Integrate metricsviz library in gVisor benchmarks.

This is part of a series of changes to add metric charts in performance
benchmarks.

All benchmarks now auto-detect if profiling metrics are enabled in the test
runtime. If profiling metrics data is present, it will generate charts of
the data.

PiperOrigin-RevId: 633094699

test/benchmarks/base/BUILD

@@ -26,6 +26,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/base",
         "//test/benchmarks/harness",
+        "//test/metricsviz",
     ],
 )
 
@@ -38,6 +39,7 @@ benchmark_test(
         "//test/benchmarks/base",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -49,6 +51,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -60,6 +63,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -71,6 +75,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -82,6 +87,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
         "@com_github_docker_docker//api/types:go_default_library",
     ],
 )

test/benchmarks/base/hackbench_test.go

@@ -22,6 +22,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarHackbench runs hackbench on the runtime.
@@ -69,6 +70,7 @@ func BenchmarkHackbench(b *testing.B) {
 			); err != nil {
 				b.Fatalf("run failed with: %v", err)
 			}
+			defer metricsviz.FromContainerLogs(ctx, b, container)
 
 			cmd := tc.MakeCmd(b)
 			b.ResetTimer()

test/benchmarks/base/size_test.go

@@ -24,6 +24,7 @@ import (
 	"gvisor.dev/gvisor/test/benchmarks/base"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkSizeEmpty creates N empty containers and reads memory usage from
@@ -58,6 +59,9 @@ func BenchmarkSizeEmpty(b *testing.B) {
 			base.CleanUpContainers(ctx, containers)
 			b.Fatalf("failed to run container: %v", err)
 		}
+		if i == 0 {
+			defer metricsviz.FromContainerLogs(ctx, b, container)
+		}
 		if _, err := container.WaitForOutputSubmatch(ctx, "Hello", 5*time.Second); err != nil {
 			base.CleanUpContainers(ctx, containers)
 			b.Fatalf("failed to read container output: %v", err)
@@ -110,6 +114,9 @@ func BenchmarkSizeNginx(b *testing.B) {
 			Cmd:     []string{"nginx", "-c", "/etc/nginx/nginx_gofer.conf"},
 		})
 	defer base.CleanUpContainers(ctx, servers)
+	if len(servers) > 0 {
+		defer metricsviz.FromContainerLogs(ctx, b, servers[0])
+	}
 
 	// DropCaches after servers are created.
 	harness.DropCaches(machine)
@@ -162,6 +169,9 @@ func BenchmarkSizeNode(b *testing.B) {
 			Cmd:     nodeCmd,
 		})
 	defer base.CleanUpContainers(ctx, servers)
+	if len(servers) > 0 {
+		defer metricsviz.FromContainerLogs(ctx, b, servers[0])
+	}
 
 	// DropCaches after servers are created.
 	harness.DropCaches(machine)

test/benchmarks/base/startup_test.go

@@ -23,6 +23,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/base"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkStartEmpty times startup time for an empty container.
@@ -46,6 +47,9 @@ func BenchmarkStartupEmpty(b *testing.B) {
 			b.Fatalf("failed to start container: %v", err)
 		}
 		b.StopTimer()
+		if i == 0 {
+			metricsviz.FromContainerLogs(ctx, b, container)
+		}
 		container.CleanUp(ctx)
 		harness.DebugLog(b, "Ran container: %d", i)
 	}
@@ -113,6 +117,7 @@ func runServerWorkload(ctx context.Context, b *testing.B, args base.ServerArgs)
 			server := args.Machine.GetContainer(ctx, b)
 			defer func() {
 				b.StopTimer()
+				metricsviz.FromContainerLogs(ctx, b, server)
 				// Cleanup servers as we run so that we can go indefinitely.
 				server.CleanUp(ctx)
 				b.StartTimer()

test/benchmarks/base/sysbench_test.go

@@ -22,6 +22,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 type testCase struct {
@@ -30,7 +31,7 @@ type testCase struct {
 	test    tools.Sysbench
 }
 
-// BenchmarSysbench runs sysbench on the runtime.
+// BenchmarkSysbench runs sysbench on the runtime.
 func BenchmarkSysbench(b *testing.B) {
 	testCases := []testCase{
 		{
@@ -108,6 +109,7 @@ func BenchmarkSysbench(b *testing.B) {
 			ctx := context.Background()
 			sysbench := machine.GetContainer(ctx, b)
 			defer sysbench.CleanUp(ctx)
+			defer metricsviz.FromContainerLogs(ctx, b, sysbench)
 
 			cmd := tc.test.MakeCmd(b)
 			b.ResetTimer()

test/benchmarks/base/syscallbench_test.go

@@ -22,6 +22,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarSyscallbench runs a syscall b.N times on the runtime.
@@ -67,6 +68,7 @@ func BenchmarkSyscallbench(b *testing.B) {
 			); err != nil {
 				b.Fatalf("run failed with: %v", err)
 			}
+			defer metricsviz.FromContainerLogs(ctx, b, container)
 			b.Run(name, func(b *testing.B) {
 				cmd := []string{"syscallbench", fmt.Sprintf("--loops=%d", b.N), fmt.Sprintf("--syscall=%d", tc.syscallArg)}
 				b.ResetTimer()
@@ -114,6 +116,7 @@ func BenchmarkSyscallUnderSeccomp(b *testing.B) {
 			); err != nil {
 				b.Fatalf("run failed with: %v", err)
 			}
+			defer metricsviz.FromContainerLogs(ctx, b, container)
 			b.Run(name, func(b *testing.B) {
 				cmd := []string{"syscallbench", "--syscall=1", fmt.Sprintf("--loops=%d", b.N)}
 				if tc.Value == "true" {

test/benchmarks/base/usage_test.go

@@ -26,6 +26,7 @@ import (
 	"gvisor.dev/gvisor/test/benchmarks/base"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkSizeEmpty creates N alpine containers and reads memory usage using `docker stats`.
@@ -48,6 +49,9 @@ func BenchmarkSizeEmpty(b *testing.B) {
 		}, "sh", "-c", "echo Hello && sleep 1000"); err != nil {
 			b.Fatalf("failed to run container: %v", err)
 		}
+		if i == 0 {
+			defer metricsviz.FromContainerLogs(ctx, b, container)
+		}
 		if _, err := container.WaitForOutputSubmatch(ctx, "Hello", 5*time.Second); err != nil {
 			b.Fatalf("failed to read container output: %v", err)
 		}
@@ -92,6 +96,9 @@ func BenchmarkSizeNginx(b *testing.B) {
 			b.Fatalf("failed to start server: %v", err)
 		}
 		defer server.CleanUp(ctx)
+		if i == 0 {
+			defer metricsviz.FromContainerLogs(ctx, b, server)
+		}
 		stats, err := server.Stats(ctx)
 		if err != nil {
 			b.Fatalf("failed to get container stats: %v", err)
@@ -139,6 +146,9 @@ func BenchmarkSizeNode(b *testing.B) {
 			b.Fatalf("failed to start server: %v", err)
 		}
 		defer server.CleanUp(ctx)
+		if i == 0 {
+			defer metricsviz.FromContainerLogs(ctx, b, server)
+		}
 		stats, err := server.Stats(ctx)
 		if err != nil {
 			b.Fatalf("failed to get container stats: %v", err)

test/benchmarks/fs/BUILD

@@ -29,6 +29,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
         "@com_github_docker_docker//api/types/mount:go_default_library",
     ],
 )

test/benchmarks/fs/fio_test.go

@@ -25,6 +25,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // Fio benchmarks run fio on the runtime under test. There are 4 basic test
@@ -192,6 +193,7 @@ func doFioBenchmark(b *testing.B, testCases []tools.Fio) {
 				ctx := context.Background()
 				container := machine.GetContainer(ctx, b)
 				cu := cleanup.Make(func() {
+					metricsviz.FromContainerLogs(ctx, b, container)
 					container.CleanUp(ctx)
 				})
 				defer cu.Clean()

test/benchmarks/media/BUILD

@@ -20,5 +20,6 @@ benchmark_test(
     deps = [
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
+        "//test/metricsviz",
     ],
 )

test/benchmarks/media/ffmpeg_test.go

@@ -21,6 +21,7 @@ import (
 
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkFfmpeg runs ffmpeg in a container and records runtime.
@@ -39,19 +40,24 @@ func BenchmarkFfmpeg(b *testing.B) {
 	b.StopTimer()
 
 	for i := 0; i < b.N; i++ {
-		container := machine.GetContainer(ctx, b)
-		defer container.CleanUp(ctx)
-		if err := harness.DropCaches(machine); err != nil {
-			b.Skipf("failed to drop caches: %v. You probably need root.", err)
-		}
+		func() {
+			container := machine.GetContainer(ctx, b)
+			defer container.CleanUp(ctx)
+			if i == 0 {
+				defer metricsviz.FromContainerLogs(ctx, b, container)
+			}
+			if err := harness.DropCaches(machine); err != nil {
+				b.Skipf("failed to drop caches: %v. You probably need root.", err)
+			}
 
-		b.StartTimer()
-		if _, err := container.Run(ctx, dockerutil.RunOpts{
-			Image: "benchmarks/ffmpeg",
-		}, cmd...); err != nil {
-			b.Fatalf("failed to run container: %v", err)
-		}
-		b.StopTimer()
+			b.StartTimer()
+			if _, err := container.Run(ctx, dockerutil.RunOpts{
+				Image: "benchmarks/ffmpeg",
+			}, cmd...); err != nil {
+				b.Fatalf("failed to run container: %v", err)
+			}
+			b.StopTimer()
+		}()
 	}
 }
 

test/benchmarks/ml/BUILD

@@ -21,5 +21,6 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )

test/benchmarks/ml/tensorflow_test.go

@@ -21,6 +21,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 func BenchmarkTensorflowDashboard(b *testing.B) {
@@ -70,22 +71,27 @@ func doTensorflowTest(b *testing.B, workloads map[string]string) {
 			b.StopTimer()
 
 			for i := 0; i < b.N; i++ {
-				container := machine.GetContainer(ctx, b)
-				defer container.CleanUp(ctx)
-				if err := harness.DropCaches(machine); err != nil {
-					b.Skipf("failed to drop caches: %v. You probably need root.", err)
-				}
+				func() {
+					container := machine.GetContainer(ctx, b)
+					defer container.CleanUp(ctx)
+					if i == 0 {
+						defer metricsviz.FromContainerLogs(ctx, b, container)
+					}
+					if err := harness.DropCaches(machine); err != nil {
+						b.Skipf("failed to drop caches: %v. You probably need root.", err)
+					}
 
-				// Run tensorflow.
-				b.StartTimer()
-				if out, err := container.Run(ctx, dockerutil.RunOpts{
-					Image:   "benchmarks/tensorflow",
-					Env:     []string{"PYTHONPATH=$PYTHONPATH:/TensorFlow-Examples/examples"},
-					WorkDir: "/TensorFlow-Examples/examples",
-				}, "python", file); err != nil {
-					b.Errorf("failed to run container: %v logs: %s", err, out)
-				}
-				b.StopTimer()
+					// Run tensorflow.
+					b.StartTimer()
+					if out, err := container.Run(ctx, dockerutil.RunOpts{
+						Image:   "benchmarks/tensorflow",
+						Env:     []string{"PYTHONPATH=$PYTHONPATH:/TensorFlow-Examples/examples"},
+						WorkDir: "/TensorFlow-Examples/examples",
+					}, "python", file); err != nil {
+						b.Errorf("failed to run container: %v logs: %s", err, out)
+					}
+					b.StopTimer()
+				}()
 			}
 		})
 	}

test/benchmarks/network/BUILD

@@ -16,6 +16,7 @@ go_library(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -31,6 +32,7 @@ benchmark_test(
         "//pkg/test/testutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -45,6 +47,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 
@@ -59,6 +62,7 @@ benchmark_test(
         "//pkg/test/dockerutil",
         "//test/benchmarks/harness",
         "//test/benchmarks/tools",
+        "//test/metricsviz",
     ],
 )
 

test/benchmarks/network/iperf_test.go

@@ -24,6 +24,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/testutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 func BenchmarkIperfOneConnection(b *testing.B) {
@@ -69,8 +70,10 @@ func BenchmarkIperfOneConnection(b *testing.B) {
 			// Set up the containers.
 			server := bm.serverFunc(ctx, b)
 			defer server.CleanUp(ctx)
+			defer metricsviz.FromNamedContainerLogs(ctx, b, server, "server")
 			client := bm.clientFunc(ctx, b)
 			defer client.CleanUp(ctx)
+			defer metricsviz.FromNamedContainerLogs(ctx, b, server, "client")
 
 			// iperf server listens on port 5001 by default.
 			port := 5001
@@ -180,8 +183,10 @@ func BenchmarkIperfManyConnections(b *testing.B) {
 			// Set up the containers.
 			server := bm.serverFunc(ctx, b)
 			defer server.CleanUp(ctx)
+			defer metricsviz.FromNamedContainerLogs(ctx, b, server, "server")
 			client := bm.clientFunc(ctx, b)
 			defer client.CleanUp(ctx)
+			defer metricsviz.FromNamedContainerLogs(ctx, b, client, "client")
 
 			// iperf server listens on port 5001 by default.
 			port := 5001

test/benchmarks/network/network.go

@@ -22,6 +22,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // runStaticServer runs static serving workloads (httpd, nginx).
@@ -46,6 +47,7 @@ func runStaticServer(b *testing.B, serverOpts dockerutil.RunOpts, serverCmd []st
 	defer client.CleanUp(ctx)
 	server := serverMachine.GetContainer(ctx, b)
 	defer server.CleanUp(ctx)
+	defer metricsviz.FromContainerLogs(ctx, b, server)
 
 	// Start the server.
 	if err := server.Spawn(ctx, serverOpts, serverCmd...); err != nil {

test/benchmarks/network/node_test.go

@@ -23,6 +23,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkNode runs requests using 'hey' against a Node server run on
@@ -78,6 +79,7 @@ func runNode(b *testing.B, hey *tools.Hey) {
 		b.Fatalf("failed to spawn redis instance: %v", err)
 	}
 	defer redis.CleanUp(ctx)
+	defer metricsviz.FromContainerLogs(ctx, b, redis)
 
 	if out, err := redis.WaitForOutput(ctx, "Ready to accept connections", 3*time.Second); err != nil {
 		b.Fatalf("failed to start redis server: %v %s", err, out)

test/benchmarks/network/ruby_test.go

@@ -24,6 +24,7 @@ import (
 	"gvisor.dev/gvisor/pkg/test/dockerutil"
 	"gvisor.dev/gvisor/test/benchmarks/harness"
 	"gvisor.dev/gvisor/test/benchmarks/tools"
+	"gvisor.dev/gvisor/test/metricsviz"
 )
 
 // BenchmarkRuby runs requests using 'hey' against a ruby application server.
@@ -76,6 +77,7 @@ func runRuby(b *testing.B, hey *tools.Hey) {
 		b.Fatalf("failed to spawn redis instance: %v", err)
 	}
 	defer redis.CleanUp(ctx)
+	defer metricsviz.FromContainerLogs(ctx, b, redis)
 
 	if out, err := redis.WaitForOutput(ctx, "Ready to accept connections", 3*time.Second); err != nil {
 		b.Fatalf("failed to start redis server: %v %s", err, out)