Reap toolsrv processes on shutdown

cmd/olliesrv/internal/toolclient/spawn.go

@@ -6,6 +6,7 @@ import (
 	"context"
 	"crypto/rand"
 	"encoding/hex"
+	"errors"
 	"fmt"
 	"io"
 	"net"
@@ -27,6 +28,8 @@ type Process struct {
 	Secret     string
 	Info       ProcessInfo
 	cancel     context.CancelFunc
+	waitOnce   sync.Once
+	waitErr    error
 }
 
 // ProcessInfo contains metadata about the toolsrv process.
@@ -94,7 +97,9 @@ func Spawn(ctx context.Context, cwd string, opts ...Option) (*Process, error) {
 	// Wait for socket to be ready
 	if err := waitForSocket(socketPath, 5*time.Second); err != nil {
 		cancel()
-		cmd.Process.Kill()
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
+		_ = os.Remove(socketPath)
 		return nil, fmt.Errorf("wait for socket: %w", err)
 	}
 
@@ -115,10 +120,15 @@ func (p *Process) Kill() error {
 	if p.cancel != nil {
 		p.cancel()
 	}
+	var err error
 	if p.Cmd != nil && p.Cmd.Process != nil {
-		return p.Cmd.Process.Kill()
+		err = p.Cmd.Process.Kill()
 	}
-	return nil
+	waitErr := p.Wait()
+	if err != nil && !errors.Is(err, os.ErrProcessDone) {
+		return err
+	}
+	return waitErr
 }
 
 // Close terminates the process (alias for Kill).
@@ -128,11 +138,17 @@ func (p *Process) Close() error {
 
 // Wait waits for the process to exit.
 func (p *Process) Wait() error {
-	if p.Cmd != nil {
-		return p.Cmd.Wait()
-	}
+	if p.Cmd == nil {
 		return nil
 	}
+	p.waitOnce.Do(func() {
+		p.waitErr = p.Cmd.Wait()
+		if p.SocketPath != "" {
+			_ = os.Remove(p.SocketPath)
+		}
+	})
+	return p.waitErr
+}
 
 // RemoteConfig configures remote toolsrv spawning.
 type RemoteConfig struct {
@@ -219,13 +235,15 @@ exec "$CACHE_DIR/bin/toolsrv" %s
 	// Send tarball of config dir over stdin
 	if err := writeTarball(stdin, cfgDir); err != nil {
 		cancel()
-		cmd.Process.Kill()
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
 		return nil, fmt.Errorf("send tarball: %w", err)
 	}
 	// Send bootstrap script
 	if _, err := stdin.Write([]byte(bootstrap)); err != nil {
 		cancel()
-		cmd.Process.Kill()
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
 		return nil, fmt.Errorf("send bootstrap: %w", err)
 	}
 	stdin.Close()
@@ -248,14 +266,16 @@ exec "$CACHE_DIR/bin/toolsrv" %s
 	case <-readyCh:
 	case <-time.After(30 * time.Second):
 		cancel()
-		cmd.Process.Kill()
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
 		return nil, fmt.Errorf("remote toolsrv startup timeout")
 	}
 
 	// Wait for local socket forwarding to be ready
 	if err := waitForSocket(localSock, 5*time.Second); err != nil {
 		cancel()
-		cmd.Process.Kill()
+		_ = cmd.Process.Kill()
+		_ = cmd.Wait()
 		return nil, fmt.Errorf("wait for forwarded socket: %w", err)
 	}