Improvements for osqueryinstance's errgroup

ee/errgroup/errgroup.go

@@ -0,0 +1,163 @@
+package errgroup
+
+import (
+	"context"
+	"fmt"
+	"log/slog"
+	"time"
+
+	"github.com/pkg/errors"
+	"golang.org/x/sync/errgroup"
+)
+
+type LoggedErrgroup struct {
+	errgroup *errgroup.Group
+	cancel   context.CancelFunc
+	doneCtx  context.Context // nolint:containedctx
+	slogger  *slog.Logger
+}
+
+const (
+	maxShutdownGoroutineDuration = 3 * time.Second
+)
+
+func NewLoggedErrgroup(ctx context.Context, slogger *slog.Logger) *LoggedErrgroup {
+	ctx, cancel := context.WithCancel(ctx)
+	e, doneCtx := errgroup.WithContext(ctx)
+
+	return &LoggedErrgroup{
+		errgroup: e,
+		cancel:   cancel,
+		doneCtx:  doneCtx,
+		slogger:  slogger,
+	}
+}
+
+// StartGoroutine starts the given goroutine in the errgroup, ensuring that we log its start and exit.
+func (l *LoggedErrgroup) StartGoroutine(ctx context.Context, goroutineName string, goroutine func() error) {
+	l.errgroup.Go(func() (err error) {
+		slogger := l.slogger.With("goroutine_name", goroutineName)
+
+		// Catch any panicking goroutines and log them. We also want to make sure
+		// we return an error from this goroutine overall if it panics.
+		defer func() {
+			if r := recover(); r != nil {
+				slogger.Log(ctx, slog.LevelError,
+					"panic occurred in goroutine",
+					"err", r,
+				)
+				if recoveredErr, ok := r.(error); ok {
+					slogger.Log(ctx, slog.LevelError,
+						"panic stack trace",
+						"stack_trace", fmt.Sprintf("%+v", errors.WithStack(recoveredErr)),
+					)
+					err = recoveredErr
+				}
+			}
+		}()
+
+		slogger.Log(ctx, slog.LevelInfo,
+			"starting goroutine in errgroup",
+		)
+
+		err = goroutine()
+
+		slogger.Log(ctx, slog.LevelInfo,
+			"exiting goroutine in errgroup",
+			"goroutine_err", err,
+		)
+
+		return err
+	})
+}
+
+// StartRepeatedGoroutine starts the given goroutine in the errgroup, ensuring that we log its start and exit.
+// If the delay is non-zero, the goroutine will not start until after the delay interval has elapsed. The goroutine
+// will run on the given interval, and will continue to run until it returns an error or the errgroup shuts down.
+func (l *LoggedErrgroup) StartRepeatedGoroutine(ctx context.Context, goroutineName string, interval time.Duration, delay time.Duration, goroutine func() error) {
+	l.StartGoroutine(ctx, goroutineName, func() error {
+		slogger := l.slogger.With("goroutine_name", goroutineName)
+
+		if delay != 0*time.Second {
+			select {
+			case <-time.After(delay):
+				slogger.Log(ctx, slog.LevelDebug,
+					"exiting delay before starting repeated goroutine",
+				)
+			case <-l.doneCtx.Done():
+				return nil
+			}
+		}
+
+		ticker := time.NewTicker(interval)
+		defer ticker.Stop()
+
+		for {
+			// Run goroutine immediately
+			if err := goroutine(); err != nil {
+				slogger.Log(ctx, slog.LevelInfo,
+					"exiting repeated goroutine in errgroup due to error",
+					"goroutine_err", err,
+				)
+				return err
+			}
+
+			// Wait for next interval or for errgroup shutdown
+			select {
+			case <-l.doneCtx.Done():
+				slogger.Log(ctx, slog.LevelInfo,
+					"exiting repeated goroutine in errgroup due to shutdown",
+				)
+				return nil
+			case <-ticker.C:
+				continue
+			}
+		}
+	})
+}
+
+// AddShutdownGoroutine adds the given goroutine to the errgroup, ensuring that we log its start and exit.
+// The goroutine will not execute until the errgroup has received a signal to exit.
+func (l *LoggedErrgroup) AddShutdownGoroutine(ctx context.Context, goroutineName string, goroutine func() error) {
+	l.StartGoroutine(ctx, goroutineName, func() error {
+		slogger := l.slogger.With("goroutine_name", goroutineName)
+
+		// Wait for errgroup to exit
+		<-l.doneCtx.Done()
+
+		goroutineStart := time.Now()
+		err := goroutine()
+		elapsedTime := time.Since(goroutineStart)
+
+		// Log anything amiss about the shutdown goroutine -- did it return an error? Did it take too long?
+		if err != nil {
+			slogger.Log(ctx, slog.LevelWarn,
+				"shutdown routine returned err",
+				"goroutine_run_time", elapsedTime.String(),
+				"goroutine_err", err,
+			)
+		} else if elapsedTime > maxShutdownGoroutineDuration {
+			slogger.Log(ctx, slog.LevelWarn,
+				"noticed slow shutdown routine",
+				"goroutine_run_time", elapsedTime.String(),
+			)
+		}
+
+		// We don't want to actually return the error here, to avoid causing an otherwise successful call
+		// to `Shutdown` => `Wait` to return an error. Shutdown routine errors don't matter for the success
+		// of the errgroup overall.
+		return l.doneCtx.Err()
+	})
+}
+
+func (l *LoggedErrgroup) Shutdown() {
+	l.cancel()
+}
+
+func (l *LoggedErrgroup) Wait() error {
+	return l.errgroup.Wait()
+}
+
+func (l *LoggedErrgroup) Exited() <-chan struct{} {
+	return l.doneCtx.Done()
+}

ee/errgroup/errgroup_test.go

@@ -0,0 +1,130 @@
+package errgroup
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/kolide/launcher/pkg/log/multislogger"
+	"github.com/stretchr/testify/require"
+)
+
+func TestWait(t *testing.T) {
+	t.Parallel()
+
+	err1 := errors.New("errgroup_test: 1")
+	err2 := errors.New("errgroup_test: 2")
+
+	for _, tt := range []struct {
+		testCaseName string
+		errs         []error
+		expectedErr  error
+	}{
+		{
+			testCaseName: "no error on exit",
+			errs:         []error{nil},
+			expectedErr:  nil,
+		},
+		{
+			testCaseName: "only first routine has error on exit",
+			errs:         []error{err1, nil},
+			expectedErr:  err1,
+		},
+		{
+			testCaseName: "only second routine has error on exit",
+			errs:         []error{nil, err2},
+			expectedErr:  err2,
+		},
+		{
+			testCaseName: "multiple routines have error on exit",
+			errs:         []error{err1, nil, err2},
+			expectedErr:  err1,
+		},
+	} {
+		tt := tt
+		t.Run(tt.testCaseName, func(t *testing.T) {
+			t.Parallel()
+
+			ctx, cancel := context.WithCancel(context.TODO())
+			defer cancel()
+
+			eg := NewLoggedErrgroup(ctx, multislogger.NewNopLogger())
+
+			for i, err := range tt.errs {
+				err := err
+				eg.StartGoroutine(ctx, strconv.Itoa(i), func() error { return err })
+				time.Sleep(500 * time.Millisecond) // try to enforce ordering of goroutines
+			}
+
+			// We should get the expected error when we wait for the routines to exit
+			require.Equal(t, tt.expectedErr, eg.Wait(), "incorrect error returned by errgroup")
+
+			// We expect that the errgroup shuts down
+			canceled := false
+			select {
+			case <-eg.Exited():
+				canceled = true
+			default:
+			}
+
+			require.True(t, canceled, "errgroup did not exit")
+		})
+	}
+}
+
+func TestShutdown(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(context.TODO())
+	defer cancel()
+
+	eg := NewLoggedErrgroup(ctx, multislogger.NewNopLogger())
+
+	eg.StartGoroutine(ctx, "test_goroutine", func() error {
+		return nil
+	})
+
+	// We should get the expected error when we wait for the routines to exit
+	eg.Shutdown()
+	require.Nil(t, eg.Wait(), "should not have returned error on shutdown")
+
+	// We expect that the errgroup shuts down
+	canceled := false
+	select {
+	case <-eg.Exited():
+		canceled = true
+	default:
+	}
+
+	require.True(t, canceled, "errgroup did not exit")
+}
+
+func Test_HandlesPanic(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(context.TODO())
+	defer cancel()
+
+	eg := NewLoggedErrgroup(ctx, multislogger.NewNopLogger())
+
+	eg.StartGoroutine(ctx, "test_goroutine", func() error {
+		testArr := make([]int, 0)
+		fmt.Println(testArr[2]) // cause out-of-bounds panic
+		return nil
+	})
+
+	// We expect that the errgroup shuts down -- the test should not panic
+	eg.Shutdown()
+	require.Error(t, eg.Wait(), "should have returned error from panicking goroutine")
+	canceled := false
+	select {
+	case <-eg.Exited():
+		canceled = true
+	default:
+	}
+
+	require.True(t, canceled, "errgroup did not exit")
+}