1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
//go:build linux
package cgroups
import (
"errors"
"fmt"
"io/fs"
"path/filepath"
"strings"
"time"
"github.com/containerd/cgroups/v3/cgroup2"
"github.com/opencontainers/runtime-spec/specs-go"
"github.com/prometheus/client_golang/prometheus"
"github.com/sirupsen/logrus"
"gitlab.com/gitlab-org/gitaly/v16/internal/gitaly/config"
cgroupscfg "gitlab.com/gitlab-org/gitaly/v16/internal/gitaly/config/cgroups"
"gitlab.com/gitlab-org/gitaly/v16/internal/log"
)
type cgroupV2Handler struct {
cfg cgroupscfg.Config
cpuUsage *prometheus.GaugeVec
cpuCFSPeriods *prometheus.Desc
cpuCFSThrottledPeriods *prometheus.Desc
cpuCFSThrottledTime *prometheus.Desc
procs *prometheus.GaugeVec
pid int
}
func newV2Handler(cfg cgroupscfg.Config, pid int) *cgroupV2Handler {
return &cgroupV2Handler{
cfg: cfg,
pid: pid,
cpuUsage: prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "gitaly_cgroup_cpu_usage_total",
Help: "CPU Usage of Cgroup",
},
[]string{"path", "type"},
),
cpuCFSPeriods: prometheus.NewDesc(
"gitaly_cgroup_cpu_cfs_periods_total",
"Number of elapsed enforcement period intervals",
[]string{"path"}, nil,
),
cpuCFSThrottledPeriods: prometheus.NewDesc(
"gitaly_cgroup_cpu_cfs_throttled_periods_total",
"Number of throttled period intervals",
[]string{"path"}, nil,
),
cpuCFSThrottledTime: prometheus.NewDesc(
"gitaly_cgroup_cpu_cfs_throttled_seconds_total",
"Total time duration the Cgroup has been throttled",
[]string{"path"}, nil,
),
procs: prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "gitaly_cgroup_procs_total",
Help: "Total number of procs",
},
[]string{"path", "subsystem"},
),
}
}
func (cvh *cgroupV2Handler) setupParent(parentResources *specs.LinuxResources) error {
if _, err := cgroup2.NewManager(cvh.cfg.Mountpoint, "/"+cvh.currentProcessCgroup(), cgroup2.ToResources(parentResources)); err != nil {
return fmt.Errorf("failed creating parent cgroup: %w", err)
}
return nil
}
func (cvh *cgroupV2Handler) setupRepository(reposResources *specs.LinuxResources) error {
for i := 0; i < int(cvh.cfg.Repositories.Count); i++ {
if _, err := cgroup2.NewManager(
cvh.cfg.Mountpoint,
"/"+cvh.repoPath(i),
cgroup2.ToResources(reposResources),
); err != nil {
return fmt.Errorf("failed creating repository cgroup: %w", err)
}
}
return nil
}
func (cvh *cgroupV2Handler) addToCgroup(pid int, cgroupPath string) error {
control, err := cgroup2.Load("/"+cgroupPath, cgroup2.WithMountpoint(cvh.cfg.Mountpoint))
if err != nil {
return fmt.Errorf("failed loading %s cgroup: %w", cgroupPath, err)
}
if err := control.AddProc(uint64(pid)); err != nil {
// Command could finish so quickly before we can add it to a cgroup, so
// we don't consider it an error.
if strings.Contains(err.Error(), "no such process") {
return nil
}
return fmt.Errorf("failed adding process to cgroup: %w", err)
}
return nil
}
func (cvh *cgroupV2Handler) collect(ch chan<- prometheus.Metric) {
if !cvh.cfg.MetricsEnabled {
return
}
for i := 0; i < int(cvh.cfg.Repositories.Count); i++ {
repoPath := cvh.repoPath(i)
logger := log.Default().WithField("cgroup_path", repoPath)
control, err := cgroup2.Load("/"+repoPath, cgroup2.WithMountpoint(cvh.cfg.Mountpoint))
if err != nil {
logger.WithError(err).Warn("unable to load cgroup controller")
return
}
if metrics, err := control.Stat(); err != nil {
logger.WithError(err).Warn("unable to get cgroup stats")
} else {
cpuUserMetric := cvh.cpuUsage.WithLabelValues(repoPath, "user")
cpuUserMetric.Set(float64(metrics.CPU.UserUsec))
ch <- cpuUserMetric
ch <- prometheus.MustNewConstMetric(
cvh.cpuCFSPeriods,
prometheus.CounterValue,
float64(metrics.CPU.NrPeriods),
repoPath,
)
ch <- prometheus.MustNewConstMetric(
cvh.cpuCFSThrottledPeriods,
prometheus.CounterValue,
float64(metrics.CPU.NrThrottled),
repoPath,
)
ch <- prometheus.MustNewConstMetric(
cvh.cpuCFSThrottledTime,
prometheus.CounterValue,
float64(metrics.CPU.ThrottledUsec)/float64(time.Second),
repoPath,
)
cpuKernelMetric := cvh.cpuUsage.WithLabelValues(repoPath, "kernel")
cpuKernelMetric.Set(float64(metrics.CPU.SystemUsec))
ch <- cpuKernelMetric
}
if subsystems, err := control.Controllers(); err != nil {
logger.WithError(err).Warn("unable to get cgroup hierarchy")
} else {
processes, err := control.Procs(true)
if err != nil {
logger.WithError(err).
Warn("unable to get process list")
continue
}
for _, subsystem := range subsystems {
procsMetric := cvh.procs.WithLabelValues(repoPath, subsystem)
procsMetric.Set(float64(len(processes)))
ch <- procsMetric
}
}
}
}
func (cvh *cgroupV2Handler) cleanup() error {
processCgroupPath := cvh.currentProcessCgroup()
control, err := cgroup2.Load("/"+processCgroupPath, cgroup2.WithMountpoint(cvh.cfg.Mountpoint))
if err != nil {
return fmt.Errorf("failed loading cgroup %s: %w", processCgroupPath, err)
}
if err := control.Delete(); err != nil {
return fmt.Errorf("failed cleaning up cgroup %s: %w", processCgroupPath, err)
}
return nil
}
func (cvh *cgroupV2Handler) repoPath(groupID int) string {
return filepath.Join(cvh.currentProcessCgroup(), fmt.Sprintf("repos-%d", groupID))
}
func (cvh *cgroupV2Handler) currentProcessCgroup() string {
return config.GetGitalyProcessTempDir(cvh.cfg.HierarchyRoot, cvh.pid)
}
func pruneOldCgroupsV2(cfg cgroupscfg.Config, logger logrus.FieldLogger) {
if err := config.PruneOldGitalyProcessDirectories(
logger,
filepath.Join(cfg.Mountpoint, cfg.HierarchyRoot),
); err != nil {
var pathError *fs.PathError
if !errors.As(err, &pathError) {
logger.WithError(err).Error("failed to clean up cpu cgroups")
}
}
}
|
