fix: limit the number of buckets for prom metrics (#114)

Author: OlivierDehaene

load_tests/load.js

@@ -2,7 +2,7 @@ import {check} from 'k6';
 import http from 'k6/http';
 import {Trend} from 'k6/metrics';
 
-const host = __ENV.HOST || '127.0.0.1:3000';
+const host = __ENV.HOST || '127.0.0.1:8080';
 
 const totalTime = new Trend('total_time', true);
 const tokenizationTIme = new Trend('tokenization_time', true);
@@ -41,7 +41,7 @@ export default function () {
     });
 
     const headers = {'Content-Type': 'application/json'};
-    const res = http.post(`http://${host}`, payload, {
+    const res = http.post(`http://${host}/predict`, payload, {
         headers, timeout: '20m'
     });
 

load_tests/load_grpc.js

@@ -2,7 +2,7 @@ import {check} from 'k6';
 import grpc from 'k6/experimental/grpc';
 import {Trend} from 'k6/metrics';
 
-const host = __ENV.HOST || '127.0.0.1:3000';
+const host = __ENV.HOST || '127.0.0.1:8080';
 
 const totalTime = new Trend('total_time', true);
 const tokenizationTIme = new Trend('tokenization_time', true);

load_tests/load_grpc_stream.js

@@ -1,7 +1,7 @@
 import grpc from 'k6/experimental/grpc';
 import {Counter, Trend} from 'k6/metrics';
 
-const host = __ENV.HOST || '127.0.0.1:3000';
+const host = __ENV.HOST || '127.0.0.1:8080';
 
 const streamCounter = new Counter('stream_counter');
 const totalTime = new Trend('total_time', true);

router/src/prometheus.rs

@@ -15,17 +15,18 @@ pub(crate) fn prometheus_builer(max_input_length: usize) -> Result<PrometheusBui
 
     // Input Length buckets
     let input_length_matcher = Matcher::Full(String::from("te_request_input_length"));
-    let input_length_buckets: Vec<f64> = (0..100)
-        .map(|x| (max_input_length as f64 / 100.0) * (x + 1) as f64)
+    let input_length_buckets: Vec<f64> = (0..20)
+        .map(|x| 2.0_f64.powi(x))
+        .filter(|x| (*x as usize) <= max_input_length)
         .collect();
 
     // Batch size buckets
     let batch_size_matcher = Matcher::Full(String::from("te_batch_next_size"));
-    let batch_size_buckets: Vec<f64> = (0..2048).map(|x| (x + 1) as f64).collect();
+    let batch_size_buckets: Vec<f64> = (0..13).map(|x| 2.0_f64.powi(x)).collect();
 
     // Batch tokens buckets
     let batch_tokens_matcher = Matcher::Full(String::from("te_batch_next_tokens"));
-    let batch_tokens_buckets: Vec<f64> = (0..100_000).map(|x| (x + 1) as f64).collect();
+    let batch_tokens_buckets: Vec<f64> = (0..21).map(|x| 2.0_f64.powi(x)).collect();
 
     // Prometheus handler
     PrometheusBuilder::new()