fix(metrics): removed influx and added l2_state_size data (#196)

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## Next release
 
+- fix(metrics): removed influx and added l2_state_size data
 - fix: command to start the Madara client
 - refactor: database error unification
 - feat: raise file-descriptor limit

crates/client/db/Cargo.toml

@@ -14,6 +14,7 @@ targets = ["x86_64-unknown-linux-gnu"]
 [dependencies]
 
 # Deoxys
+dc-metrics = { workspace = true }
 dp-block = { workspace = true }
 dp-class = { workspace = true }
 dp-convert = { workspace = true }

crates/client/db/src/db_metrics.rs

@@ -0,0 +1,15 @@
+use dc_metrics::{IntGaugeVec, MetricsRegistry, Opts, PrometheusError};
+
+#[derive(Clone, Debug)]
+pub struct DbMetrics {
+    pub column_sizes: IntGaugeVec,
+}
+
+impl DbMetrics {
+    pub fn register(registry: &MetricsRegistry) -> Result<Self, PrometheusError> {
+        Ok(Self {
+            column_sizes: registry
+                .register(IntGaugeVec::new(Opts::new("column_sizes", "Sizes of RocksDB columns"), &["column"])?)?,
+        })
+    }
+}

crates/client/db/src/lib.rs

@@ -10,6 +10,7 @@ use block_db::ChainInfo;
 use bonsai_db::{BonsaiDb, DatabaseKeyMapping};
 use bonsai_trie::id::BasicId;
 use bonsai_trie::{BonsaiStorage, BonsaiStorageConfig};
+use db_metrics::DbMetrics;
 use rocksdb::backup::{BackupEngine, BackupEngineOptions};
 
 pub mod block_db;
@@ -23,6 +24,7 @@ pub mod bonsai_db;
 pub mod class_db;
 pub mod contract_db;
 pub mod db_block_id;
+pub mod db_metrics;
 pub mod storage_updates;
 
 pub use error::{DeoxysStorageError, TrieType};
@@ -447,6 +449,21 @@ impl DeoxysBackend {
             log: Column::BonsaiClassesLog,
         })
     }
+
+    pub fn get_storage_size(&self, db_metrics: &DbMetrics) -> u64 {
+        let mut storage_size = 0;
+
+        for &column in Column::ALL.iter() {
+            let cf_handle = self.db.get_column(column);
+            let cf_metadata = self.db.get_column_family_metadata_cf(&cf_handle);
+            let column_size = cf_metadata.size;
+            storage_size += column_size;
+
+            db_metrics.column_sizes.with_label_values(&[column.rocksdb_name()]).set(column_size as i64);
+        }
+
+        storage_size
+    }
 }
 
 pub mod bonsai_identifier {

crates/client/metrics/src/lib.rs

@@ -15,7 +15,7 @@ pub use prometheus::{
         AtomicF64 as F64, AtomicI64 as I64, AtomicU64 as U64, GenericCounter as Counter,
         GenericCounterVec as CounterVec, GenericGauge as Gauge, GenericGaugeVec as GaugeVec,
     },
-    exponential_buckets, Error as PrometheusError, Histogram, HistogramOpts, HistogramVec, Opts, Registry,
+    exponential_buckets, Error as PrometheusError, Histogram, HistogramOpts, HistogramVec, IntGaugeVec, Opts, Registry,
 };
 
 #[derive(Clone, Debug)]

crates/client/sync/src/l2.rs

@@ -5,6 +5,7 @@ use std::sync::{Arc, Mutex};
 use std::time::Instant;
 
 use anyhow::{bail, Context};
+use dc_db::db_metrics::DbMetrics;
 use dc_db::DeoxysBackend;
 use dc_db::DeoxysStorageError;
 use dc_telemetry::{TelemetryHandle, VerbosityLevel};
@@ -59,6 +60,7 @@ async fn l2_verify_and_apply_task(
     verify: bool,
     backup_every_n_blocks: Option<u64>,
     block_metrics: BlockMetrics,
+    db_metrics: DbMetrics,
     starting_block: u64,
     sync_timer: Arc<Mutex<Option<Instant>>>,
     telemetry: TelemetryHandle,
@@ -118,7 +120,16 @@ async fn l2_verify_and_apply_task(
         })
         .await?;
 
-        update_sync_metrics(block_n, &block_header, starting_block, &block_metrics, sync_timer.clone()).await?;
+        update_sync_metrics(
+            block_n,
+            &block_header,
+            starting_block,
+            &block_metrics,
+            &db_metrics,
+            sync_timer.clone(),
+            &backend,
+        )
+        .await?;
 
         let sw = PerfStopwatch::new();
         if backend.maybe_flush(false)? {
@@ -287,11 +298,13 @@ pub struct L2SyncConfig {
 }
 
 /// Spawns workers to fetch blocks and state updates from the feeder.
+#[allow(clippy::too_many_arguments)]
 pub async fn sync(
     backend: &Arc<DeoxysBackend>,
     provider: SequencerGatewayProvider,
     config: L2SyncConfig,
     block_metrics: BlockMetrics,
+    db_metrics: DbMetrics,
     starting_block: u64,
     chain_id: Felt,
     telemetry: TelemetryHandle,
@@ -329,6 +342,7 @@ pub async fn sync(
         config.verify,
         config.backup_every_n_blocks,
         block_metrics,
+        db_metrics,
         starting_block,
         Arc::clone(&sync_timer),
         telemetry,
@@ -353,7 +367,9 @@ async fn update_sync_metrics(
     block_header: &Header,
     starting_block: u64,
     block_metrics: &BlockMetrics,
+    db_metrics: &DbMetrics,
     sync_timer: Arc<Mutex<Option<Instant>>>,
+    backend: &DeoxysBackend,
 ) -> anyhow::Result<()> {
     // Update Block sync time metrics
     let elapsed_time;
@@ -381,6 +397,12 @@ async fn update_sync_metrics(
     block_metrics.l1_gas_price_wei.set(f64::from_u128(block_header.l1_gas_price.eth_l1_gas_price).unwrap_or(0f64));
     block_metrics.l1_gas_price_strk.set(f64::from_u128(block_header.l1_gas_price.strk_l1_gas_price).unwrap_or(0f64));
 
+    if block_number % 200 == 0 {
+        let storage_size = backend.get_storage_size(db_metrics);
+        let size_gb = storage_size as f64 / (1024 * 1024 * 1024) as f64;
+        block_metrics.l2_state_size.set(size_gb);
+    }
+
     Ok(())
 }
 

crates/client/sync/src/lib.rs

@@ -20,7 +20,7 @@ pub mod starknet_sync_worker {
     use std::{sync::Arc, time::Duration};
 
     use anyhow::Context;
-    use dc_db::DeoxysBackend;
+    use dc_db::{db_metrics::DbMetrics, DeoxysBackend};
     use dc_telemetry::TelemetryHandle;
     use reqwest::Url;
     use starknet_providers::SequencerGatewayProvider;
@@ -38,6 +38,7 @@ pub mod starknet_sync_worker {
         starting_block: Option<u64>,
         backup_every_n_blocks: Option<u64>,
         block_metrics: BlockMetrics,
+        db_metrics: DbMetrics,
         chain_id: Felt,
         telemetry: TelemetryHandle,
         pending_block_poll_interval: Duration,
@@ -89,6 +90,7 @@ pub mod starknet_sync_worker {
                     pending_block_poll_interval,
                 },
                 block_metrics,
+                db_metrics,
                 starting_block,
                 chain_id,
                 telemetry,

crates/client/sync/src/metrics/block_metrics.rs

@@ -7,6 +7,7 @@ pub struct BlockMetrics {
     pub l2_sync_time: Gauge<F64>,
     pub l2_avg_sync_time: Gauge<F64>,
     pub l2_latest_sync_time: Gauge<F64>,
+    pub l2_state_size: Gauge<F64>,
     pub transaction_count: Gauge<F64>,
     pub event_count: Gauge<F64>,
     // L1 network metrics
@@ -25,6 +26,8 @@ impl BlockMetrics {
                 .register(Gauge::new("deoxys_l2_avg_sync_time", "Gauge for deoxys L2 average sync time")?)?,
             l2_latest_sync_time: registry
                 .register(Gauge::new("deoxys_l2_latest_sync_time", "Gauge for deoxys L2 latest sync time")?)?,
+            l2_state_size: registry
+                .register(Gauge::new("deoxys_l2_state_size", "Gauge for node storage usage in GB")?)?,
             l1_block_number: registry
                 .register(Gauge::new("deoxys_l1_block_number", "Gauge for deoxys L1 block number")?)?,
             transaction_count: registry

crates/node/src/service/sync.rs

@@ -2,6 +2,7 @@ use std::sync::Arc;
 use std::time::Duration;
 
 use anyhow::Context;
+use dc_db::db_metrics::DbMetrics;
 use dc_db::{DatabaseService, DeoxysBackend};
 use dc_metrics::MetricsRegistry;
 use dc_sync::fetch::fetchers::FetchConfig;
@@ -24,6 +25,7 @@ pub struct SyncService {
     l1_core_address: H160,
     starting_block: Option<u64>,
     block_metrics: BlockMetrics,
+    db_metrics: DbMetrics,
     chain_id: Felt,
     start_params: Option<TelemetryHandle>,
     disabled: bool,
@@ -38,6 +40,7 @@ impl SyncService {
         telemetry: TelemetryHandle,
     ) -> anyhow::Result<Self> {
         let block_metrics = BlockMetrics::register(&metrics_handle)?;
+        let db_metrics = DbMetrics::register(&metrics_handle)?;
         let fetch_config = config.block_fetch_config();
 
         let l1_endpoint = if !config.sync_l1_disabled {
@@ -59,6 +62,7 @@ impl SyncService {
             starting_block: config.starting_block,
             backup_every_n_blocks: config.backup_every_n_blocks,
             block_metrics,
+            db_metrics,
             chain_id: config.network.chain_id(),
             start_params: Some(telemetry),
             disabled: config.sync_disabled,
@@ -76,6 +80,7 @@ impl SyncService {
             l1_core_address,
             starting_block,
             block_metrics,
+            db_metrics,
             chain_id,
             pending_block_poll_interval,
             ..
@@ -92,6 +97,7 @@ impl SyncService {
                 starting_block,
                 backup_every_n_blocks,
                 block_metrics,
+                db_metrics,
                 chain_id,
                 telemetry,
                 pending_block_poll_interval,

infra/docker-compose.yml

@@ -93,36 +93,6 @@ services:
       - ./grafana/dashboards:/var/lib/grafana/dashboards
       - ./grafana/datasources/datasource.yml:/etc/grafana/provisioning/datasources/main.yml
 
-  influxdb:
-    container_name: influxdb
-    image: influxdb:latest
-    restart: unless-stopped
-    ports:
-      - "8086:8086"
-    volumes:
-      - ./influxdb:/var/lib/influxdb
-    environment:
-      - INFLUXDB_DB=mydb
-      - INFLUXDB_HTTP_AUTH_ENABLED=true
-      - INFLUXDB_ADMIN_USER=admin
-      - INFLUXDB_ADMIN_PASSWORD=adminpassword
-      - INFLUXDB_USER=user
-      - INFLUXDB_USER_PASSWORD=userpassword
-
-  telegraf:
-    container_name: telegraf
-    image: telegraf:latest
-    restart: unless-stopped
-    volumes:
-      - ./telegraf/etc/telegraf.conf:/etc/telegraf/telegraf.conf:ro
-      - /var/lib/deoxys:/var/lib/deoxys
-    depends_on:
-      - influxdb
-    links:
-      - influxdb
-    ports:
-      - "8125:8125"
-
 networks:
   loki-net:
     external: true