PGO: optimize collected profile file size (#14256) (#14267)

(cherry picked from commit 3833927) # Conflicts: # .github/workflows/benchmarks.yml # systemtest/benchtest/profiles.go # testing/benchmark/Makefile Co-authored-by: Kostiantyn Masliuk <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
elastic · Oct 11, 2024 · 7097284 · 7097284
1 parent 05081aa
commit 7097284
Show file tree

Hide file tree

Showing 3 changed files with 32 additions and 12 deletions.
diff --git a/.github/workflows/benchmarks.yml b/.github/workflows/benchmarks.yml
@@ -226,7 +226,7 @@ jobs:
           git_commit_gpgsign: true
 
       - name: Open PGO PR
-        if: ${{ env.RUN_STANDALONE == 'true' && github.ref == 'refs/heads/main' }}
+        if: ${{ env.RUN_STANDALONE == 'true' }}
         run: ${{ github.workspace }}/.ci/scripts/push-pgo-pr.sh
         env:
           WORKSPACE_PATH: ${{ github.workspace }}

diff --git a/systemtest/benchtest/profiles.go b/systemtest/benchtest/profiles.go
@@ -18,9 +18,11 @@
 package benchtest
 
 import (
+	"compress/gzip"
 	"context"
 	"fmt"
 	"io"
+	"math/rand/v2"
 	"net/http"
 	"os"
 	"strconv"
@@ -88,12 +90,30 @@ func (p *profiles) recordCPU() error {
 	if benchConfig.CPUProfile == "" {
 		return nil
 	}
-	duration := benchConfig.Benchtime
-	profile, err := fetchProfile("/debug/pprof/profile", duration)
-	if err != nil {
-		return fmt.Errorf("failed to fetch CPU profile: %w", err)
+	// Limit profiling time to random 5% of overall time.
+	// This should not seriously affect the profile quality,
+	// since we merge the final profile form multiple sources,
+	// but prevent profile size from swelling.
+	var done bool
+	const tickets = 20
+	duration := benchConfig.Benchtime / tickets
+	for i := range tickets {
+		if done || (rand.N(tickets-i)+i+1) < tickets {
+			time.Sleep(duration)
+			continue
+		}
+		profile, err := fetchProfile("/debug/pprof/profile", duration)
+		if err != nil {
+			return fmt.Errorf("failed to fetch CPU profile: %w", err)
+		}
+		// We don't need the address in the profile, so discard it to reduce the size.
+		if err := profile.Aggregate(true, true, true, true, false); err != nil {
+			return fmt.Errorf("failed to fetch CPU profile: %w", err)
+		}
+		profile = profile.Compact()
+		p.cpu = append(p.cpu, profile)
+		done = true
 	}
-	p.cpu = append(p.cpu, profile)
 	return nil
 }
 
@@ -168,14 +188,15 @@ func (p *profiles) writeDeltas(filename string, deltas []*profile.Profile) error
 		return err
 	}
 	defer f.Close()
-	return merged.Write(f)
+	w, err := gzip.NewWriterLevel(f, gzip.BestCompression)
+	if err != nil {
+		return err
+	}
+	defer w.Close()
+	return merged.WriteUncompressed(w)
 }
 
 func (p *profiles) mergeBenchmarkProfiles(profiles []*profile.Profile) (*profile.Profile, error) {
-	for i, profile := range profiles {
-		benchmarkName := p.benchmarkNames[i]
-		profile.SetLabel("benchmark", []string{benchmarkName})
-	}
 	merged, err := profile.Merge(profiles)
 	if err != nil {
 		return nil, fmt.Errorf("error merging profiles: %w", err)

diff --git a/testing/benchmark/Makefile b/testing/benchmark/Makefile
@@ -30,7 +30,6 @@ SSH_OPTS ?= -o LogLevel=ERROR -o StrictHostKeyChecking=no -o ServerAliveInterval
 SSH_KEY ?= ~/.ssh/id_rsa_terraform
 WORKER_IP = $(shell terraform output -raw public_ip)
 APM_SERVER_IP = $(shell terraform output -raw apm_server_ip)
-RUN_STANDALONE = $(shell echo var.run_standalone | terraform console | tr -d '"')
 
 SHELL = /bin/bash
 .SHELLFLAGS = -o pipefail -c