make cgroup errors skippable

metric/system/cgroup/util.go

@@ -168,7 +168,7 @@ func SupportedSubsystems(rootfs resolve.Resolver) (map[string]struct{}, error) {
 // The returned map contains the subsystem name as a key and the value is the
 // mountpoint.
 func SubsystemMountpoints(rootfs resolve.Resolver, subsystems map[string]struct{}) (Mountpoints, error) {
-
+	// TODO: will we run into mount namespace issues if we use /proc/self/mountinfo?
 	mountinfo, err := os.Open(rootfs.ResolveHostFS("/proc/self/mountinfo"))
 	if err != nil {
 		return Mountpoints{}, err

metric/system/process/process.go

@@ -257,12 +257,14 @@ func (procStats *Stats) pidFill(pid int, filter bool) (ProcState, bool, error) {
 	if procStats.EnableCgroups {
 		cgStats, err := procStats.cgroups.GetStatsForPid(status.Pid.ValueOr(0))
 		if err != nil {
-			return status, true, fmt.Errorf("cgroups.GetStatsForPid: %w", err)
-		}
-		status.Cgroup = cgStats
-		if ok {
-			status.Cgroup.FillPercentages(last.Cgroup, status.SampleTime, last.SampleTime)
+			procStats.logger.Debugf("Non-fatal error fetching cgroups metrics for pid %d, metrics are valid but partial: %s", pid, err)
+		} else {
+			status.Cgroup = cgStats
+			if ok {
+				status.Cgroup.FillPercentages(last.Cgroup, status.SampleTime, last.SampleTime)
+			}
 		}
+
 	} // end cgroups processor
 
 	status, err = FillMetricsRequiringMoreAccess(pid, status)

metric/system/process/process_linux_test.go

@@ -234,6 +234,25 @@ func TestParseProcStat(t *testing.T) {
 	assert.Equal(t, want, got, "")
 }
 
+func TestCgroupsBadCgroupsConfig(t *testing.T) {
+	_ = logp.DevelopmentSetup()
+	testStats := Stats{CPUTicks: true,
+		EnableCgroups: true,
+		EnableNetwork: true,
+		Hostfs:        resolve.NewTestResolver("/"),
+		Procs:         []string{".*"},
+		CgroupOpts:    cgroup.ReaderOptions{RootfsMountpoint: resolve.NewTestResolver("testdata")}, // procs here have no cgroup data, leading to errors
+	}
+	err := testStats.Init()
+	require.NoError(t, err)
+
+	// make sure we still have proc data despite cgroups errors
+	procs, _, err := testStats.Get()
+	require.NoError(t, err)
+	t.Logf("got %d procs", len(procs))
+	require.NotEmpty(t, procs)
+}
+
 func TestParseIO(t *testing.T) {
 	path := resolve.NewTestResolver("testdata/")
 	data, err := getIOData(path, 42)

metric/system/process/testdata/proc/cgroups

@@ -0,0 +1,15 @@
+#subsys_name    hierarchy       num_cgroups     enabled
+cpuset  0       176     1
+cpu     0       176     1
+cpuacct 0       176     1
+blkio   0       176     1
+memory  0       176     1
+devices 0       176     1
+freezer 0       176     1
+net_cls 0       176     1
+perf_event      0       176     1
+net_prio        0       176     1
+hugetlb 0       176     1
+pids    0       176     1
+rdma    0       176     1
+misc    0       176     1

metric/system/process/testdata/proc/self/mountinfo

@@ -0,0 +1 @@
+29 23 0:26 / /sys/fs/cgroup rw,nosuid,nodev,noexec,relatime shared:4 - cgroup2 cgroup2 rw,seclabel,nsdelegate,memory_recursiveprot