-
Notifications
You must be signed in to change notification settings - Fork 607
/
cluster.go
215 lines (180 loc) · 5.13 KB
/
cluster.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
package membership
import (
"errors"
"sync"
"google.golang.org/grpc"
"github.com/coreos/etcd/raft/raftpb"
"github.com/docker/swarmkit/api"
"github.com/gogo/protobuf/proto"
)
var (
// ErrIDExists is thrown when a node wants to join the existing cluster but its ID already exists
ErrIDExists = errors.New("membership: can't add node to cluster, node id is a duplicate")
// ErrIDRemoved is thrown when a node tries to perform an operation on an existing cluster but was removed
ErrIDRemoved = errors.New("membership: node was removed during cluster lifetime")
// ErrIDNotFound is thrown when we try an operation on a member that does not exist in the cluster list
ErrIDNotFound = errors.New("membership: member not found in cluster list")
// ErrConfigChangeInvalid is thrown when a configuration change we received looks invalid in form
ErrConfigChangeInvalid = errors.New("membership: ConfChange type should be either AddNode, RemoveNode or UpdateNode")
// ErrCannotUnmarshalConfig is thrown when a node cannot unmarshal a configuration change
ErrCannotUnmarshalConfig = errors.New("membership: cannot unmarshal configuration change")
)
// Cluster represents a set of active
// raft Members
type Cluster struct {
mu sync.RWMutex
members map[uint64]*Member
// removed contains the list of removed Members,
// those ids cannot be reused
removed map[uint64]bool
}
// Member represents a raft Cluster Member
type Member struct {
*api.RaftMember
api.RaftClient
Conn *grpc.ClientConn
}
// NewCluster creates a new Cluster neighbors
// list for a raft Member
func NewCluster() *Cluster {
// TODO(abronan): generate Cluster ID for federation
return &Cluster{
members: make(map[uint64]*Member),
removed: make(map[uint64]bool),
}
}
// Members returns the list of raft Members in the Cluster.
func (c *Cluster) Members() map[uint64]*Member {
members := make(map[uint64]*Member)
c.mu.RLock()
for k, v := range c.members {
members[k] = v
}
c.mu.RUnlock()
return members
}
// Removed returns the list of raft Members removed from the Cluster.
func (c *Cluster) Removed() []uint64 {
c.mu.RLock()
removed := make([]uint64, 0, len(c.removed))
for k := range c.removed {
removed = append(removed, k)
}
c.mu.RUnlock()
return removed
}
// GetMember returns informations on a given Member.
func (c *Cluster) GetMember(id uint64) *Member {
c.mu.RLock()
defer c.mu.RUnlock()
return c.members[id]
}
// AddMember adds a node to the Cluster Memberlist.
func (c *Cluster) AddMember(member *Member) error {
c.mu.Lock()
defer c.mu.Unlock()
if c.removed[member.RaftID] {
return ErrIDRemoved
}
c.members[member.RaftID] = member
return nil
}
// RemoveMember removes a node from the Cluster Memberlist.
func (c *Cluster) RemoveMember(id uint64) error {
c.mu.Lock()
defer c.mu.Unlock()
if c.members[id] != nil {
conn := c.members[id].Conn
if conn != nil {
_ = conn.Close()
}
delete(c.members, id)
}
c.removed[id] = true
return nil
}
// ReplaceMemberConnection replaces the member's GRPC connection and GRPC
// client.
func (c *Cluster) ReplaceMemberConnection(id uint64, newConn *Member) error {
c.mu.Lock()
defer c.mu.Unlock()
oldMember, ok := c.members[id]
if !ok {
return ErrIDNotFound
}
oldMember.Conn.Close()
newMember := *oldMember
newMember.Conn = newConn.Conn
newMember.RaftClient = newConn.RaftClient
c.members[id] = &newMember
return nil
}
// IsIDRemoved checks if a Member is in the remove set.
func (c *Cluster) IsIDRemoved(id uint64) bool {
c.mu.RLock()
defer c.mu.RUnlock()
return c.removed[id]
}
// Clear resets the list of active Members and removed Members.
func (c *Cluster) Clear() {
c.mu.Lock()
c.members = make(map[uint64]*Member)
c.removed = make(map[uint64]bool)
c.mu.Unlock()
}
// ValidateConfigurationChange takes a proposed ConfChange and
// ensures that it is valid.
func (c *Cluster) ValidateConfigurationChange(cc raftpb.ConfChange) error {
c.mu.Lock()
defer c.mu.Unlock()
if c.removed[cc.NodeID] {
return ErrIDRemoved
}
switch cc.Type {
case raftpb.ConfChangeAddNode:
if c.members[cc.NodeID] != nil {
return ErrIDExists
}
case raftpb.ConfChangeRemoveNode:
if c.members[cc.NodeID] == nil {
return ErrIDNotFound
}
case raftpb.ConfChangeUpdateNode:
if c.members[cc.NodeID] == nil {
return ErrIDNotFound
}
default:
return ErrConfigChangeInvalid
}
m := &api.RaftMember{}
if err := proto.Unmarshal(cc.Context, m); err != nil {
return ErrCannotUnmarshalConfig
}
return nil
}
// CanRemoveMember checks if removing a Member would not result in a loss
// of quorum, this check is needed before submitting a configuration change
// that might block or harm the Cluster on Member recovery
func (c *Cluster) CanRemoveMember(from uint64, id uint64) bool {
members := c.Members()
nreachable := 0 // reachable managers after removal
for _, m := range members {
if m.RaftID == id {
continue
}
// Local node from where the remove is issued
if m.RaftID == from {
nreachable++
continue
}
connState, err := m.Conn.State()
if err == nil && connState == grpc.Ready {
nreachable++
}
}
nquorum := (len(members)-1)/2 + 1
if nreachable < nquorum {
return false
}
return true
}