VC: metrics (#3915)

* Initial commit.
Enable MetricsHttpServerRef and configuration.

* Add metrics.

* Add headers.
Add compilation issue fixes.

beacon_chain/conf.nim

@@ -811,6 +811,22 @@ type
       desc: "A file specifying the authorizition token required for accessing the keymanager API"
       name: "keymanager-token-file" .}: Option[InputFile]
 
+    metricsEnabled* {.
+      desc: "Enable the metrics server"
+      defaultValue: false
+      name: "metrics" .}: bool
+
+    metricsAddress* {.
+      desc: "Listening address of the metrics server"
+      defaultValue: defaultAdminListenAddress
+      defaultValueDesc: $defaultAdminListenAddressDesc
+      name: "metrics-address" .}: ValidIpAddress
+
+    metricsPort* {.
+      desc: "Listening HTTP port of the metrics server"
+      defaultValue: 8008
+      name: "metrics-port" .}: Port
+
     graffiti* {.
       desc: "The graffiti value that will appear in proposed blocks. " &
             "You can use a 0x-prefixed hex encoded string to specify " &

beacon_chain/nimbus_validator_client.nim

@@ -4,10 +4,14 @@
 #   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
 #   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
 # at your option. This file may not be copied, modified, or distributed except according to those terms.
+import metrics, metrics/chronos_httpserver
 import validator_client/[common, fallback_service, duties_service,
                          attestation_service, fork_service,
                          sync_committee_service, doppelganger_service]
 
+type
+  ValidatorClientError* = object of CatchableError
+
 proc initGenesis(vc: ValidatorClientRef): Future[RestGenesis] {.async.} =
   info "Initializing genesis", nodes_count = len(vc.beaconNodes)
   var nodes = vc.beaconNodes
@@ -110,6 +114,40 @@ proc initClock(vc: ValidatorClientRef): Future[BeaconClock] {.async.} =
     await sleepAsync(genesisTime.offset)
   return res
 
+proc initMetrics(vc: ValidatorClientRef): Future[bool] {.async.} =
+  if vc.config.metricsEnabled:
+    let
+      metricsAddress = vc.config.metricsAddress
+      metricsPort = vc.config.metricsPort
+      url = "http://" & $metricsAddress & ":" & $metricsPort & "/metrics"
+    info "Starting metrics HTTP server", url = url
+    let server =
+      block:
+        let res = MetricsHttpServerRef.new($metricsAddress, metricsPort)
+        if res.isErr():
+          error "Could not start metrics HTTP server", url = url,
+                error_msg = res.error()
+          return false
+        res.get()
+    vc.metricsServer = some(server)
+    try:
+      await server.start()
+    except MetricsError as exc:
+      error "Could not start metrics HTTP server", url = url,
+            error_msg = exc.msg, error_name = exc.name
+      return false
+  return true
+
+proc shutdownMetrics(vc: ValidatorClientRef) {.async.} =
+  if vc.config.metricsEnabled:
+    if vc.metricsServer.isSome():
+      info "Shutting down metrics HTTP server"
+      await vc.metricsServer.get().close()
+
+proc shutdownSlashingProtection(vc: ValidatorClientRef) =
+  info "Closing slashing protection", path = vc.config.validatorsDir()
+  vc.attachedValidators.slashingProtection.close()
+
 proc onSlotStart(vc: ValidatorClientRef, wallTime: BeaconTime,
                  lastSlot: Slot): Future[bool] {.async.} =
   ## Called at the beginning of a slot - usually every slot, but sometimes might
@@ -148,8 +186,19 @@ proc asyncInit(vc: ValidatorClientRef) {.async.} =
 
   vc.beaconClock = await vc.initClock()
 
-  if not(await initValidators(vc)):
-    fatal "Could not initialize local validators"
+  if not(await initMetrics(vc)):
+    raise newException(ValidatorClientError,
+                       "Could not initialize metrics server")
+
+  try:
+    if not(await initValidators(vc)):
+      await vc.shutdownMetrics()
+      raise newException(ValidatorClientError,
+                         "Could not initialize local validators")
+  except CancelledError:
+    debug "Initialization process interrupted"
+    await vc.shutdownMetrics()
+    return
 
   info "Initializing slashing protection", path = vc.config.validatorsDir()
   vc.attachedValidators.slashingProtection =
@@ -165,10 +214,16 @@ proc asyncInit(vc: ValidatorClientRef) {.async.} =
     vc.doppelgangerService = await DoppelgangerServiceRef.init(vc)
     vc.attestationService = await AttestationServiceRef.init(vc)
     vc.syncCommitteeService = await SyncCommitteeServiceRef.init(vc)
+  except CatchableError as exc:
+    warn "Unexpected error encountered while initializing",
+          error_name = exc.name, error_msg = exc.msg
+    await vc.shutdownMetrics()
+    vc.shutdownSlashingProtection()
   except CancelledError:
     debug "Initialization process interrupted"
-    info "Closing slashing protection", path = vc.config.validatorsDir()
-    vc.attachedValidators.slashingProtection.close()
+    await vc.shutdownMetrics()
+    vc.shutdownSlashingProtection()
+    return
 
 proc asyncRun(vc: ValidatorClientRef) {.async.} =
   vc.fallbackService.start()
@@ -190,8 +245,8 @@ proc asyncRun(vc: ValidatorClientRef) {.async.} =
     debug "Main loop failed with an error", err_name = $exc.name,
           err_msg = $exc.msg
 
-  info "Closing slashing protection", path = vc.config.validatorsDir()
-  vc.attachedValidators.slashingProtection.close()
+  await vc.shutdownMetrics()
+  vc.shutdownSlashingProtection()
   debug "Stopping main processing loop"
   var pending: seq[Future[void]]
   if not(vc.runSlotLoopFut.finished()):

beacon_chain/validator_client/api.nim

@@ -1,3 +1,10 @@
+# beacon_chain
+# Copyright (c) 2021-2022 Status Research & Development GmbH
+# Licensed and distributed under either of
+#   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
+#   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
+# at your option. This file may not be copied, modified, or distributed except according to those terms.
+
 import chronicles
 import ../spec/eth2_apis/eth2_rest_serialization,
        ../spec/datatypes/[phase0, altair]

beacon_chain/validator_client/attestation_service.nim

@@ -6,7 +6,7 @@
 # at your option. This file may not be copied, modified, or distributed except according to those terms.
 
 import std/sets
-import chronicles
+import metrics, chronicles
 import "."/[common, api, block_service]
 
 const
@@ -14,6 +14,16 @@ const
 
 logScope: service = ServiceName
 
+declareCounter beacon_attestations_sent,
+  "Number of attestations sent by the node"
+
+declareCounter beacon_aggregates_sent,
+  "Number of beacon chain attestations sent by the node"
+
+declareHistogram beacon_attestation_sent_delay,
+  "Time(s) between expected and actual attestation send moment",
+  buckets = DelayBuckets
+
 type
   AggregateItem* = object
     aggregator_index: uint64
@@ -108,6 +118,8 @@ proc serveAttestation(service: AttestationServiceRef, adata: AttestationData,
 
   let delay = vc.getDelay(adata.slot.attestation_deadline())
   if res:
+    beacon_attestations_sent.inc()
+    beacon_attestation_sent_delay.observe(delay.toFloatSeconds())
     notice "Attestation published", attestation = shortLog(attestation),
                                     validator = shortLog(validator),
                                     validator_index = vindex,
@@ -191,6 +203,7 @@ proc serveAggregateAndProof*(service: AttestationServiceRef,
       return false
 
   if res:
+    beacon_aggregates_sent.inc()
     notice "Aggregated attestation published",
            attestation = shortLog(signedProof.message.aggregate),
            validator = shortLog(validator),

beacon_chain/validator_client/block_service.nim

@@ -1,9 +1,23 @@
+# beacon_chain
+# Copyright (c) 2021-2022 Status Research & Development GmbH
+# Licensed and distributed under either of
+#   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
+#   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
+# at your option. This file may not be copied, modified, or distributed except according to those terms.
+
 import ".."/spec/forks
 import common, api
-import chronicles
+import chronicles, metrics
 
 logScope: service = "block_service"
 
+declareCounter beacon_blocks_sent,
+  "Number of beacon blocks sent by this node"
+
+declareHistogram beacon_blocks_sent_delay,
+  "Time(s) between expected and actual block send moment",
+  buckets = DelayBuckets
+
 proc publishBlock(vc: ValidatorClientRef, currentSlot, slot: Slot,
                   validator: AttachedValidator) {.async.} =
   let
@@ -113,6 +127,9 @@ proc publishBlock(vc: ValidatorClientRef, currentSlot, slot: Slot,
               err_name = exc.name, err_msg = exc.msg
         return
     if res:
+      let delay = vc.getDelay(slot.block_deadline())
+      beacon_blocks_sent.inc()
+      beacon_blocks_sent_delay.observe(delay.toFloatSeconds())
       notice "Block published", blockRoot = shortLog(blockRoot),
              blck = shortLog(beaconBlock), signature = shortLog(signature),
              validator = shortLog(validator)

beacon_chain/validator_client/common.nim

@@ -9,6 +9,7 @@ import std/[tables, os, sets, sequtils]
 import chronos, presto, presto/client as presto_client, chronicles, confutils,
        json_serialization/std/[options, net],
        stew/[base10, results, byteutils]
+import metrics, metrics/chronos_httpserver
 
 # Local modules
 import
@@ -36,6 +37,9 @@ const
   TIME_DELAY_FROM_SLOT* = 79.milliseconds
   SUBSCRIPTION_BUFFER_SLOTS* = 2'u64
 
+  DelayBuckets* = [-Inf, -4.0, -2.0, -1.0, -0.5, -0.1, -0.05,
+                   0.05, 0.1, 0.5, 1.0, 2.0, 4.0, 8.0, Inf]
+
 type
   ServiceState* {.pure.} = enum
     Initialized, Running, Error, Closing, Closed
@@ -137,6 +141,7 @@ type
 
   ValidatorClient* = object
     config*: ValidatorClientConf
+    metricsServer*: Option[MetricsHttpServerRef]
     graffitiBytes*: GraffitiBytes
     beaconNodes*: seq[BeaconNodeServerRef]
     fallbackService*: FallbackServiceRef

beacon_chain/validator_client/duties_service.nim

@@ -1,3 +1,10 @@
+# beacon_chain
+# Copyright (c) 2021-2022 Status Research & Development GmbH
+# Licensed and distributed under either of
+#   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
+#   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
+# at your option. This file may not be copied, modified, or distributed except according to those terms.
+
 import std/[sets, sequtils]
 import chronicles
 import common, api, block_service

beacon_chain/validator_client/fallback_service.nim

@@ -1,3 +1,10 @@
+# beacon_chain
+# Copyright (c) 2021-2022 Status Research & Development GmbH
+# Licensed and distributed under either of
+#   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
+#   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
+# at your option. This file may not be copied, modified, or distributed except according to those terms.
+
 import common
 
 const

beacon_chain/validator_client/fork_service.nim

@@ -1,3 +1,10 @@
+# beacon_chain
+# Copyright (c) 2021-2022 Status Research & Development GmbH
+# Licensed and distributed under either of
+#   * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
+#   * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
+# at your option. This file may not be copied, modified, or distributed except according to those terms.
+
 import std/algorithm
 import chronicles
 import common, api

beacon_chain/validator_client/sync_committee_service.nim

@@ -7,7 +7,7 @@
 
 import
   std/sets,
-  chronicles,
+  metrics, chronicles,
   "."/[common, api, block_service],
   ../spec/datatypes/[phase0, altair, bellatrix],
   ../spec/eth2_apis/rest_types
@@ -17,6 +17,16 @@ const
 
 logScope: service = ServiceName
 
+declareCounter beacon_sync_committee_messages_sent,
+  "Number of sync committee messages sent by the node"
+
+declareHistogram beacon_sync_committee_message_sent_delay,
+  "Time(s) between expected and actual sync committee message send moment",
+  buckets = DelayBuckets
+
+declareCounter beacon_sync_committee_contributions_sent,
+  "Number of sync committee contributions sent by the node"
+
 type
   ContributionItem* = object
     aggregator_index: uint64
@@ -82,6 +92,8 @@ proc serveSyncCommitteeMessage*(service: SyncCommitteeServiceRef,
 
   let delay = vc.getDelay(message.slot.sync_committee_message_deadline())
   if res:
+    beacon_sync_committee_messages_sent.inc()
+    beacon_sync_committee_message_sent_delay.observe(delay.toFloatSeconds())
     notice "Sync committee message published",
            message = shortLog(message),
            validator = shortLog(validator),
@@ -193,6 +205,7 @@ proc serveContributionAndProof*(service: SyncCommitteeServiceRef,
       false
 
   if res:
+    beacon_sync_committee_contributions_sent.inc()
     notice "Sync contribution published",
            validator = shortLog(validator),
            validator_index = validatorIdx