fix(agent): give the override file the owner of the stack (#24)

CLAUDE.md

@@ -98,6 +98,9 @@ DATABASE_URL=postgres://... go test -tags integration ./internal/master/store/
   without the other has caused a fix commit already.
 - **`<summary>` C# convention from global CLAUDE.md does not apply here** — this
   is Go. Use idiomatic GoDoc (`// FuncName does X.`).
+- **The override file gets the owner of the existing override, else of the stack's first
+  compose file.** The agent runs as root and renames a new file into place; it chowns it
+  afterwards. A failure is a `warning:` line in the job output, never a failed job.
 - **Compose loads only one override file.** Of `compose.override.yml`,
   `compose.override.yaml`, `docker-compose.override.yml`, `docker-compose.override.yaml`
   only the first that exists is read. Hence the refusal described above.

internal/agent/updater/compose.go

@@ -2,11 +2,15 @@ package updater
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
+	"io/fs"
 	"os"
 	"os/exec"
+	"path/filepath"
 	"strings"
+	"syscall"
 	"time"
 
 	"github.com/docker/docker/api/types/container"
@@ -27,6 +31,26 @@ const (
 // job's error message; the full output is in the job log.
 const errorOutputTail = 2 << 10
 
+// chown is os.Chown; tests replace it.
+var chown = os.Chown
+
+// stackOwner returns the uid and gid the override file should have: those of
+// the existing override if there is one, else those of the stack's compose file.
+func stackOwner(overridePath, composeFile string) (uid, gid int, err error) {
+	fi, err := os.Stat(overridePath)
+	if errors.Is(err, fs.ErrNotExist) {
+		fi, err = os.Stat(composeFile)
+	}
+	if err != nil {
+		return 0, 0, err
+	}
+	st, ok := fi.Sys().(*syscall.Stat_t)
+	if !ok {
+		return 0, 0, errors.New("file owner is not available on this platform")
+	}
+	return int(st.Uid), int(st.Gid), nil
+}
+
 // tailWriter adapts a tailBuffer to io.Writer.
 type tailWriter struct{ b *tailBuffer }
 
@@ -152,9 +176,19 @@ func (e *ComposeExecutor) Execute(ctx context.Context, job protocol.JobStart, ou
 	}
 	fmt.Fprintln(out, resolveNote(job.Ref, pinned))
 
+	// The agent runs as root and WriteOverride renames a new file into place,
+	// so the override would end up root-owned. Hand it back to the stack's
+	// owner. A failure here is only a warning: the update itself is fine.
+	overridePath := filepath.Join(dir, OverrideName)
+	uid, gid, ownerErr := stackOwner(overridePath, files[0])
 	if _, err := WriteOverride(dir, service, pinned); err != nil {
 		return "", err
 	}
+	if ownerErr != nil {
+		fmt.Fprintf(out, "warning: could not read the owner of %s or %s to give it to %s: %v\n", overridePath, files[0], overridePath, ownerErr)
+	} else if err := chown(overridePath, uid, gid); err != nil {
+		fmt.Fprintf(out, "warning: could not give %s the owner %d:%d: %v\n", overridePath, uid, gid, err)
+	}
 
 	base := composeBase(project, workingDir, files, envs)
 	for _, step := range [][]string{{"pull", service}, {"up", "-d", service}} {

internal/agent/updater/compose_test.go

@@ -7,6 +7,7 @@ import (
 	"os"
 	"path/filepath"
 	"strings"
+	"syscall"
 	"testing"
 
 	"github.com/docker/docker/api/types/container"
@@ -64,6 +65,48 @@ func (r *recorder) run(_ context.Context, dir string, out io.Writer, args ...str
 	return nil
 }
 
+// stubChown replaces chown for the test. Tests that use it must not run in parallel.
+func stubChown(t *testing.T, f func(path string, uid, gid int) error) {
+	t.Helper()
+	orig := chown
+	chown = f
+	t.Cleanup(func() { chown = orig })
+}
+
+func TestStackOwner(t *testing.T) {
+	dir := t.TempDir()
+	override := filepath.Join(dir, OverrideName)
+	compose := filepath.Join(dir, "compose.yaml")
+
+	t.Run("falls back to the compose file when there is no override", func(t *testing.T) {
+		require.NoError(t, os.WriteFile(compose, nil, 0o644))
+		uid, gid, err := stackOwner(override, compose)
+		require.NoError(t, err)
+		assert.Equal(t, os.Getuid(), uid)
+		assert.Equal(t, os.Getgid(), gid)
+	})
+	t.Run("prefers the existing override", func(t *testing.T) {
+		require.NoError(t, os.WriteFile(override, nil, 0o644))
+		require.NoError(t, os.Remove(compose))
+		uid, _, err := stackOwner(override, compose)
+		require.NoError(t, err, "the compose file is not consulted")
+		assert.Equal(t, os.Getuid(), uid)
+	})
+	t.Run("fails when neither file exists", func(t *testing.T) {
+		require.NoError(t, os.Remove(override))
+		_, _, err := stackOwner(override, compose)
+		assert.Error(t, err)
+	})
+}
+
+func TestChown_RealToOwnIDs(t *testing.T) {
+	f := filepath.Join(t.TempDir(), "f")
+	require.NoError(t, os.WriteFile(f, nil, 0o644))
+	uid, gid, err := stackOwner(f, f)
+	require.NoError(t, err)
+	assert.NoError(t, chown(f, uid, gid))
+}
+
 func TestComposeExecutor(t *testing.T) {
 	const d = "sha256:d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1d1"
 	root, _ := filepath.EvalSymlinks(t.TempDir())
@@ -108,6 +151,43 @@ func TestComposeExecutor(t *testing.T) {
 		assert.Equal(t, pinned, serviceImage(doc, "web"))
 		assert.Contains(t, out.String(), "resolved "+pinned)
 	})
+	t.Run("the override is chowned to the owner of the compose file", func(t *testing.T) {
+		t.Cleanup(cleanOverride)
+		composeFile := filepath.Join(stack, "compose.yaml")
+		require.NoError(t, os.WriteFile(composeFile, []byte("services: {}\n"), 0o644))
+		t.Cleanup(func() { os.Remove(composeFile) })
+		fi, err := os.Stat(composeFile)
+		require.NoError(t, err)
+		st := fi.Sys().(*syscall.Stat_t)
+		var gotPath string
+		var gotUID, gotGID int
+		stubChown(t, func(path string, uid, gid int) error {
+			gotPath, gotUID, gotGID = path, uid, gid
+			return nil
+		})
+		var out strings.Builder
+		_, err = setup(fakeCompose{rw, []container.Summary{ct}}, &fakeDist{digest: d}, &recorder{}).Execute(context.Background(), job, &out)
+		require.NoError(t, err)
+		assert.Equal(t, overridePath, gotPath)
+		assert.Equal(t, int(st.Uid), gotUID)
+		assert.Equal(t, int(st.Gid), gotGID)
+		assert.NotContains(t, out.String(), "warning")
+	})
+	t.Run("a chown failure is a warning and the job still succeeds", func(t *testing.T) {
+		t.Cleanup(cleanOverride)
+		composeFile := filepath.Join(stack, "compose.yaml")
+		require.NoError(t, os.WriteFile(composeFile, []byte("services: {}\n"), 0o644))
+		t.Cleanup(func() { os.Remove(composeFile) })
+		stubChown(t, func(string, int, int) error { return errors.New("operation not permitted") })
+		var out strings.Builder
+		got, err := setup(fakeCompose{rw, []container.Summary{ct}}, &fakeDist{digest: d}, &recorder{}).Execute(context.Background(), job, &out)
+		require.NoError(t, err)
+		assert.Equal(t, pinned, got)
+		assert.Contains(t, out.String(), "warning: could not give "+overridePath+" the owner ")
+		assert.Contains(t, out.String(), "operation not permitted")
+		_, doc := readOverride(t, stack)
+		assert.Equal(t, pinned, serviceImage(doc, "web"))
+	})
 	t.Run("a pinned ref is written verbatim and not resolved", func(t *testing.T) {
 		t.Cleanup(cleanOverride)
 		for _, ref := range []string{"gitea.shcizo.se/shcizo/slusk:1.3@" + d, "gitea.shcizo.se/shcizo/slusk@" + d} {