leaderless gossip-based service discovery w/ consul compatibility
Make error logging more accurate
+38
-9
+4
-5
agent.go
+4
-5
agent.go
···
98
98
99
99
// Join asks the Serf instance to join. See the Serf.Join function.
100
100
func (a *Agent) Join(addrs []string) (n int, err error) {
101
-
log.Printf("[INFO] cascade: joining: %v", addrs)
101
+
log.Printf("[INFO] cascade: issuing join request to '%v'", addrs)
102
102
// we always ignore old events because cascade don't
103
103
// care about the past
104
104
n, err = a.serf.Join(addrs, true)
105
-
if n > 0 {
106
-
log.Printf("[INFO] cascade: joined: %d nodes", n)
107
-
}
108
105
if err != nil {
109
-
log.Printf("[WARN] cascade: error joining: %v", err)
106
+
return n, fmt.Errorf("[ERR] cascade: error joining: %v\n", err)
110
107
}
108
+
// TODO: when joining fails, we don't get an error here - serf & memberlist
109
+
// just print to stdout and serf.Join returns without issue.
111
110
return n, err
112
111
}
113
112
+18
-4
main.go
+18
-4
main.go
···
24
24
const (
25
25
// gracefulTimeout controls how long we wait before forcefully terminating
26
26
// note that this value interacts with serfConfig.LeavePropagateDelay
27
-
gracefulTimeout = 5 * time.Second
27
+
gracefulTimeout = 10 * time.Second
28
28
)
29
29
30
30
func main() {
···
36
36
defer agent.Shutdown()
37
37
// join any specified startup nodes
38
38
if err := startupJoin(config, agent); err != nil {
39
-
log.Panic(err)
39
+
log.Fatal(err)
40
+
}
41
+
if os.Getenv("CASCADE_DEBUG") != "" {
42
+
go debugPrints(agent)
40
43
}
41
44
if err := handleSignals(config, agent); err != nil {
42
45
log.Panic(err)
46
+
}
47
+
}
48
+
49
+
func debugPrints(agent *Agent) {
50
+
for {
51
+
for _, m := range agent.serf.Members() {
52
+
log.Printf("[DEBUG] cascade: name '%s' addr '%s' status '%s' \n",
53
+
m.Name, m.Addr, m.Status)
54
+
}
55
+
time.Sleep(time.Second * 5)
43
56
}
44
57
}
45
58
···
114
127
return nil
115
128
}
116
129
117
-
log.Printf("Joining cluster...")
118
130
n, err := agent.Join(config.StartJoin)
119
131
if err != nil {
120
132
return err
121
133
}
134
+
if n > 0 {
135
+
log.Printf("[INFO] cascade: join request issued to %d nodes", n)
136
+
}
122
137
123
-
log.Printf("Join completed. Synced with %d initial agents", n)
124
138
return nil
125
139
}
126
140
+16
make-test-cluster
+16
make-test-cluster
···
1
+
#!/bin/sh
2
+
#
3
+
# make a little cascade cluster for development & debugging
4
+
5
+
export CASCADE_DEBUG=true
6
+
7
+
trap "trap - SIGTERM && kill -- -$$" SIGINT SIGTERM EXIT
8
+
# error case:
9
+
# CASCADE_BIND='127.0.0.1' go run . &
10
+
# sleep 5
11
+
# CASCADE_BIND='127.0.0.2' CASCADE_JOIN='127.0.0.1' go run . &
12
+
# happy case:
13
+
CASCADE_NAME=foo CASCADE_BIND='127.0.0.1' go run . &
14
+
sleep 5
15
+
CASCADE_NAME=bar CASCADE_BIND='127.0.0.2' CASCADE_JOIN='127.0.0.1' go run . &
16
+
wait