chore(dashmate): expose metrics from Drive

dashpay · Apr 26, 2024 · d867fe1 · d867fe1
1 parent 3909b28
commit d867fe1
Show file tree

Hide file tree

Showing 10 changed files with 71 additions and 53 deletions.
diff --git a/packages/dashmate/configs/defaults/getBaseConfigFactory.js b/packages/dashmate/configs/defaults/getBaseConfigFactory.js
@@ -255,6 +255,11 @@ export default function getBaseConfigFactory(homeDir) {
               dkgInterval: 288,
               llmqSize: 400,
             },
+            metrics: {
+              enabled: false,
+              host: '127.0.0.1',
+              port: 29090,
+            },
             epochTime: 788400,
           },
           tenderdash: {

diff --git a/packages/dashmate/configs/getConfigFileMigrationsFactory.js b/packages/dashmate/configs/getConfigFileMigrationsFactory.js
@@ -524,51 +524,56 @@ export default function getConfigFileMigrationsFactory(homeDir, defaultConfigs)
       '1.0.0-dev.12': (configFile) => {
         Object.entries(configFile.configs)
           .forEach(([name, options]) => {
-            // Do nothing if it's already migrated for some reason
-            if (!options.platform.dapi.envoy) {
-              return;
+            // Update tenderdash config
+            options.platform.drive.tenderdash.docker.image = base.get('platform.drive.tenderdash.docker.image');
+            options.platform.drive.tenderdash.mempool.maxConcurrentCheckTx = base.get('platform.drive.tenderdash.mempool.maxConcurrentCheckTx');
+
+            // Add metrics to Drive ABCI
+            options.platform.drive.abci.metrics = base.get('platform.drive.abci.metrics');
+
+            // Envoy -> Gateway
+            if (options.platform.dapi.envoy) {
+              options.platform.gateway = lodash.cloneDeep(options.platform.dapi.envoy);
+
+              // add new options
+              options.platform.gateway.maxConnections = base.get('platform.gateway.maxConnections');
+              options.platform.gateway.maxHeapSizeInBytes = base.get('platform.gateway.maxHeapSizeInBytes');
+              options.platform.gateway.metrics = base.get('platform.gateway.metrics');
+              options.platform.gateway.admin = base.get('platform.gateway.admin');
+              options.platform.gateway.upstreams = base.get('platform.gateway.upstreams');
+              options.platform.gateway.log = base.get('platform.gateway.log');
+
+              // http -> listeners
+              options.platform.gateway.listeners = lodash.cloneDeep(
+                base.get('platform.gateway.listeners'),
+              );
+
+              options.platform.gateway.listeners.dapiAndDrive.host = options.platform.dapi.envoy
+                .http.host;
+              options.platform.gateway.listeners.dapiAndDrive.port = options.platform.dapi.envoy
+                .http.port;
+
+              delete options.platform.gateway.http;
+
+              // update rate limiter
+              options.platform.gateway.rateLimiter.docker = base.get('platform.gateway.rateLimiter.docker');
+              options.platform.gateway.rateLimiter.unit = base.get('platform.gateway.rateLimiter.unit');
+              options.platform.gateway.rateLimiter.requestsPerUnit = base.get('platform.gateway.rateLimiter.requestsPerUnit');
+              options.platform.gateway.rateLimiter.blacklist = base.get('platform.gateway.rateLimiter.blacklist');
+              options.platform.gateway.rateLimiter.whitelist = base.get('platform.gateway.rateLimiter.whitelist');
+              options.platform.gateway.rateLimiter.metrics = base.get('platform.gateway.rateLimiter.metrics');
+
+              delete options.platform.gateway.rateLimiter.fillInterval;
+              delete options.platform.gateway.rateLimiter.maxTokens;
+              delete options.platform.gateway.rateLimiter.tokensPerFill;
+
+              // delete envoy
+              delete options.platform.dapi.envoy;
+
+              // update image
+              options.platform.gateway.docker.image = base.get('platform.gateway.docker.image');
             }
 
-            options.platform.gateway = lodash.cloneDeep(options.platform.dapi.envoy);
-
-            // add new options
-            options.platform.gateway.maxConnections = base.get('platform.gateway.maxConnections');
-            options.platform.gateway.maxHeapSizeInBytes = base.get('platform.gateway.maxHeapSizeInBytes');
-            options.platform.gateway.metrics = base.get('platform.gateway.metrics');
-            options.platform.gateway.admin = base.get('platform.gateway.admin');
-            options.platform.gateway.upstreams = base.get('platform.gateway.upstreams');
-            options.platform.gateway.log = base.get('platform.gateway.log');
-
-            // http -> listeners
-            options.platform.gateway.listeners = lodash.cloneDeep(
-              base.get('platform.gateway.listeners'),
-            );
-
-            options.platform.gateway.listeners.dapiAndDrive.host = options.platform.dapi.envoy
-              .http.host;
-            options.platform.gateway.listeners.dapiAndDrive.port = options.platform.dapi.envoy
-              .http.port;
-
-            delete options.platform.gateway.http;
-
-            // update rate limiter
-            options.platform.gateway.rateLimiter.docker = base.get('platform.gateway.rateLimiter.docker');
-            options.platform.gateway.rateLimiter.unit = base.get('platform.gateway.rateLimiter.unit');
-            options.platform.gateway.rateLimiter.requestsPerUnit = base.get('platform.gateway.rateLimiter.requestsPerUnit');
-            options.platform.gateway.rateLimiter.blacklist = base.get('platform.gateway.rateLimiter.blacklist');
-            options.platform.gateway.rateLimiter.whitelist = base.get('platform.gateway.rateLimiter.whitelist');
-            options.platform.gateway.rateLimiter.metrics = base.get('platform.gateway.rateLimiter.metrics');
-
-            delete options.platform.gateway.rateLimiter.fillInterval;
-            delete options.platform.gateway.rateLimiter.maxTokens;
-            delete options.platform.gateway.rateLimiter.tokensPerFill;
-
-            // delete envoy
-            delete options.platform.dapi.envoy;
-
-            // update image
-            options.platform.gateway.docker.image = base.get('platform.gateway.docker.image');
-
             // rename non conventional field
             if (options.platform.drive.abci.tokioConsole.retention_secs) {
               options.platform.drive.abci.tokioConsole.retention = options.platform.drive.abci
@@ -603,10 +608,6 @@ export default function getConfigFileMigrationsFactory(homeDir, defaultConfigs)
                   fs.rmSync(oldFilePath, { recursive: true });
                 }
               }
-
-              // Update tenderdash config
-              options.platform.drive.tenderdash.docker.image = base.get('platform.drive.tenderdash.docker.image');
-              options.platform.drive.tenderdash.mempool.maxConcurrentCheckTx = base.get('platform.drive.tenderdash.mempool.maxConcurrentCheckTx');
             }
           });
 

diff --git a/packages/dashmate/docker-compose.yml b/packages/dashmate/docker-compose.yml
@@ -72,7 +72,7 @@ services:
       - DB_PATH=/var/lib/dash/rs-drive-abci/db
       - ABCI_CONSENSUS_BIND_ADDRESS=tcp://0.0.0.0:26658
       - GRPC_BIND_ADDRESS=0.0.0.0:26670
-      - PROMETHEUS_BIND_ADDRESS=http://0.0.0.0:29090
+      - PROMETHEUS_BIND_ADDRESS=${PLATFORM_DRIVE_ABCI_METRICS_URL}
       - TOKIO_CONSOLE_ENABLED=${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_ENABLED:?err}
       - TOKIO_CONSOLE_ADDRESS=0.0.0.0:${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_PORT:?err}
       - TOKIO_CONSOLE_RETENTION_SECS=${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_RETENTION:?err}
@@ -81,14 +81,14 @@ services:
       - 26658
       - 26659
       - 26670
-      - 29090
     env_file:
       # Logger settings
       - ${DASHMATE_HOME_DIR:?err}/${CONFIG_NAME:?err}/platform/drive/abci/logger.env
     profiles:
       - platform
     ports:
       - ${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_HOST:?err}:${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_PORT:?err}:${PLATFORM_DRIVE_ABCI_TOKIO_CONSOLE_PORT:?err}
+      - ${PLATFORM_DRIVE_ABCI_METRICS_HOST:?err}:${PLATFORM_DRIVE_ABCI_METRICS_PORT:?err}:29090
 
   drive_tenderdash:
     image: ${PLATFORM_DRIVE_TENDERDASH_DOCKER_IMAGE:?err}

diff --git a/packages/dashmate/src/config/configJsonSchema.js b/packages/dashmate/src/config/configJsonSchema.js
@@ -806,9 +806,12 @@ export default {
                   type: 'integer',
                   minimum: 180,
                 },
+                metrics: {
+                  $ref: '#/definitions/enabledHostPort',
+                },
               },
               additionalProperties: false,
-              required: ['docker', 'logs', 'tokioConsole', 'validatorSet', 'chainLock', 'epochTime'],
+              required: ['docker', 'logs', 'tokioConsole', 'validatorSet', 'chainLock', 'epochTime', 'metrics'],
             },
             tenderdash: {
               type: 'object',

diff --git a/packages/dashmate/src/config/generateEnvsFactory.js b/packages/dashmate/src/config/generateEnvsFactory.js
@@ -76,6 +76,11 @@ export default function generateEnvsFactory(configFile, homeDir, getConfigProfil
       tenderdashLogDirectoryPath = path.dirname(tenderdashLogFilePath);
     }
 
+    let driveAbciMetricsUrl = '';
+    if (config.get('platform.drive.abci.metrics.enabled')) {
+      driveAbciMetricsUrl = 'http://0.0.0.0:29090';
+    }
+
     return {
       DASHMATE_HOME_DIR: homeDir.getPath(),
       LOCAL_UID: uid,
@@ -93,6 +98,7 @@ export default function generateEnvsFactory(configFile, homeDir, getConfigProfil
       DASHMATE_HELPER_DOCKER_IMAGE,
       PLATFORM_DRIVE_TENDERDASH_LOG_DIRECTORY_PATH: tenderdashLogDirectoryPath,
       PLATFORM_GATEWAY_RATE_LIMITER_METRICS_DISABLED: config.get('platform.gateway.rateLimiter.metrics.enabled'),
+      PLATFORM_DRIVE_ABCI_METRICS_URL: driveAbciMetricsUrl,
       ...convertObjectToEnvs(config.getOptions()),
     };
   }

diff --git a/packages/dashmate/src/listr/tasks/setup/setupLocalPresetTaskFactory.js b/packages/dashmate/src/listr/tasks/setup/setupLocalPresetTaskFactory.js
@@ -223,6 +223,7 @@ export default function setupLocalPresetTaskFactory(
                   config.set('platform.drive.tenderdash.node.key', key);
 
                   config.set('platform.drive.abci.tokioConsole.port', config.get('platform.drive.abci.tokioConsole.port') + (i * 100));
+                  config.set('platform.drive.abci.metrics.port', config.get('platform.drive.abci.metrics.port') + (i * 100));
                   config.set('platform.gateway.admin.port', config.get('platform.gateway.admin.port') + (i * 100));
                   config.set('platform.gateway.listeners.dapiAndDrive.port', config.get('platform.gateway.listeners.dapiAndDrive.port') + (i * 100));
                   config.set('platform.gateway.metrics.port', config.get('platform.gateway.metrics.port') + (i * 100));

diff --git a/packages/rs-drive-abci/.env.local b/packages/rs-drive-abci/.env.local
@@ -1,6 +1,6 @@
 # ABCI host and port to listen
 ABCI_CONSENSUS_BIND_ADDRESS="tcp://0.0.0.0:26658"
-PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
+#PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
 GRPC_BIND_ADDRESS="0.0.0.0:26670"
 
 # stderr logging for humans

diff --git a/packages/rs-drive-abci/.env.mainnet b/packages/rs-drive-abci/.env.mainnet
@@ -1,6 +1,6 @@
 # ABCI host and port to listen
 ABCI_CONSENSUS_BIND_ADDRESS="tcp://0.0.0.0:26658"
-PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
+#PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
 GRPC_BIND_ADDRESS="0.0.0.0:26670"
 
 # stderr logging for humans

diff --git a/packages/rs-drive-abci/.env.testnet b/packages/rs-drive-abci/.env.testnet
@@ -1,6 +1,6 @@
 # ABCI host and port to listen
 ABCI_CONSENSUS_BIND_ADDRESS="tcp://0.0.0.0:26658"
-PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
+#PROMETHEUS_BIND_ADDRESS="http://0.0.0.0:29090"
 GRPC_BIND_ADDRESS="0.0.0.0:26670"
 
 # stderr logging for humans

diff --git a/packages/rs-drive-abci/src/main.rs b/packages/rs-drive-abci/src/main.rs
@@ -262,6 +262,8 @@ fn start_prometheus(config: &PlatformConfig) -> Result<Option<Prometheus>, Strin
         .filter(|s| !s.is_empty());
 
     if let Some(addr) = prometheus_addr {
+        tracing::info!("Expose prometheus metrics on {}", addr);
+
         let addr = url::Url::parse(&addr).map_err(|e| e.to_string())?;
         Ok(Some(Prometheus::new(addr).map_err(|e| e.to_string())?))
     } else {