update metric tests

3 files changed, 71 insertions, 8 deletions

diff --git a/internal/cgroups/mock_linux_test.go b/internal/cgroups/mock_linux_test.go
index b941bda23..f21180e19 100644
--- a/internal/cgroups/mock_linux_test.go
+++ b/internal/cgroups/mock_linux_test.go
@@ -112,5 +112,15 @@ func (m *mockCgroup) setupMockCgroupFiles(
 				require.NoError(t, os.WriteFile(shardControlFilePath, []byte(content), 0o644))
 			}
 		}
+
+		for shard := uint(0); shard < manager.cfg.Git.Count; shard++ {
+			shardPath := filepath.Join(cgroupPath, fmt.Sprintf("git-commands-%d", shard))
+			require.NoError(t, os.MkdirAll(shardPath, 0o755))
+
+			for filename, content := range contentByFilename {
+				shardControlFilePath := filepath.Join(shardPath, filename)
+				require.NoError(t, os.WriteFile(shardControlFilePath, []byte(content), 0o644))
+			}
+		}
 	}
 }
diff --git a/internal/cgroups/v1_linux.go b/internal/cgroups/v1_linux.go
index f5404fdfc..48d3aeb70 100644
--- a/internal/cgroups/v1_linux.go
+++ b/internal/cgroups/v1_linux.go
@@ -20,7 +20,8 @@ type CGroupV1Manager struct {
 	cfg                         cgroupscfg.Config
 	hierarchy                   func() ([]cgroups.Subsystem, error)
 	memoryFailedTotal, cpuUsage *prometheus.GaugeVec
-	procs                       *prometheus.GaugeVec
+	procs, gitCgroups           *prometheus.GaugeVec
+	gitWithoutCgroup            prometheus.Counter
 	gitCmds                     map[string]cgroupscfg.Command
 	gitCgroupPool               chan string
 }
@@ -52,6 +53,19 @@ func newV1Manager(cfg cgroupscfg.Config) *CGroupV1Manager {
 			},
 			[]string{"path", "subsystem"},
 		),
+		gitCgroups: prometheus.NewGaugeVec(
+			prometheus.GaugeOpts{
+				Name: "gitaly_cgroup_git_cgroups",
+				Help: "Total number of git cgroups in use",
+			},
+			[]string{"status"},
+		),
+		gitWithoutCgroup: prometheus.NewCounter(
+			prometheus.CounterOpts{
+				Name: "gitaly_cgroup_git_without_cgroup_total",
+				Help: "Total number of git commands without a cgroup",
+			},
+		),
 		gitCmds: make(map[string]cgroupscfg.Command),
 	}
 }
@@ -139,6 +153,10 @@ func (cg *CGroupV1Manager) AddCommand(
 			return fmt.Errorf("failed adding process to cgroup: %w", err)
 		}
 	default:
+		cg.gitWithoutCgroup.Inc()
+		log.Default().WithField("git_cmd", gitCmdName).
+			WithField("relative_path", repo.GetRelativePath()).
+			Warn("git command could not be added to a cgroup")
 		return nil
 	}
 
@@ -218,6 +236,15 @@ func (cg *CGroupV1Manager) Collect(ch chan<- prometheus.Metric) {
 			ch <- procsMetric
 		}
 	}
+
+	gitCgroupsTotalMetric := cg.gitCgroups.WithLabelValues("total")
+	gitCgroupsTotalMetric.Set(float64(cg.cfg.Git.Count))
+	ch <- gitCgroupsTotalMetric
+
+	gitCgroupsInUseMetric := cg.gitCgroups.WithLabelValues("in_use")
+	gitCgroupsInUseMetric.Set(float64(int(cg.cfg.Git.Count) - len(cg.gitCgroupPool)))
+	ch <- gitCgroupsInUseMetric
+	ch <- cg.gitWithoutCgroup
 }
 
 // Describe describes the cgroup metrics that Collect provides
diff --git a/internal/cgroups/v1_linux_test.go b/internal/cgroups/v1_linux_test.go
index 8bf769e15..1ce1c6f94 100644
--- a/internal/cgroups/v1_linux_test.go
+++ b/internal/cgroups/v1_linux_test.go
@@ -186,6 +186,16 @@ func TestMetrics(t *testing.T) {
 	}
 
 	config := defaultCgroupsConfig()
+	config.Git = cgroups.Git{
+		Count: 1,
+		Commands: []cgroups.Command{
+			cgroups.Command{
+				Name:        "git-cmd-1",
+				MemoryBytes: 1048576,
+				CPUShares:   16,
+			},
+		},
+	}
 	v1Manager1 := newV1Manager(config)
 	v1Manager1.hierarchy = mock.hierarchy
 
@@ -198,12 +208,18 @@ func TestMetrics(t *testing.T) {
 	logger.SetLevel(logrus.DebugLevel)
 	ctx = ctxlogrus.ToContext(ctx, logrus.NewEntry(logger))
 
-	cmd1 := exec.Command("ls", "-hal", ".")
-	cmd2, err := command.New(ctx, cmd1, nil, nil, nil)
+	cmd, err := command.New(ctx, exec.Command("ls", "-hal", "."), nil, nil, nil)
+	require.NoError(t, err)
+	gitCmd1, err := command.New(ctx, exec.Command("ls", "-hal", "."), nil, nil, nil)
+	require.NoError(t, err)
+	gitCmd2, err := command.New(ctx, exec.Command("ls", "-hal", "."), nil, nil, nil)
 	require.NoError(t, err)
 
-	require.NoError(t, v1Manager1.AddCommand(cmd2, "git-cmd", repo))
-	require.NoError(t, cmd2.Wait())
+	require.NoError(t, v1Manager1.AddCommand(cmd, "cmd-1", repo))
+	require.NoError(t, v1Manager1.AddCommand(gitCmd1, "git-cmd-1", repo))
+	require.NoError(t, v1Manager1.AddCommand(gitCmd2, "git-cmd-1", repo))
+	require.NoError(t, cmd.Wait())
+	require.NoError(t, gitCmd1.Wait())
 
 	processCgroupPath := v1Manager1.currentProcessCgroup()
 
@@ -211,20 +227,30 @@ func TestMetrics(t *testing.T) {
 # TYPE gitaly_cgroup_cpu_usage gauge
 gitaly_cgroup_cpu_usage{path="%s",type="kernel"} 0
 gitaly_cgroup_cpu_usage{path="%s",type="user"} 0
+# HELP gitaly_cgroup_git_cgroups Total number of git cgroups in use
+# TYPE gitaly_cgroup_git_cgroups gauge
+gitaly_cgroup_git_cgroups{status="in_use"} 0
+gitaly_cgroup_git_cgroups{status="total"} 1
+# HELP gitaly_cgroup_git_without_cgroup_total Total number of git commands without a cgroup
+# TYPE gitaly_cgroup_git_without_cgroup_total counter
+gitaly_cgroup_git_without_cgroup_total 1
 # HELP gitaly_cgroup_memory_failed_total Number of memory usage hits limits
 # TYPE gitaly_cgroup_memory_failed_total gauge
 gitaly_cgroup_memory_failed_total{path="%s"} 2
 # HELP gitaly_cgroup_procs_total Total number of procs
 # TYPE gitaly_cgroup_procs_total gauge
-gitaly_cgroup_procs_total{path="%s",subsystem="memory"} 1
-gitaly_cgroup_procs_total{path="%s",subsystem="cpu"} 1
+gitaly_cgroup_procs_total{path="%s",subsystem="memory"} 2
+gitaly_cgroup_procs_total{path="%s",subsystem="cpu"} 2
 `, processCgroupPath, processCgroupPath, processCgroupPath, processCgroupPath, processCgroupPath))
 	assert.NoError(t, testutil.CollectAndCompare(
 		v1Manager1,
 		expected,
 		"gitaly_cgroup_memory_failed_total",
 		"gitaly_cgroup_cpu_usage",
-		"gitaly_cgroup_procs_total"))
+		"gitaly_cgroup_procs_total",
+		"gitaly_cgroup_git_cgroups",
+		"gitaly_cgroup_git_without_cgroup_total",
+	))
 
 	logEntry := hook.LastEntry()
 	assert.Contains(