getallpids: Ignore removed cgroups

getallpids.go

@@ -1,8 +1,12 @@
 package cgroups
 
 import (
+	"errors"
 	"io/fs"
+	"os"
 	"path/filepath"
+
+	"golang.org/x/sys/unix"
 )
 
 // GetAllPids returns all pids from the cgroup identified by path, and all its
@@ -11,17 +15,30 @@ func GetAllPids(path string) ([]int, error) {
 	var pids []int
 	err := filepath.WalkDir(path, func(p string, d fs.DirEntry, iErr error) error {
 		if iErr != nil {
+			// A descendant cgroup can be removed while we walk, ignore
+			// any such error unless it's on the root (path) cgroup here
+			if p != path && ignoreCgroupRemoved(iErr) {
+				return nil
+			}
 			return iErr
 		}
 		if !d.IsDir() {
 			return nil
 		}
 		cPids, err := readProcsFile(p)
 		if err != nil {
+			if p != path && ignoreCgroupRemoved(err) {
+				return nil
+			}
 			return err
 		}
 		pids = append(pids, cPids...)
 		return nil
 	})
 	return pids, err
 }
+
+// ignoreCgroupRemoved reports whether err indicates the cgroup was removed.
+func ignoreCgroupRemoved(err error) bool {
+	return os.IsNotExist(err) || errors.Is(err, unix.ENODEV)
+}

getallpids_test.go

@@ -1,6 +1,9 @@
 package cgroups
 
 import (
+	"os"
+	"path/filepath"
+	"slices"
 	"testing"
 )
 
@@ -15,3 +18,34 @@ func BenchmarkGetAllPids(b *testing.B) {
 	}
 	b.Logf("iter: %d, total: %d", b.N, total)
 }
+
+func TestGetAllPidsIgnoresVanishedSubcgroup(t *testing.T) {
+	TestMode = true
+	defer func() { TestMode = false }()
+
+	root := t.TempDir()
+	if err := WriteFile(root, "cgroup.procs", "1\n"); err != nil {
+		t.Fatal(err)
+	}
+	child := filepath.Join(root, "child")
+	if err := os.Mkdir(child, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := WriteFile(child, "cgroup.procs", "2\n3\n"); err != nil {
+		t.Fatal(err)
+	}
+
+	// Pretend "vanishing" cgroup, directory exists but not the file to read
+	if err := os.Mkdir(filepath.Join(root, "gone"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+
+	pids, err := GetAllPids(root)
+	if err != nil {
+		t.Fatalf("GetAllPids errored on a vanished child cgroup: %v", err)
+	}
+	slices.Sort(pids)
+	if want := []int{1, 2, 3}; !slices.Equal(pids, want) {
+		t.Errorf("GetAllPids() = %v, want %v", pids, want)
+	}
+}