RSDK-9706: Fix FTDC tracking of module restarts/crashes.

ftdc/cmd/parser.go

@@ -441,6 +441,9 @@ func main() {
 			nolintPrintln("reset range")
 			nolintPrintln("-  Unset any prior range. \"zoom out to full\"")
 			nolintPrintln()
+			nolintPrintln("r, refresh")
+			nolintPrintln("-  Regenerate the plot.png image. Useful when a current viam-server is running.")
+			nolintPrintln()
 			nolintPrintln("`quit` or Ctrl-d to exit")
 		case strings.HasPrefix(cmd, "range "):
 			pieces := strings.SplitN(cmd, " ", 3)

module/modmanager/manager.go

@@ -867,6 +867,10 @@ func (mgr *Manager) newOnUnexpectedExitHandler(mod *module) func(exitCode int) b
 				"error", err)
 		}
 
+		if mgr.ftdc != nil {
+			mgr.ftdc.Remove(mod.getFTDCName())
+		}
+
 		// If attemptRestart returns any orphaned resource names, restart failed,
 		// and we should remove orphaned resources. Since we handle process
 		// restarting ourselves, return false here so goutils knows not to attempt
@@ -1225,7 +1229,7 @@ func (m *module) stopProcess() error {
 		// of metrics will be reported. Therefore it is safe to continue monitoring the module process
 		// while it's in shutdown.
 		if m.ftdc != nil {
-			m.ftdc.Remove(m.process.ID())
+			m.ftdc.Remove(m.getFTDCName())
 		}
 	}()
 
@@ -1347,6 +1351,10 @@ func (m *module) getFullEnvironment(viamHomeDir string) map[string]string {
 	return getFullEnvironment(m.cfg, m.dataDir, viamHomeDir)
 }
 
+func (m *module) getFTDCName() string {
+	return fmt.Sprintf("proc.modules.%s", m.process.ID())
+}
+
 func (m *module) registerProcessWithFTDC() {
 	if m.ftdc == nil {
 		return
@@ -1364,7 +1372,7 @@ func (m *module) registerProcessWithFTDC() {
 		return
 	}
 
-	m.ftdc.Add(fmt.Sprintf("proc.modules.%s", m.process.ID()), statser)
+	m.ftdc.Add(m.getFTDCName(), statser)
 }
 
 func getFullEnvironment(

module/modmanager/manager_test.go

@@ -1,6 +1,7 @@
 package modmanager
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
@@ -28,6 +29,7 @@ import (
 	"go.viam.com/rdk/components/generic"
 	"go.viam.com/rdk/components/motor"
 	"go.viam.com/rdk/config"
+	"go.viam.com/rdk/ftdc"
 	"go.viam.com/rdk/logging"
 	modlib "go.viam.com/rdk/module"
 	modmanageroptions "go.viam.com/rdk/module/modmanager/options"
@@ -1379,6 +1381,90 @@ func TestBadModuleFailsFast(t *testing.T) {
 	test.That(t, err.Error(), test.ShouldContainSubstring, "module test-module exited too quickly after attempted startup")
 }
 
+// TestFTDCAfterModuleCrash is to give confidence that the FTDC sections devoted to tracking module
+// process information (e.g: CPU usage) is in sync with the Process IDs (PIDs) that are actually
+// running.
+func TestFTDCAfterModuleCrash(t *testing.T) {
+	logger := logging.NewTestLogger(t)
+	modCfgs := []config.Module{
+		{
+			Name: "test-module",
+			// testmodule2 has a `kill_module` DoCommand to force a module/process crash.
+			ExePath: rtestutils.BuildTempModule(t, "module/testmodule2"),
+			Type:    config.ModuleTypeLocal,
+		},
+	}
+
+	ctx := context.Background()
+	parentAddr := setupSocketWithRobot(t)
+	opts := modmanageroptions.Options{UntrustedEnv: false}
+
+	// Start up a mod manager with FTDC enabled. We will inspect the FTDC output for the
+	// `ElapsedTimeSecs` to assert the "pid tracking code" is working correctly.
+	ftdcData := bytes.NewBuffer(nil)
+	opts.FTDC = ftdc.NewWithWriter(ftdcData, logger)
+	// Normally a test would explicitly call `constructDatum` to control/guarantee FTDC gets
+	// data. But as a short-cut to avoid exposing methods that are currently private, we just run
+	// FTDC in the background. And sleep long enough between testing events (killing the module) to
+	// assert the right behavior.
+	opts.FTDC.Start()
+
+	// Set up a mod manager. Currently there are zero modules running.
+	mgr := setupModManager(t, ctx, parentAddr, logger, opts)
+
+	// Add a module, this will register an FTDC "section" for that module process.
+	err := mgr.Add(ctx, modCfgs...)
+	test.That(t, err, test.ShouldBeNil)
+
+	// Add a resource -- this is simply to invoke the `kill_module` command.
+	res, err := mgr.AddResource(ctx, resource.Config{
+		Name:  "foo",
+		API:   generic.API,
+		Model: resource.NewModel("rdk", "test", "helper2"),
+	}, nil)
+	test.That(t, err, test.ShouldBeNil)
+	test.That(t, mgr.IsModularResource(generic.Named("foo")), test.ShouldBeTrue)
+
+	// Kill the module a few times for good measure.
+	for idx := 0; idx < 3; idx++ {
+		_, err = res.DoCommand(ctx, map[string]interface{}{"command": "kill_module"})
+
+		// FTDC is running in the background with a one second interval. So we sleep for two seconds
+		// and cross our fingers we don't get a poor scheduler execution. The assertions are
+		// intentionally weak to minimize the risk of false positives (a test failure with correct
+		// production code).
+		time.Sleep(2 * time.Second)
+	}
+
+	mgr.Close(ctx)
+	opts.FTDC.StopAndJoin(ctx)
+
+	datums, err := ftdc.Parse(ftdcData)
+	logger.Info("Num ftdc datums: ", len(datums))
+
+	// Keep count of the number of `ElapsedTimeSecs` readings we encounter. It is a testing bug if
+	// we don't see any process FTDC metrics for the module.
+	numModuleElapsedTimeMetricsSeen := 0
+	for _, datum := range datums {
+		for _, reading := range datum.Readings {
+			if reading.MetricName == "proc.modules.test-module.ElapsedTimeSecs" {
+				logger.Infow("Reading", "timestamp", datum.Time, "elapsedTimeSecs", reading.Value)
+				numModuleElapsedTimeMetricsSeen++
+				// Dan: I don't have a good reason to believe that we can't (legitimately) observe
+				// an `ElapsedTimeSecs` of 0 here. It's more likely we'd see a 0 because we queried
+				// a bad PID.
+				//
+				// If my assumption is wrong and we get a false positive here, we can reevaluate the
+				// options for making a more robust test.
+				test.That(t, reading.Value, test.ShouldBeGreaterThan, 0)
+			}
+		}
+	}
+
+	// Assert that we saw at least one datapoint before considering the test a success.
+	test.That(t, numModuleElapsedTimeMetricsSeen, test.ShouldBeGreaterThan, 0)
+}
+
 func TestModularDiscoverFunc(t *testing.T) {
 	ctx := context.Background()
 	logger := logging.NewTestLogger(t)