Add topology export trigger for party metrics (#2497)

The trigger is disabled by default.

[ci]

Signed-off-by: Simon Meier <[email protected]>

Author: meiersi-da

apps/app/src/test/scala/org/lfdecentralizedtrust/splice/integration/tests/ValidatorIntegrationTest.scala

@@ -12,6 +12,7 @@ import org.lfdecentralizedtrust.splice.integration.tests.SpliceTests.Integration
 import org.lfdecentralizedtrust.splice.util.WalletTestUtil
 import org.lfdecentralizedtrust.splice.validator.config.ValidatorAppBackendConfig
 import com.digitalasset.canton.config.CantonRequireTypes.InstanceName
+import com.digitalasset.canton.config.NonNegativeFiniteDuration
 import com.digitalasset.canton.config.RequireTypes.PositiveInt
 import com.digitalasset.canton.logging.SuppressionRule
 import com.digitalasset.canton.sequencing.SubmissionRequestAmplification
@@ -21,6 +22,7 @@ import org.apache.pekko.http.scaladsl.Http
 import org.apache.pekko.http.scaladsl.client.RequestBuilding.{Get, Post}
 import org.apache.pekko.http.scaladsl.model.StatusCodes
 import org.apache.pekko.http.scaladsl.model.headers.{Authorization, OAuth2BearerToken}
+import org.lfdecentralizedtrust.splice.config.ConfigTransforms
 import org.slf4j.event.Level
 
 import scala.concurrent.Future
@@ -70,6 +72,13 @@ class ValidatorIntegrationTest extends IntegrationTest with WalletTestUtil {
             )
           })
       })
+      // The topology metrics trigger is disabled by default.
+      // Enable it here to check that it starts and runs without errors
+      .addConfigTransform((_, config) =>
+        ConfigTransforms.updateAllAutomationConfigs(
+          _.copy(topologyMetricsPollingInterval = Some(NonNegativeFiniteDuration.ofSeconds(1)))
+        )(config)
+      )
 
   "start and restart cleanly" in { implicit env =>
     initDsoWithSv1Only()

apps/common/src/main/scala/org/lfdecentralizedtrust/splice/config/AutomationConfig.scala

@@ -46,6 +46,11 @@ case class AutomationConfig(
       */
     domainIngestionPollingInterval: NonNegativeFiniteDuration =
       NonNegativeFiniteDuration ofSeconds 30,
+    /** Polling interval to recompute and export topology metrics.
+      *
+      * Set to None to disable the topology metrics trigger.
+      */
+    topologyMetricsPollingInterval: Option[NonNegativeFiniteDuration] = None,
     /** Maximal number of retries that the time-based triggers retry transient failures w/o raising a warning.
       */
     maxNumSilentPollingRetries: Int = 3,

apps/validator/src/main/scala/org/lfdecentralizedtrust/splice/validator/ValidatorApp.scala

@@ -853,6 +853,7 @@ class ValidatorApp(
         config.maxVettingDelay,
         config.parameters,
         config.latestPackagesOnly,
+        metrics,
         loggerFactory,
       )
       _ <- MonadUtil.sequentialTraverse_(config.appInstances.toList)({ case (name, instance) =>

apps/validator/src/main/scala/org/lfdecentralizedtrust/splice/validator/automation/TopologyMetricsTrigger.scala

@@ -0,0 +1,46 @@
+// Copyright (c) 2024 Digital Asset (Switzerland) GmbH and/or its affiliates. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+package org.lfdecentralizedtrust.splice.validator.automation
+
+import com.digitalasset.canton.topology.store.TopologyStoreId
+import com.digitalasset.canton.topology.transaction.PartyToParticipant
+import com.digitalasset.canton.tracing.TraceContext
+import io.opentelemetry.api.trace.Tracer
+import org.lfdecentralizedtrust.splice.automation.{PollingTrigger, TriggerContext}
+import org.lfdecentralizedtrust.splice.environment.ParticipantAdminConnection
+import org.lfdecentralizedtrust.splice.environment.TopologyAdminConnection.TopologyResult
+import org.lfdecentralizedtrust.splice.scan.admin.api.client.ScanConnection
+import org.lfdecentralizedtrust.splice.validator.metrics.ValidatorAppMetrics
+
+import scala.concurrent.{ExecutionContext, Future}
+
+class TopologyMetricsTrigger(
+    override protected val context: TriggerContext,
+    metrics: ValidatorAppMetrics,
+    scanConnection: ScanConnection,
+    participantAdminConnection: ParticipantAdminConnection,
+)(implicit
+    override val ec: ExecutionContext,
+    override val tracer: Tracer,
+) extends PollingTrigger {
+
+  override def performWorkIfAvailable()(implicit
+      tc: TraceContext
+  ): Future[Boolean] =
+    for {
+      synchronizerId <- scanConnection.getAmuletRulesDomain()(tc)
+      partyMappings <- participantAdminConnection.listPartyToParticipant(
+        store = Some(TopologyStoreId.SynchronizerStore(synchronizerId))
+      )
+    } yield {
+      updateMetrics(partyMappings)
+      false
+    }
+
+  private def updateMetrics(
+      partyMappings: Seq[TopologyResult[PartyToParticipant]]
+  ) = {
+    metrics.numberOfPartiesGauge.updateValue(partyMappings.size.toDouble)
+  }
+}

apps/validator/src/main/scala/org/lfdecentralizedtrust/splice/validator/automation/ValidatorAutomationService.scala

@@ -42,6 +42,7 @@ import io.opentelemetry.api.trace.Tracer
 import monocle.Monocle.toAppliedFocusOps
 import org.apache.pekko.stream.Materializer
 import org.lfdecentralizedtrust.splice.store.AppStoreWithIngestion.SpliceLedgerConnectionPriority
+import org.lfdecentralizedtrust.splice.validator.metrics.ValidatorAppMetrics
 
 import java.nio.file.Path
 import scala.concurrent.ExecutionContextExecutor
@@ -77,6 +78,7 @@ class ValidatorAutomationService(
     maxVettingDelay: NonNegativeFiniteDuration,
     params: SpliceParametersConfig,
     latestPackagesOnly: Boolean,
+    metrics: ValidatorAppMetrics,
     override protected val loggerFactory: NamedLoggerFactory,
 )(implicit
     ec: ExecutionContextExecutor,
@@ -98,6 +100,19 @@ class ValidatorAutomationService(
       : org.lfdecentralizedtrust.splice.validator.automation.ValidatorAutomationService.type =
     ValidatorAutomationService
 
+  automationConfig.topologyMetricsPollingInterval.foreach(topologyPollingInterval =>
+    registerTrigger(
+      new TopologyMetricsTrigger(
+        triggerContext
+          .focus(_.config.pollingInterval)
+          .replace(topologyPollingInterval),
+        metrics,
+        scanConnection,
+        participantAdminConnection,
+      )
+    )
+  )
+
   walletManagerOpt.foreach { walletManager =>
     registerTrigger(
       new WalletAppInstallTrigger(

apps/validator/src/main/scala/org/lfdecentralizedtrust/splice/validator/metrics/ValidatorAppMetrics.scala

@@ -3,15 +3,28 @@
 
 package org.lfdecentralizedtrust.splice.validator.metrics
 
-import com.daml.metrics.api.MetricHandle.LabeledMetricsFactory
+import com.daml.metrics.api.MetricHandle.{Gauge, LabeledMetricsFactory}
+import com.daml.metrics.api.{MetricInfo, MetricsContext}
+import com.daml.metrics.api.MetricQualification.Saturation
 import org.lfdecentralizedtrust.splice.BaseSpliceMetrics
 import com.digitalasset.canton.metrics.DbStorageHistograms
 
 /** Modelled after [[com.digitalasset.canton.synchronizer.metrics.DomainMetrics]].
-  *
-  * This is only a bare-bones implementation so the code compiles so far.
   */
 class ValidatorAppMetrics(
     metricsFactory: LabeledMetricsFactory,
     storageHistograms: DbStorageHistograms,
-) extends BaseSpliceMetrics("validator", metricsFactory, storageHistograms) {}
+) extends BaseSpliceMetrics("validator", metricsFactory, storageHistograms) {
+
+  val numberOfPartiesGauge: Gauge[Double] =
+    metricsFactory.gauge[Double](
+      MetricInfo(
+        prefix :+ "synchronizer-topology-num-parties",
+        summary = "Total number of parties",
+        description = "The total number of parties allocated on the global synchronizer.",
+        qualification = Saturation,
+      ),
+      Double.NaN,
+    )(MetricsContext.Empty)
+
+}

docs/src/release_notes.rst

@@ -15,6 +15,13 @@ Upcoming
 
     - All app & UI images now use a non-root user.
 
+  - Validator
+
+     - Add a trigger to export the metric ``validator_synchronizer_topology_num_parties``
+       that counts the number of parties allocated on the Global Synchronizer.
+       The trigger does not run by default. To enable it, add the following to your validator app config:
+       ``canton.validator-apps.<your-validator-name>.automation.topology-metrics-polling-interval = 5m``.
+
   - SV
 
     - Deployment