leaderless gossip-based service discovery w/ consul compatibility
fix runaway cpu bug, simplify setupAgent
+35
-29
+6
README
+6
README
+2
-9
agent.go
+2
-9
agent.go
···
8
8
"github.com/hashicorp/serf/serf"
9
9
)
10
10
11
-
// Agent starts and manages a Serf & adds
12
-
// service discovery
11
+
// Agent starts and manages a Serf & adds service discovery (TODO)
13
12
type Agent struct {
14
-
// conf *Config
15
13
serfConf *serf.Config
16
14
eventCh chan serf.Event
17
-
18
-
// i doubt we care about handling events,
19
-
// but i'm leaving this here just in case
20
-
// eventHandlerList []EventHandler
21
-
// eventHandlersLock sync.Mutex
22
15
23
16
// This is the underlying Serf we are wrapping
24
17
serf *serf.Serf
···
53
46
log.Printf("[INFO] cascade: Received event: %s", e.String())
54
47
55
48
case <-serfShutdownCh:
56
-
log.Printf("[WARN] cascade: Serf shutdown detected, quitting")
49
+
log.Printf("[INFO] cascade: serf shutdown detected, quitting eventloop")
57
50
a.Shutdown()
58
51
return
59
52
+7
config.go
+7
config.go
···
3
3
import (
4
4
"fmt"
5
5
"net"
6
+
"os"
6
7
)
7
8
8
9
const DefaultBindPort int = 4449
9
10
10
11
func DefaultConfig() *Config {
12
+
hostname, err := os.Hostname()
13
+
if err != nil {
14
+
panic(err)
15
+
}
16
+
11
17
return &Config{
12
18
BindAddr: "0.0.0.0",
19
+
NodeName: hostname,
13
20
}
14
21
}
15
22
+19
-20
main.go
+19
-20
main.go
···
66
66
67
67
// Bail fast if not doing a graceful leave
68
68
if !graceful {
69
-
log.Printf("[WARN] cascade: non-graceful leave detected")
70
-
return nil
69
+
log.Fatal("[WARN] cascade: non-graceful leave detected")
71
70
}
72
71
73
72
// Attempt a graceful leave
···
129
128
if err != nil {
130
129
log.Panic(err)
131
130
}
131
+
132
132
serfConfig := serf.DefaultConfig()
133
133
serfConfig.NodeName = config.NodeName
134
134
serfConfig.ProtocolVersion = uint8(serf.ProtocolVersionMax)
135
-
serfConfig.CoalescePeriod = 3 * time.Second
136
-
serfConfig.QuiescentPeriod = time.Second
137
-
serfConfig.QueryResponseSizeLimit = 1024
138
-
serfConfig.QuerySizeLimit = 1024
139
-
serfConfig.UserEventSizeLimit = 512
140
-
serfConfig.UserCoalescePeriod = 3 * time.Second
141
-
serfConfig.UserQuiescentPeriod = time.Second
142
-
// TODO: look at reconnect/tombstone settings w more scrutiny
143
-
serfConfig.ReconnectInterval = 0
144
-
serfConfig.ReconnectTimeout = 0
145
-
serfConfig.TombstoneTimeout = 0
146
-
serfConfig.BroadcastTimeout = 0
147
-
// TODO: what are the implications of true here o_O
148
-
serfConfig.EnableNameConflictResolution = true
149
-
// hardcode DefaultWANConfig because cascade is designed to be
150
-
// used as a single global system.
135
+
// TODO: how should cascade handle name conflicts?
136
+
// defaulting to just blowing up for now, but
137
+
// we _could_ take the tailscale route & append
138
+
// -1 or whatever to the node. that would be more user friendly.
139
+
// TODO: some of these serf settings were pulled
140
+
// from consul[1]. re-examine them eventually.
141
+
serfConfig.EnableNameConflictResolution = false
142
+
serfConfig.ReconnectTimeout = 3 * 24 * time.Hour
143
+
serfConfig.QueueDepthWarning = 1000000
144
+
serfConfig.MinQueueDepth = 4096
145
+
serfConfig.LeavePropagateDelay = 3 * time.Second
146
+
151
147
serfConfig.MemberlistConfig = memberlist.DefaultWANConfig()
148
+
serfConfig.MemberlistConfig.DeadNodeReclaimTime = 30 * time.Second
152
149
serfConfig.MemberlistConfig.BindAddr = bindIP
153
150
serfConfig.MemberlistConfig.BindPort = bindPort
154
-
serfConfig.MemberlistConfig.AdvertiseAddr = ""
155
-
serfConfig.MemberlistConfig.AdvertisePort = 0
156
151
157
152
agent := Create(serfConfig)
158
153
return agent
159
154
}
155
+
156
+
// [1]: sources for consul serf tweaks
157
+
// https://github.com/hashicorp/consul/blob/v1.14.4/agent/consul/config.go
158
+
// https://github.com/hashicorp/consul/blob/v1.14.4/lib/serf/serf.go