuitFailure)
 
+    nodeHealthMonitor.setNode(wakunode2.node)
+
     wakunode2.startApp().isOkOr:
       error "Starting app failed", error = error
       quit(QuitFailure)
 
+    if conf.rest and not restServerRes.isErr():
+      wakunode2.restServer = restServerRes.value
+
     wakunode2.setupMonitoringAndExternalInterfaces().isOkOr:
       error "Starting monitoring and external interfaces failed", error = error
       quit(QuitFailure)
 
+    nodeHealthMonitor.setOverallHealth(HealthStatus.READY)
+
     debug "Setting up shutdown hooks"
     ## Setup shutdown hooks for this process.
     ## Stop node gracefully on shutdown.
 
     proc asyncStopper(node: App) {.async: (raises: [Exception]).} =
+      nodeHealthMonitor.setOverallHealth(HealthStatus.SHUTTING_DOWN)
       await node.stop()
       quit(QuitSuccess)
 

scripts/chkhealth.sh

@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+
+# optional argument to specgify the ip address
+ip_address=$1
+plain_text_out=false
+
+# Parse command line arguments
+POSITIONAL_ARGS=()
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+    -p|--plain)
+      plain_text_out=true
+      shift # past argument
+      ;;
+    -*|--*)
+      echo "Unknown option $1"
+      exit 1
+      ;;
+    *)
+      POSITIONAL_ARGS+=("$1") # save positional arg
+      shift # past argument
+      ;;
+  esac
+done
+
+set -- "${POSITIONAL_ARGS[@]}" # restore positional parameters
+
+# Check if an IP address is provided as an argument
+if [[ -n "$1" ]]; then
+  ip_address="$1"
+else
+  ip_address="localhost:8645"
+fi
+
+# check if curl is available
+if ! command -v curl &> /dev/null
+then
+    echo "curl could not be found"
+    exit 1
+fi
+
+response=$(curl -s GET http://${ip_address}/health)
+
+if [[ -z "${response}" ]]; then
+  echo -e "$(date +'%H:%M:%S')\tnode health status is: unknown\n"
+  exit 1
+fi
+
+if ! command -v jq &> /dev/null || [[ "$plain_text_out" = true ]]; then
+  echo -e "$(date +'%H:%M:%S')\tnode health status is: ${response}\n"
+else
+  echo -e "$(date +'%H:%M:%S')\tnode health status is:\n"
+  echo "${response}" | jq .
+fi

tests/all_tests_waku.nim

@@ -77,7 +77,8 @@ import
   ./wakunode_rest/test_rest_filter,
   ./wakunode_rest/test_rest_lightpush,
   ./wakunode_rest/test_rest_admin,
-  ./wakunode_rest/test_rest_cors
+  ./wakunode_rest/test_rest_cors,
+  ./wakunode_rest/test_rest_health
 
 import ./waku_rln_relay/test_all
 

tests/wakunode_rest/test_rest_health.nim

@@ -19,6 +19,7 @@ import
   ../../waku/waku_api/rest/health/handlers as health_api,
   ../../waku/waku_api/rest/health/client as health_api_client,
   ../../waku/waku_rln_relay,
+  ../../waku/node/health_monitor,
   ../testlib/common,
   ../testlib/testutils,
   ../testlib/wakucore,
@@ -35,17 +36,20 @@ proc testWakuNode(): WakuNode =
 
 suite "Waku v2 REST API - health":
   # TODO: better test for health
-  xasyncTest "Get node health info - GET /health":
+  asyncTest "Get node health info - GET /health":
     # Given
     let node = testWakuNode()
+    let healthMonitor = WakuNodeHealthMonitor()
     await node.start()
     await node.mountRelay()
 
+    healthMonitor.setOverallHealth(HealthStatus.INITIALIZING)
+
     let restPort = Port(58001)
     let restAddress = parseIpAddress("0.0.0.0")
     let restServer = WakuRestServerRef.init(restAddress, restPort).tryGet()
 
-    installHealthApiHandler(restServer.router, node)
+    installHealthApiHandler(restServer.router, healthMonitor)
     restServer.start()
     let client = newRestHttpClient(initTAddress(restAddress, restPort))
 
@@ -54,9 +58,10 @@ suite "Waku v2 REST API - health":
 
     # Then
     check:
-      response.status == 503
-      $response.contentType == $MIMETYPE_TEXT
-      response.data == "Node is not ready"
+      response.status == 200
+      $response.contentType == $MIMETYPE_JSON
+      response.data ==
+        HealthReport(nodeHealth: HealthStatus.INITIALIZING, protocolsHealth: @[])
 
     # now kick in rln (currently the only check for health)
     await node.mountRlnRelay(
@@ -67,15 +72,19 @@ suite "Waku v2 REST API - health":
         rlnRelayTreePath: genTempPath("rln_tree", "wakunode"),
       )
     )
-
+    healthMonitor.setNode(node)
+    healthMonitor.setOverallHealth(HealthStatus.READY)
     # When
     response = await client.healthCheck()
 
     # Then
     check:
       response.status == 200
-      $response.contentType == $MIMETYPE_TEXT
-      response.data == "Node is healthy"
+      $response.contentType == $MIMETYPE_JSON
+      response.data.nodeHealth == HealthStatus.READY
+      response.data.protocolsHealth.len() == 1
+      response.data.protocolsHealth[0].protocol == "Rln Relay"
+      response.data.protocolsHealth[0].health == HealthStatus.READY
 
     await restServer.stop()
     await restServer.closeWait()