use dedicated agent

Author: margaretjgu

.buildkite/benchmark.sh

@@ -5,6 +5,11 @@ set -euo pipefail
 repo_pwd="$PWD"
 mkdir -p benchmark-output/{pr,base}
 
+warmup_cpu() {
+  node -e "let s=0; for(let i=0;i<5e7;i++)s+=Math.sqrt(i); console.log('CPU warm')" > /dev/null
+  sleep 2
+}
+
 run_benchmark() {
   local target="$1"
 
@@ -14,6 +19,7 @@ run_benchmark() {
   fi
 
   npm install --silent
+  warmup_cpu
   npm run benchmark
   mv benchmark*.json "$repo_pwd/benchmark-output/$target"
 

.buildkite/benchmark.yml

@@ -1,9 +1,9 @@
 ---
 agents:
-  memory: "2G"
+  memory: "4G"
 
 steps:
-  - label: ":elasticsearch: :javascript: ES JavaScript Transport Benchmark (Linux)"
+  - label: "ES JavaScript Transport Benchmark (Linux)"
     command: ./.buildkite/benchmark.sh
     agents:
-      image: node:lts
+      image: node:22

scripts/compare-benchmark-json.mjs

@@ -305,14 +305,14 @@ function generateSummary() {
   let summary = `## Summary\n\n`
 
   if (regressions.failures.length === 0 && regressions.warnings.length === 0) {
-    summary += `**PASSED** - No significant performance regressions detected.\n\n`
+    summary += `No significant changes detected.\n\n`
     return summary
   }
 
   if (regressions.failures.length > 0) {
-    summary += `**FAILED** - ${regressions.failures.length} regression(s) detected\n\n`
-    summary += '| Metric | Regression |\n'
-    summary += '|--------|------------|\n'
+    summary += `**${regressions.failures.length} metric(s) exceeded threshold**\n\n`
+    summary += '| Metric | Change |\n'
+    summary += '|--------|--------|\n'
     for (const { metric, change } of regressions.failures) {
       summary += `| ${metric} | ${change >= 0 ? '+' : ''}${change.toFixed(1)}% |\n`
     }
@@ -329,14 +329,6 @@ function generateSummary() {
     summary += '\n'
   }
 
-  summary += `### Thresholds\n\n`
-  summary += `| Type | Warning | Failure |\n`
-  summary += `|------|---------|---------|\n`
-  summary += `| Latency | >${THRESHOLDS.latency.warning}% slower | >${THRESHOLDS.latency.failure}% slower |\n`
-  summary += `| Throughput | >${Math.abs(THRESHOLDS.throughput.warning)}% slower | >${Math.abs(THRESHOLDS.throughput.failure)}% slower |\n`
-  summary += `| Memory | >${THRESHOLDS.memory.warning}% larger | >${THRESHOLDS.memory.failure}% larger |\n`
-  summary += `| GC | >${THRESHOLDS.gc.warning}% more | >${THRESHOLDS.gc.failure}% more |\n\n`
-
   return summary
 }
 

test/benchmark/benchmark.mjs

@@ -3,7 +3,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { writeFileSync } from 'node:fs'
-import { run, bench, group } from 'mitata'
+import { run, bench, group, configure } from 'mitata'
 import {
   Transport,
   WeightedConnectionPool,
@@ -13,6 +13,11 @@ import {
   UndiciConnection
 } from '../../index.js'
 
+configure({
+  warmup: { iterations: 1000, time: 500_000_000 },
+  run: { iterations: 5000, time: 1_000_000_000 }
+})
+
 group('Transport#constructor - UndiciConnection', () => {
   bench('WeightedConnectionPool', () => {
     const connectionPool = new WeightedConnectionPool({ Connection: UndiciConnection })
@@ -59,7 +64,7 @@ group('Transport#constructor - HttpConnection', () => {
   .gc('inner')
 })
 
-const { layout, benchmarks } = await run()
+const { layout, benchmarks } = await run({ format: 'quiet' })
 
 const output = {}