-
Notifications
You must be signed in to change notification settings - Fork 1.1k
/
stats_server.go
346 lines (317 loc) · 11.5 KB
/
stats_server.go
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
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
package statsserver
import (
"context"
"path/filepath"
"sync"
"time"
"github.com/containernetworking/plugins/pkg/ns"
cstorage "github.com/containers/storage"
"github.com/cri-o/cri-o/internal/lib/sandbox"
"github.com/cri-o/cri-o/internal/oci"
"github.com/cri-o/cri-o/pkg/config"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
"github.com/vishvananda/netlink"
types "k8s.io/cri-api/pkg/apis/runtime/v1"
)
// StatsServer is responsible for maintaining a list of container and sandbox stats.
// If collectionPeriod is > 0, it maintains this list by updating the stats on collectionPeriod frequency.
// Otherwise, it only updates the stats as they're requested.
type StatsServer struct {
shutdown chan struct{}
alreadyShutdown bool
collectionPeriod time.Duration
sboxStats map[string]*types.PodSandboxStats
ctrStats map[string]*types.ContainerStats
parentServerIface
sync.Mutex
}
// parentServerIface is an interface for requesting information from the parent ContainerServer.
// This is to be able to decouple stats server from the ContainerServer package, while also preventing
// data duplication (mainly in the active list of sandboxes), and avoid circular dependencies to boot.
type parentServerIface interface {
Runtime() *oci.Runtime
Store() cstorage.Store
ListSandboxes() []*sandbox.Sandbox
GetSandbox(string) *sandbox.Sandbox
Config() *config.Config
}
// New returns a new StatsServer, deriving the needed information from the provided parentServerIface.
func New(cs parentServerIface) *StatsServer {
ss := &StatsServer{
shutdown: make(chan struct{}, 1),
alreadyShutdown: false,
collectionPeriod: time.Duration(cs.Config().StatsCollectionPeriod) * time.Second,
sboxStats: make(map[string]*types.PodSandboxStats),
ctrStats: make(map[string]*types.ContainerStats),
parentServerIface: cs,
}
go ss.updateLoop()
return ss
}
// updateLoop updates the current list of stats every collectionPeriod seconds.
// If collectionPeriod is 0, it does nothing.
func (ss *StatsServer) updateLoop() {
if ss.collectionPeriod == 0 {
// fetch stats on-demand
return
}
for {
select {
case <-ss.shutdown:
return
case <-time.After(ss.collectionPeriod):
}
ss.update()
}
}
// update updates the list of container and sandbox stats.
// It does so by updating the stats of every sandbox, which in turn
// updates the stats for each container it has.
func (ss *StatsServer) update() {
ss.Lock()
defer ss.Unlock()
for _, sb := range ss.ListSandboxes() {
ss.updateSandbox(sb)
}
}
// updateSandbox updates the StatsServer's entry for this sandbox, as well as each child container.
// It first populates the stats from the CgroupParent, then calculates network usage, updates
// each of its children container stats by calling into the runtime, and finally calculates the CPUNanoCores.
func (ss *StatsServer) updateSandbox(sb *sandbox.Sandbox) *types.PodSandboxStats {
if sb == nil {
return nil
}
sandboxStats := &types.PodSandboxStats{
Attributes: &types.PodSandboxAttributes{
Id: sb.ID(),
Labels: sb.Labels(),
Metadata: sb.Metadata(),
Annotations: sb.Annotations(),
},
Linux: &types.LinuxPodSandboxStats{},
}
if err := ss.Config().CgroupManager().PopulateSandboxCgroupStats(sb.CgroupParent(), sandboxStats); err != nil {
logrus.Errorf("Error getting sandbox stats %s: %v", sb.ID(), err)
}
if err := ss.populateNetworkUsage(sandboxStats, sb); err != nil {
logrus.Errorf("Error adding network stats for sandbox %s: %v", sb.ID(), err)
}
containerStats := make([]*types.ContainerStats, 0, len(sb.Containers().List()))
for _, c := range sb.Containers().List() {
if c.StateNoLock().Status == oci.ContainerStateStopped {
continue
}
cStats, err := ss.Runtime().ContainerStats(context.TODO(), c, sb.CgroupParent())
if err != nil {
logrus.Errorf("Error getting container stats %s: %v", c.ID(), err)
continue
}
ss.populateWritableLayer(cStats, c)
if oldcStats, ok := ss.ctrStats[c.ID()]; ok {
updateUsageNanoCores(oldcStats.Cpu, cStats.Cpu)
}
}
sandboxStats.Linux.Containers = containerStats
if old, ok := ss.sboxStats[sb.ID()]; ok {
updateUsageNanoCores(old.Linux.Cpu, sandboxStats.Linux.Cpu)
}
ss.sboxStats[sb.ID()] = sandboxStats
return sandboxStats
}
// updateContainer calls into the runtime handler to update the container stats,
// as well as populates the writable layer by calling into the container storage.
// If this container already existed in the stats server, the CPU nano cores are calculated as well.
func (ss *StatsServer) updateContainer(c *oci.Container, sb *sandbox.Sandbox) *types.ContainerStats {
if c == nil || sb == nil {
return nil
}
if c.StateNoLock().Status == oci.ContainerStateStopped {
return nil
}
cStats, err := ss.Runtime().ContainerStats(context.TODO(), c, sb.CgroupParent())
if err != nil {
logrus.Errorf("Error getting container stats %s: %v", c.ID(), err)
return nil
}
ss.populateWritableLayer(cStats, c)
if oldcStats, ok := ss.ctrStats[c.ID()]; ok {
updateUsageNanoCores(oldcStats.Cpu, cStats.Cpu)
}
ss.ctrStats[c.ID()] = cStats
return cStats
}
// updateUsageNanoCores calculates the usage nano cores by averaging the CPU usage between the timestamps
// of the old usage and the recently gathered usage.
func updateUsageNanoCores(old, current *types.CpuUsage) {
if old == nil || current == nil || old.UsageCoreNanoSeconds == nil || current.UsageCoreNanoSeconds == nil {
return
}
nanoSeconds := current.Timestamp - old.Timestamp
usageNanoCores := uint64(float64(current.UsageCoreNanoSeconds.Value-old.UsageCoreNanoSeconds.Value) /
float64(nanoSeconds) * float64(time.Second/time.Nanosecond))
current.UsageNanoCores = &types.UInt64Value{
Value: usageNanoCores,
}
}
// populateWritableLayer attempts to populate the container stats writable layer.
func (ss *StatsServer) populateWritableLayer(stats *types.ContainerStats, container *oci.Container) {
writableLayer, err := ss.writableLayerForContainer(container)
if err != nil {
logrus.Error(err)
}
stats.WritableLayer = writableLayer
}
// writableLayerForContainer gathers information about the container's writable layer.
// It does so by calling into the GraphDriver's endpoint to get the UsedBytes and InodesUsed.
func (ss *StatsServer) writableLayerForContainer(container *oci.Container) (*types.FilesystemUsage, error) {
writableLayer := &types.FilesystemUsage{
Timestamp: time.Now().UnixNano(),
FsId: &types.FilesystemIdentifier{Mountpoint: container.MountPoint()},
}
driver, err := ss.Store().GraphDriver()
if err != nil {
return writableLayer, errors.Wrapf(err, "Unable to get graph driver for disk usage for container %s", container.ID())
}
id := filepath.Base(filepath.Dir(container.MountPoint()))
usage, err := driver.ReadWriteDiskUsage(id)
if err != nil {
return writableLayer, errors.Wrapf(err, "Unable to get disk usage for container %s", container.ID())
}
writableLayer.UsedBytes = &types.UInt64Value{Value: uint64(usage.Size)}
writableLayer.InodesUsed = &types.UInt64Value{Value: uint64(usage.InodeCount)}
return writableLayer, nil
}
// populateNetworkUsage gathers information about the network from within the sandbox's network namespace.
func (ss *StatsServer) populateNetworkUsage(stats *types.PodSandboxStats, sb *sandbox.Sandbox) error {
return ns.WithNetNSPath(sb.NetNsPath(), func(_ ns.NetNS) error {
links, err := netlink.LinkList()
if err != nil {
logrus.Errorf("Unable to retrieve network namespace links: %v", err)
return err
}
stats.Linux.Network = &types.NetworkUsage{
Interfaces: make([]*types.NetworkInterfaceUsage, 0, len(links)-1),
}
for i := range links {
iface, err := linkToInterface(links[i])
if err != nil {
logrus.Errorf("Failed to %v for pod %s", err, sb.ID())
continue
}
// TODO FIXME or DefaultInterfaceName?
if i == 0 {
stats.Linux.Network.DefaultInterface = iface
} else {
stats.Linux.Network.Interfaces = append(stats.Linux.Network.Interfaces, iface)
}
}
return nil
})
}
// linkToInterface translates information found from the netlink package
// into CRI the NetworkInterfaceUsage structure.
func linkToInterface(link netlink.Link) (*types.NetworkInterfaceUsage, error) {
attrs := link.Attrs()
if attrs == nil {
return nil, errors.New("get stats for iface")
}
if attrs.Statistics == nil {
return nil, errors.Errorf("get stats for iface %s", attrs.Name)
}
return &types.NetworkInterfaceUsage{
Name: attrs.Name,
RxBytes: &types.UInt64Value{Value: attrs.Statistics.RxBytes},
RxErrors: &types.UInt64Value{Value: attrs.Statistics.RxErrors},
TxBytes: &types.UInt64Value{Value: attrs.Statistics.TxBytes},
TxErrors: &types.UInt64Value{Value: attrs.Statistics.TxErrors},
}, nil
}
// StatsForSandbox returns the stats for the given sandbox
func (ss *StatsServer) StatsForSandbox(sb *sandbox.Sandbox) *types.PodSandboxStats {
ss.Lock()
defer ss.Unlock()
return ss.statsForSandbox(sb)
}
// StatsForSandboxes returns the stats for the given list of sandboxes
func (ss *StatsServer) StatsForSandboxes(sboxes []*sandbox.Sandbox) []*types.PodSandboxStats {
ss.Lock()
defer ss.Unlock()
stats := make([]*types.PodSandboxStats, 0, len(sboxes))
for _, sb := range sboxes {
if stat := ss.statsForSandbox(sb); stat != nil {
stats = append(stats, stat)
}
}
return stats
}
// statsForSandbox is an internal, non-locking version of StatsForSandbox
// that returns (and occasionally gathers) the stats for the given sandbox.
func (ss *StatsServer) statsForSandbox(sb *sandbox.Sandbox) *types.PodSandboxStats {
if ss.collectionPeriod == 0 {
return ss.updateSandbox(sb)
}
sboxStat, ok := ss.sboxStats[sb.ID()]
if ok {
return sboxStat
}
// Cache miss, try again
return ss.updateSandbox(sb)
}
// RemoveStatsForSandbox removes the saved entry for the specified sandbox
// to prevent the map from always growing.
func (ss *StatsServer) RemoveStatsForSandbox(sb *sandbox.Sandbox) {
ss.Lock()
defer ss.Unlock()
delete(ss.sboxStats, sb.ID())
}
// StatsForContainer returns the stats for the given container
func (ss *StatsServer) StatsForContainer(c *oci.Container, sb *sandbox.Sandbox) *types.ContainerStats {
ss.Lock()
defer ss.Unlock()
return ss.statsForContainer(c, sb)
}
// StatsForContainers returns the stats for the given list of containers
func (ss *StatsServer) StatsForContainers(ctrs []*oci.Container) []*types.ContainerStats {
ss.Lock()
defer ss.Unlock()
stats := make([]*types.ContainerStats, 0, len(ctrs))
for _, c := range ctrs {
sb := ss.GetSandbox(c.Sandbox())
if sb == nil {
logrus.Errorf("Unexpectedly failed to get sandbox %s for container %s", c.Sandbox(), c.ID())
continue
}
if stat := ss.statsForContainer(c, sb); stat != nil {
stats = append(stats, stat)
}
}
return stats
}
// statsForContainer is an internal, non-locking version of StatsForContainer
// that returns (and occasionally gathers) the stats for the given container.
func (ss *StatsServer) statsForContainer(c *oci.Container, sb *sandbox.Sandbox) *types.ContainerStats {
if ss.collectionPeriod == 0 {
return ss.updateContainer(c, sb)
}
ctrStat, ok := ss.ctrStats[c.ID()]
if ok {
return ctrStat
}
return ss.updateContainer(c, sb)
}
// RemoveStatsForContainer removes the saved entry for the specified container
// to prevent the map from always growing.
func (ss *StatsServer) RemoveStatsForContainer(c *oci.Container) {
ss.Lock()
defer ss.Unlock()
delete(ss.ctrStats, c.ID())
}
// Shutdown tells the updateLoop to stop updating.
func (ss *StatsServer) Shutdown() {
if ss.alreadyShutdown {
return
}
close(ss.shutdown)
ss.alreadyShutdown = true
}