-
Notifications
You must be signed in to change notification settings - Fork 611
/
services.go
247 lines (211 loc) · 7.32 KB
/
services.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
package orchestrator
import (
"sort"
"github.com/docker/go-events"
"github.com/docker/swarmkit/api"
"github.com/docker/swarmkit/log"
"github.com/docker/swarmkit/manager/state"
"github.com/docker/swarmkit/manager/state/store"
"golang.org/x/net/context"
)
// This file provices service-level orchestration. It observes changes to
// services and creates and destroys tasks as necessary to match the service
// specifications. This is different from task-level orchestration, which
// responds to changes in individual tasks (or nodes which run them).
func (r *ReplicatedOrchestrator) initCluster(readTx store.ReadTx) error {
clusters, err := store.FindClusters(readTx, store.ByName("default"))
if err != nil {
return err
}
if len(clusters) != 1 {
// we'll just pick it when it is created.
return nil
}
r.cluster = clusters[0]
return nil
}
func (r *ReplicatedOrchestrator) initServices(readTx store.ReadTx) error {
services, err := store.FindServices(readTx, store.All)
if err != nil {
return err
}
for _, s := range services {
if isReplicatedService(s) {
r.reconcileServices[s.ID] = s
}
}
return nil
}
func (r *ReplicatedOrchestrator) handleServiceEvent(ctx context.Context, event events.Event) {
switch v := event.(type) {
case state.EventDeleteService:
if !isReplicatedService(v.Service) {
return
}
deleteServiceTasks(ctx, r.store, v.Service)
r.restarts.ClearServiceHistory(v.Service.ID)
case state.EventCreateService:
if !isReplicatedService(v.Service) {
return
}
r.reconcileServices[v.Service.ID] = v.Service
case state.EventUpdateService:
if !isReplicatedService(v.Service) {
return
}
r.reconcileServices[v.Service.ID] = v.Service
}
}
func (r *ReplicatedOrchestrator) tickServices(ctx context.Context) {
if len(r.reconcileServices) > 0 {
for _, s := range r.reconcileServices {
r.reconcile(ctx, s)
}
r.reconcileServices = make(map[string]*api.Service)
}
}
func (r *ReplicatedOrchestrator) resolveService(ctx context.Context, task *api.Task) *api.Service {
if task.ServiceID == "" {
return nil
}
var service *api.Service
r.store.View(func(tx store.ReadTx) {
service = store.GetService(tx, task.ServiceID)
})
return service
}
type tasksByRunningState []*api.Task
func (ts tasksByRunningState) Len() int { return len(ts) }
func (ts tasksByRunningState) Swap(i, j int) { ts[i], ts[j] = ts[j], ts[i] }
func (ts tasksByRunningState) Less(i, j int) bool {
return ts[i].Status.State == api.TaskStateRunning && ts[j].Status.State != api.TaskStateRunning
}
type taskWithIndex struct {
task *api.Task
// index is a counter that counts this task as the nth instance of
// the service on its node. This is used for sorting the tasks so that
// when scaling down we leave tasks more evenly balanced.
index int
}
type tasksByIndex []taskWithIndex
func (ts tasksByIndex) Len() int { return len(ts) }
func (ts tasksByIndex) Swap(i, j int) { ts[i], ts[j] = ts[j], ts[i] }
func (ts tasksByIndex) Less(i, j int) bool {
if ts[i].index < 0 {
return false
}
return ts[i].index < ts[j].index
}
func (r *ReplicatedOrchestrator) reconcile(ctx context.Context, service *api.Service) {
var (
tasks []*api.Task
err error
)
r.store.View(func(tx store.ReadTx) {
tasks, err = store.FindTasks(tx, store.ByServiceID(service.ID))
})
if err != nil {
log.G(ctx).WithError(err).Errorf("reconcile failed finding tasks")
return
}
runningTasks := make([]*api.Task, 0, len(tasks))
runningInstances := make(map[uint64]struct{}) // this could be a bitfield...
for _, t := range tasks {
// Technically the check below could just be
// t.DesiredState <= api.TaskStateRunning, but ignoring tasks
// with DesiredState == NEW simplifies the drainer unit tests.
if t.DesiredState > api.TaskStateNew && t.DesiredState <= api.TaskStateRunning {
runningTasks = append(runningTasks, t)
runningInstances[t.Slot] = struct{}{}
}
}
numTasks := len(runningTasks)
deploy := service.Spec.GetMode().(*api.ServiceSpec_Replicated)
specifiedInstances := int(deploy.Replicated.Replicas)
switch {
case specifiedInstances > numTasks:
log.G(ctx).Debugf("Service %s was scaled up from %d to %d instances", service.ID, numTasks, specifiedInstances)
// Update all current tasks then add missing tasks
r.updater.Update(ctx, r.cluster, service, runningTasks)
_, err = r.store.Batch(func(batch *store.Batch) error {
r.addTasks(ctx, batch, service, runningInstances, specifiedInstances-numTasks)
return nil
})
if err != nil {
log.G(ctx).WithError(err).Errorf("reconcile batch failed")
}
case specifiedInstances < numTasks:
// Update up to N tasks then remove the extra
log.G(ctx).Debugf("Service %s was scaled down from %d to %d instances", service.ID, numTasks, specifiedInstances)
// Preferentially remove tasks on the nodes that have the most
// copies of this service, to leave a more balanced result.
// First sort tasks such that tasks which are currently running
// (in terms of observed state) appear before non-running tasks.
// This will cause us to prefer to remove non-running tasks, all
// other things being equal in terms of node balance.
sort.Sort(tasksByRunningState(runningTasks))
// Assign each task an index that counts it as the nth copy of
// of the service on its node (1, 2, 3, ...), and sort the
// tasks by this counter value.
instancesByNode := make(map[string]int)
tasksWithIndices := make(tasksByIndex, 0, numTasks)
for _, t := range runningTasks {
if t.NodeID != "" {
instancesByNode[t.NodeID]++
tasksWithIndices = append(tasksWithIndices, taskWithIndex{task: t, index: instancesByNode[t.NodeID]})
} else {
tasksWithIndices = append(tasksWithIndices, taskWithIndex{task: t, index: -1})
}
}
sort.Sort(tasksWithIndices)
sortedTasks := make([]*api.Task, 0, numTasks)
for _, t := range tasksWithIndices {
sortedTasks = append(sortedTasks, t.task)
}
r.updater.Update(ctx, r.cluster, service, sortedTasks[:specifiedInstances])
_, err = r.store.Batch(func(batch *store.Batch) error {
r.removeTasks(ctx, batch, service, sortedTasks[specifiedInstances:])
return nil
})
if err != nil {
log.G(ctx).WithError(err).Errorf("reconcile batch failed")
}
case specifiedInstances == numTasks:
// Simple update, no scaling - update all tasks.
r.updater.Update(ctx, r.cluster, service, runningTasks)
}
}
func (r *ReplicatedOrchestrator) addTasks(ctx context.Context, batch *store.Batch, service *api.Service, runningInstances map[uint64]struct{}, count int) {
instance := uint64(0)
for i := 0; i < count; i++ {
// Find an instance number that is missing a running task
for {
instance++
if _, ok := runningInstances[instance]; !ok {
break
}
}
err := batch.Update(func(tx store.Tx) error {
return store.CreateTask(tx, newTask(r.cluster, service, instance))
})
if err != nil {
log.G(ctx).Errorf("Failed to create task: %v", err)
}
}
}
func (r *ReplicatedOrchestrator) removeTasks(ctx context.Context, batch *store.Batch, service *api.Service, tasks []*api.Task) {
for _, t := range tasks {
err := batch.Update(func(tx store.Tx) error {
// TODO(aaronl): optimistic update?
t = store.GetTask(tx, t.ID)
if t != nil {
t.DesiredState = api.TaskStateShutdown
return store.UpdateTask(tx, t)
}
return nil
})
if err != nil {
log.G(ctx).WithError(err).Errorf("removing task %s failed", t.ID)
}
}
}