Make the time_elapsed_db_query to be a histogram vec

Author: maximopalopoli

aggregation_mode/gateway/src/http.rs

@@ -260,7 +260,7 @@ impl GatewayServer {
                 let duration = start.elapsed();
                 state
                     .metrics
-                    .register_db_response_time_post(duration.as_secs_f64());
+                    .register_db_response_time_post("sp1-post", duration.as_secs_f64());
 
                 HttpResponse::Ok().json(AppResponse::new_sucessfull(
                     serde_json::json!({ "task_id": task_id.to_string() }),

aggregation_mode/gateway/src/metrics.rs

@@ -1,30 +1,30 @@
 use actix_web::{web, App, HttpResponse, HttpServer, Responder};
-use prometheus::{self, histogram_opts, Encoder, Histogram, Registry, TextEncoder};
+use prometheus::{self, histogram_opts, Encoder, HistogramVec, Registry, TextEncoder};
 use std::sync::Arc;
 
 #[derive(Clone, Debug)]
 pub struct GatewayMetrics {
     pub registry: Registry,
-    pub time_elapsed_db_post: Histogram,
+    pub time_elapsed_db_query: HistogramVec,
 }
 
 impl GatewayMetrics {
     pub fn start(metrics_port: u16) -> Result<Arc<Self>, prometheus::Error> {
         let registry = Registry::new();
 
-        let time_elapsed_db_post = Histogram::with_opts(histogram_opts!(
-            "time_elapsed_db_post",
-            "Time elapsed in DB posts"
-        ))?;
+        let time_elapsed_db_query = HistogramVec::new(
+            histogram_opts!("time_elapsed_db_query", "Time elapsed in DB posts"),
+            &["query"],
+        )?;
 
-        registry.register(Box::new(time_elapsed_db_post.clone()))?;
+        registry.register(Box::new(time_elapsed_db_query.clone()))?;
 
         // Arc is used because metrics are a shared resource accessed by both the background and metrics HTTP
         // server and the application code, across multiple Actix worker threads. The server outlives start(),
         // so the data must be static and safely shared between threads.
         let metrics = Arc::new(Self {
             registry,
-            time_elapsed_db_post,
+            time_elapsed_db_query,
         });
 
         let server_metrics = metrics.clone();
@@ -57,7 +57,9 @@ impl GatewayMetrics {
             .body(buffer)
     }
 
-    pub fn register_db_response_time_post(&self, value: f64) {
-        self.time_elapsed_db_post.observe(value);
+    pub fn register_db_response_time_post(&self, query: &str, value: f64) {
+        self.time_elapsed_db_query
+            .with_label_values(&[query])
+            .observe(value);
     }
 }

grafana/provisioning/dashboards/aligned/aggregation_mode_gateway.json

@@ -18,7 +18,7 @@
   "editable": true,
   "fiscalYearStartMonth": 0,
   "graphTooltip": 0,
-  "id": 2,
+  "id": 8,
   "links": [],
   "liveNow": false,
   "panels": [
@@ -388,7 +388,7 @@
             "uid": "prometheus"
           },
           "editorMode": "code",
-          "expr": "histogram_quantile(0.50, sum by (le) (rate(time_elapsed_db_post_bucket[$__rate_interval])))",
+          "expr": "histogram_quantile(0.50, sum by (le) (rate(time_elapsed_db_query_bucket{query=\"sp1-post\"}[$__rate_interval])))",
           "instant": false,
           "legendFormat": "__auto",
           "range": true,
@@ -576,7 +576,7 @@
             "uid": "prometheus"
           },
           "editorMode": "code",
-          "expr": "sum(rate(time_elapsed_db_post_count[$__rate_interval]))",
+          "expr": "sum(rate(time_elapsed_db_query_count[$__rate_interval]))",
           "instant": false,
           "legendFormat": "__auto",
           "range": true,
@@ -602,6 +602,6 @@
   "timezone": "",
   "title": "Aggregation Mode",
   "uid": "a66a5480-6a60-4b87-9d29-4f0f446edafd",
-  "version": 12,
+  "version": 1,
   "weekStart": ""
 }