rm: clean up all nodes before returning errors

Signed-off-by: CrazyMax <1951866+crazy-max@users.noreply.github.com>

Author: crazy-max

commands/rm.go

@@ -2,6 +2,7 @@ package commands
 
 import (
 	"context"
+	stderrors "errors"
 	"fmt"
 	"time"
 
@@ -90,7 +91,7 @@ func runRm(ctx context.Context, dockerCli command.Cli, in rmOptions) error {
 					return err
 				}
 
-				err1 := rm(ctx, nodes, in)
+				err1 := rm(timeoutCtx, nodes, in)
 				if err := txn.Remove(b.Name); err != nil {
 					return err
 				}
@@ -140,24 +141,36 @@ func rmCmd(dockerCli command.Cli, rootOpts *rootOptions) *cobra.Command {
 }
 
 func rm(ctx context.Context, nodes []builder.Node, in rmOptions) (err error) {
+	errCh := make(chan error, len(nodes)*3)
+	var eg errgroup.Group
 	for _, node := range nodes {
-		if node.Driver == nil {
-			continue
-		}
-		// Do not stop the buildkitd daemon when --keep-daemon is provided
-		if !in.keepDaemon {
-			if err := node.Driver.Stop(ctx, true); err != nil {
-				return err
+		eg.Go(func() error {
+			if node.Err != nil {
+				errCh <- errors.Wrapf(node.Err, "failed to load node %s", node.Name)
 			}
-		}
-		if err := node.Driver.Rm(ctx, true, !in.keepState, !in.keepDaemon); err != nil {
-			return err
-		}
-		if node.Err != nil {
-			err = node.Err
-		}
+			if node.Driver == nil {
+				return nil
+			}
+			// Do not stop the buildkitd daemon when --keep-daemon is provided
+			if !in.keepDaemon {
+				if err := node.Driver.Stop(ctx, true); err != nil {
+					errCh <- errors.Wrapf(err, "failed to stop node %s", node.Name)
+				}
+			}
+			if err := node.Driver.Rm(ctx, true, !in.keepState, !in.keepDaemon); err != nil {
+				errCh <- errors.Wrapf(err, "failed to remove node %s", node.Name)
+			}
+			return nil
+		})
+	}
+	_ = eg.Wait()
+	close(errCh)
+
+	var errs []error
+	for err := range errCh {
+		errs = append(errs, err)
 	}
-	return err
+	return stderrors.Join(errs...)
 }
 
 func rmAllInactive(ctx context.Context, txn *store.Txn, dockerCli command.Cli, in rmOptions) error {
@@ -184,7 +197,7 @@ func rmAllInactive(ctx context.Context, txn *store.Txn, dockerCli command.Cli, i
 					return nil
 				}
 				if b.Inactive() {
-					rmerr := rm(ctx, nodes, in)
+					rmerr := rm(timeoutCtx, nodes, in)
 					if err := txn.Remove(b.Name); err != nil {
 						return err
 					}

tests/rm.go

@@ -25,6 +25,7 @@ var rmTests = []func(t *testing.T, sb integration.Sandbox){
 	testRmMulti,
 	testRmInvalidBuildkitdConfig,
 	testRmAllInactiveInvalidBuildkitdConfig,
+	testRmUnreachableEndpoint,
 }
 
 func testRm(t *testing.T, sb integration.Sandbox) {
@@ -34,6 +35,7 @@ func testRm(t *testing.T, sb integration.Sandbox) {
 
 	out, err := rmCmd(sb, withArgs("default"))
 	require.Error(t, err, out) // can't remove a docker builder
+	require.Contains(t, out, "context builder cannot be removed")
 
 	out, err = createCmd(sb, withArgs("--driver", "docker-container"))
 	require.NoError(t, err, out)
@@ -155,6 +157,43 @@ func testRmAllInactiveInvalidBuildkitdConfig(t *testing.T, sb integration.Sandbo
 	builderName = ""
 }
 
+func testRmUnreachableEndpoint(t *testing.T, sb integration.Sandbox) {
+	if !isDockerContainerWorker(sb) {
+		t.Skip("only testing with docker-container worker")
+	}
+
+	out, err := createCmd(sb, withArgs("--driver", "docker-container"))
+	require.NoError(t, err, out)
+	builderName := strings.TrimSpace(out)
+
+	out, err = inspectCmd(sb, withArgs(builderName, "--bootstrap"))
+	require.NoError(t, err, out)
+
+	t.Cleanup(func() {
+		if builderName == "" {
+			return
+		}
+		_, _ = rmCmd(sb, withArgs("--keep-daemon", builderName))
+	})
+
+	var goodContainer string
+	updateStoredBuilder(t, sb, builderName, func(ng *store.NodeGroup) {
+		require.NotEmpty(t, ng.Nodes)
+		goodContainer = driver.BuilderName(ng.Nodes[0].Name)
+		badNode := ng.Nodes[0]
+		badNode.Name += "-unreachable"
+		badNode.Endpoint = "tcp://127.0.0.1:1"
+		ng.Nodes = append([]store.Node{badNode}, ng.Nodes...)
+	})
+
+	out, err = rmCmd(sb, withArgs("--timeout=2s", builderName))
+	require.Error(t, err, out)
+	require.Contains(t, out, "failed to remove "+builderName)
+	requireNoStoredBuilder(t, sb, builderName)
+	requireNoContainer(t, sb, goodContainer)
+	builderName = ""
+}
+
 func updateStoredBuilder(t *testing.T, sb integration.Sandbox, name string, fn func(*store.NodeGroup)) {
 	t.Helper()