/
fault_set.go
188 lines (153 loc) · 4.78 KB
/
fault_set.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
package urkel
import (
"bytes"
"context"
"fmt"
"io"
"strings"
"github.com/stretchr/testify/require"
"google.golang.org/grpc"
v1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/client-go/kubernetes/scheme"
"k8s.io/client-go/tools/remotecommand"
)
// FaultSet dispatches a set of removable faults to apply to a collection of Pods.
type FaultSet struct {
t require.TestingT
streams map[string]Chaos_InjectFaultClient
}
// NewFaultSet returns an empty FaultSet.
func NewFaultSet(t require.TestingT) *FaultSet {
return &FaultSet{t: t, streams: make(map[string]Chaos_InjectFaultClient)}
}
// Partition Pods in |partA| from Pods in |partB|. The PartitionMode determines
// whether connections are actively reset (REJECT) or packets are passively
// ignored (DROP).
func (fs *FaultSet) Partition(partA, partB []v1.Pod, mode PartitionMode) {
fmt.Println("partition ", podNames(partA), podNames(partB), mode)
// Determine the iflink index of each pod.
var ind = make(map[string]string)
for _, p := range partA {
ind[p.Name] = execRemote(fs.t, &p, "cat", "/sys/class/net/eth0/iflink")
}
for _, p := range partB {
ind[p.Name] = execRemote(fs.t, &p, "cat", "/sys/class/net/eth0/iflink")
}
// Dispatch bi-directional Partition faults for each pairing. The bi-
// directionality is needed to properly partition w.r.t the K8s service
// address, as a unilateral partition would still allow flow to a
// partitioned pod which happens to traverse over a service address IP.
for _, a := range partA {
var ifIndex = ind[a.Name]
for _, b := range partB {
fs.install(a, Fault{
Partition: &Partition{
InterfaceIndex: ifIndex,
FromIpRange: b.Status.PodIP,
Mode: string(mode),
},
})
}
}
for _, b := range partB {
var ifIndex = ind[b.Name]
for _, a := range partA {
fs.install(b, Fault{
Partition: &Partition{
InterfaceIndex: ifIndex,
FromIpRange: a.Status.PodIP,
Mode: string(mode),
},
})
}
}
}
// TODO(johnny): Support for CPU & disk stress; filled disks; packet latency and jitter.
// Crash Pods immediately, with no grace period. Crash faults are not remove-able.
func (fs *FaultSet) Crash(pods ...v1.Pod) {
fmt.Println("crash ", podNames(pods))
var zero = new(int64)
var coreV1 = kubeClient(fs.t).CoreV1()
for _, p := range pods {
require.NoError(fs.t, coreV1.Pods(p.Namespace).Delete(p.Name, &metav1.DeleteOptions{
GracePeriodSeconds: zero,
}))
}
}
// Delete pods using their default GracePeriodSeconds policy. Deletion faults are not remove-able.
func (fs *FaultSet) Delete(pods ...v1.Pod) {
fmt.Println("delete ", podNames(pods))
var coreV1 = kubeClient(fs.t).CoreV1()
for _, p := range pods {
require.NoError(fs.t, coreV1.Pods(p.Namespace).Delete(p.Name, &metav1.DeleteOptions{}))
}
}
// RemoveAll previously installed faults.
func (fs *FaultSet) RemoveAll() {
for _, s := range fs.streams {
require.NoError(fs.t, s.CloseSend())
var _, err = s.Recv()
require.Equal(fs.t, err, io.EOF)
}
}
// install a fault of the |pod| via a daemon running on the Pod's host.
func (fs *FaultSet) install(pod v1.Pod, fault Fault) {
var addr = pod.Status.HostIP + ":1666"
var err error
var s, ok = fs.streams[addr]
if !ok {
faultConnMu.Lock()
var conn, ok = faultConns[addr]
faultConnMu.Unlock()
if !ok {
conn, err = grpc.Dial(addr, grpc.WithInsecure())
require.NoError(fs.t, err)
faultConnMu.Lock()
faultConns[addr] = conn
faultConnMu.Unlock()
}
s, err = NewChaosClient(conn).InjectFault(context.Background())
require.NoError(fs.t, err, "starting fault stream")
fs.streams[addr] = s
}
require.NoError(fs.t, s.Send(&fault))
_, err = s.Recv() // Read confirmation.
require.NoError(fs.t, err)
}
// execRemote command |args| on the given |pod|.
func execRemote(t require.TestingT, pod *v1.Pod, args ...string) string {
var stdout, stderr bytes.Buffer
req := kubeClient(t).CoreV1().RESTClient().Post().
Resource("pods").
Name(pod.Name).
Namespace(pod.Namespace).
SubResource("exec")
req.VersionedParams(&v1.PodExecOptions{
Stdin: false,
Stdout: true,
Stderr: true,
TTY: false,
Container: pod.Spec.Containers[0].Name,
Command: args,
}, scheme.ParameterCodec)
rc, err := remotecommand.NewSPDYExecutor(kubeConfig(t), "POST", req.URL())
require.NoError(t, err, "starting exec stream")
err = rc.Stream(remotecommand.StreamOptions{
Stdin: nil,
Stdout: &stdout,
Stderr: &stderr,
Tty: false,
TerminalSizeQueue: nil,
})
require.NoError(t, err, "reading exec stream")
require.Empty(t, stderr.String())
return strings.TrimSpace(stdout.String())
}
func podNames(pods []v1.Pod) []string {
var out []string
for _, p := range pods {
out = append(out, p.Name)
}
return out
}