forked from pulumi/pulumi
/
plan_executor.go
359 lines (316 loc) · 13.4 KB
/
plan_executor.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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
// Copyright 2016-2018, Pulumi Corporation.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package deploy
import (
"context"
"github.com/pkg/errors"
"github.com/pulumi/pulumi/pkg/diag"
"github.com/pulumi/pulumi/pkg/resource"
"github.com/pulumi/pulumi/pkg/resource/deploy/providers"
"github.com/pulumi/pulumi/pkg/resource/graph"
"github.com/pulumi/pulumi/pkg/util/contract"
"github.com/pulumi/pulumi/pkg/util/logging"
"github.com/pulumi/pulumi/pkg/util/result"
)
// planExecutor is responsible for taking a plan and driving it to completion.
// Its primary responsibility is to own a `stepGenerator` and `stepExecutor`, serving
// as the glue that links the two subsystems together.
type planExecutor struct {
plan *Plan // The plan that we are executing
stepGen *stepGenerator // step generator owned by this plan
stepExec *stepExecutor // step executor owned by this plan
}
// execError creates an error appropriate for returning from planExecutor.Execute.
func execError(message string, preview bool) error {
kind := "update"
if preview {
kind = "preview"
}
return errors.New(kind + " " + message)
}
// reportError reports a single error to the executor's diag stream with the indicated URN for context.
func (pe *planExecutor) reportError(urn resource.URN, err error) {
pe.plan.Diag().Errorf(diag.RawMessage(urn, err.Error()))
}
// Execute executes a plan to completion, using the given cancellation context and running a preview
// or update.
func (pe *planExecutor) Execute(callerCtx context.Context, opts Options, preview bool) error {
// Set up a goroutine that will signal cancellation to the plan's plugins if the caller context is cancelled. We do
// not hang this off of the context we create below because we do not want the failure of a single step to cause
// other steps to fail.
done := make(chan bool)
defer close(done)
go func() {
select {
case <-callerCtx.Done():
logging.V(4).Infof("planExecutor.Execute(...): signalling cancellation to providers...")
cancelErr := pe.plan.ctx.Host.SignalCancellation()
if cancelErr != nil {
logging.V(4).Infof("planExecutor.Execute(...): failed to signal cancellation to providers: %v", cancelErr)
}
case <-done:
logging.V(4).Infof("planExecutor.Execute(...): exiting provider canceller")
}
}()
// Before doing anything else, optionally refresh each resource in the base checkpoint.
if opts.Refresh {
if err := pe.refresh(callerCtx, opts, preview); err != nil {
return err
}
if opts.RefreshOnly {
return nil
}
}
// Begin iterating the source.
src, err := pe.plan.source.Iterate(callerCtx, opts, pe.plan)
if err != nil {
return err
}
// Set up a step generator for this plan.
pe.stepGen = newStepGenerator(pe.plan, opts)
// Retire any pending deletes that are currently present in this plan.
if err = pe.retirePendingDeletes(callerCtx, opts, preview); err != nil {
return err
}
// Derive a cancellable context for this plan. We will only cancel this context if some piece of the plan's
// execution fails.
ctx, cancel := context.WithCancel(callerCtx)
// Set up a step generator and executor for this plan.
pe.stepExec = newStepExecutor(ctx, cancel, pe.plan, opts, preview, false)
// We iterate the source in its own goroutine because iteration is blocking and we want the main loop to be able to
// respond to cancellation requests promptly.
type nextEvent struct {
Event SourceEvent
Error error
}
incomingEvents := make(chan nextEvent)
go func() {
for {
event, sourceErr := src.Next()
select {
case incomingEvents <- nextEvent{event, sourceErr}:
if event == nil {
return
}
case <-done:
logging.V(4).Infof("planExecutor.Execute(...): incoming events goroutine exiting")
return
}
}
}()
// The main loop. We'll continuously select for incoming events and the cancellation signal. There are
// a three ways we can exit this loop:
// 1. The SourceIterator sends us a `nil` event. This means that we're done processing source events and
// we should begin processing deletes.
// 2. The SourceIterator sends us an error. This means some error occurred in the source program and we
// should bail.
// 3. The stepExecCancel cancel context gets canceled. This means some error occurred in the step executor
// and we need to bail. This can also happen if the user hits Ctrl-C.
canceled, err := func() (bool, error) {
logging.V(4).Infof("planExecutor.Execute(...): waiting for incoming events")
for {
select {
case event := <-incomingEvents:
logging.V(4).Infof("planExecutor.Execute(...): incoming event (nil? %v, %v)", event.Event == nil, event.Error)
if event.Error != nil {
pe.reportError("", event.Error)
cancel()
return false, event.Error
}
if event.Event == nil {
deleteSteps := pe.stepGen.GenerateDeletes()
deletes := pe.stepGen.ScheduleDeletes(deleteSteps)
// ScheduleDeletes gives us a list of lists of steps. Each list of steps can safely be executed in
// parallel, but each list must execute completes before the next list can safely begin executing.
//
// This is not "true" delete parallelism, since there may be resources that could safely begin
// deleting but we won't until the previous set of deletes fully completes. This approximation is
// conservative, but correct.
for _, antichain := range deletes {
logging.V(4).Infof("planExecutor.Execute(...): beginning delete antichain")
tok := pe.stepExec.ExecuteParallel(antichain)
tok.Wait(ctx)
logging.V(4).Infof("planExecutor.Execute(...): antichain complete")
}
// We're done here - signal completion so that the step executor knows to terminate.
pe.stepExec.SignalCompletion()
return false, nil
}
if res := pe.handleSingleEvent(event.Event); res != nil {
if resErr := res.Error(); resErr != nil {
logging.V(4).Infof("planExecutor.Execute(...): error handling event: %v", resErr)
pe.reportError(pe.plan.generateEventURN(event.Event), resErr)
}
cancel()
return false, result.TODO()
}
case <-ctx.Done():
logging.V(4).Infof("planExecutor.Execute(...): context finished: %v", ctx.Err())
// NOTE: we use the presence of an error in the caller context in order to distinguish caller-initiated
// cancellation from internally-initiated cancellation.
return callerCtx.Err() != nil, nil
}
}
}()
pe.stepExec.WaitForCompletion()
logging.V(4).Infof("planExecutor.Execute(...): step executor has completed")
// Figure out if execution failed and why. Step generation and execution errors trump cancellation.
if err != nil || pe.stepExec.Errored() {
err = execError("failed", preview)
} else if canceled {
err = execError("canceled", preview)
}
return err
}
// handleSingleEvent handles a single source event. For all incoming events, it produces a chain that needs
// to be executed and schedules the chain for execution.
func (pe *planExecutor) handleSingleEvent(event SourceEvent) *result.Result {
contract.Require(event != nil, "event != nil")
var steps []Step
var res *result.Result
switch e := event.(type) {
case RegisterResourceEvent:
logging.V(4).Infof("planExecutor.handleSingleEvent(...): received RegisterResourceEvent")
steps, res = pe.stepGen.GenerateSteps(e)
case ReadResourceEvent:
logging.V(4).Infof("planExecutor.handleSingleEvent(...): received ReadResourceEvent")
steps, res = pe.stepGen.GenerateReadSteps(e)
case RegisterResourceOutputsEvent:
logging.V(4).Infof("planExecutor.handleSingleEvent(...): received register resource outputs")
pe.stepExec.ExecuteRegisterResourceOutputs(e)
return nil
}
if res != nil {
return res
}
pe.stepExec.ExecuteSerial(steps)
return nil
}
// retirePendingDeletes deletes all resources that are pending deletion. Run before the start of a plan, this pass
// ensures that the engine never sees any resources that are pending deletion from a previous plan.
//
// retirePendingDeletes re-uses the plan executor's step generator but uses its own step executor.
func (pe *planExecutor) retirePendingDeletes(callerCtx context.Context, opts Options, preview bool) error {
contract.Require(pe.stepGen != nil, "pe.stepGen != nil")
steps := pe.stepGen.GeneratePendingDeletes()
if len(steps) == 0 {
logging.V(4).Infoln("planExecutor.retirePendingDeletes(...): no pending deletions")
return nil
}
logging.V(4).Infof("planExecutor.retirePendingDeletes(...): executing %d steps", len(steps))
ctx, cancel := context.WithCancel(callerCtx)
stepExec := newStepExecutor(ctx, cancel, pe.plan, opts, preview, false)
antichains := pe.stepGen.ScheduleDeletes(steps)
// Submit the deletes for execution and wait for them all to retire.
for _, antichain := range antichains {
for _, step := range antichain {
pe.plan.Ctx().StatusDiag.Infof(diag.RawMessage(step.URN(), "completing deletion from previous update"))
}
tok := stepExec.ExecuteParallel(antichain)
tok.Wait(ctx)
}
stepExec.SignalCompletion()
stepExec.WaitForCompletion()
// Like Refresh, we use the presence of an error in the caller's context to detect whether or not we have been
// cancelled.
canceled := callerCtx.Err() != nil
if stepExec.Errored() {
return execError("failed", preview)
} else if canceled {
return execError("canceled", preview)
}
return nil
}
// refresh refreshes the state of the base checkpoint file for the current plan in memory.
func (pe *planExecutor) refresh(callerCtx context.Context, opts Options, preview bool) error {
prev := pe.plan.prev
if prev == nil || len(prev.Resources) == 0 {
return nil
}
// Create a refresh step for each resource in the old snapshot.
steps := make([]Step, len(prev.Resources))
for i := range prev.Resources {
steps[i] = NewRefreshStep(pe.plan, prev.Resources[i], nil)
}
// Fire up a worker pool and issue each refresh in turn.
ctx, cancel := context.WithCancel(callerCtx)
stepExec := newStepExecutor(ctx, cancel, pe.plan, opts, preview, true)
stepExec.ExecuteParallel(steps)
stepExec.SignalCompletion()
stepExec.WaitForCompletion()
// Rebuild this plan's map of old resources and dependency graph, stripping out any deleted resources and repairing
// dependency lists as necessary. Note that this updates the base snapshot _in memory_, so it is critical that any
// components that use the snapshot refer to the same instance and avoid reading it concurrently with this rebuild.
//
// The process of repairing dependency lists is a bit subtle. Because multiple physical resources may share a URN,
// the ability of a particular URN to be referenced in a dependency list can change based on the dependent
// resource's position in the resource list. For example, consider the following list of resources, where each
// resource is a (URN, ID, Dependencies) tuple:
//
// [ (A, 0, []), (B, 0, [A]), (A, 1, []), (A, 2, []), (C, 0, [A]) ]
//
// Let `(A, 0, [])` and `(A, 2, [])` be deleted by the refresh. This produces the following intermediate list
// before dependency lists are repaired:
//
// [ (B, 0, [A]), (A, 1, []), (C, 0, [A]) ]
//
// In order to repair the dependency lists, we iterate over the intermediate resource list, keeping track of which
// URNs refer to at least one physical resource at each point in the list, and remove any dependencies that refer
// to URNs that do not refer to any physical resources. This process produces the following final list:
//
// [ (B, 0, []), (A, 1, []), (C, 0, [A]) ]
//
// Note that the correctness of this process depends on the fact that the list of resources is a topological sort
// of its corresponding dependency graph, so a resource always appears in the list after any resources on which it
// may depend.
resources := make([]*resource.State, 0, len(prev.Resources))
referenceable := make(map[resource.URN]bool)
olds := make(map[resource.URN]*resource.State)
for _, s := range steps {
new := s.New()
if new == nil {
contract.Assert(s.Old().Custom)
contract.Assert(!providers.IsProviderType(s.Old().Type))
continue
}
// Remove any deleted resources from this resource's dependency list.
if len(new.Dependencies) != 0 {
deps := make([]resource.URN, 0, len(new.Dependencies))
for _, d := range new.Dependencies {
if referenceable[d] {
deps = append(deps, d)
}
}
new.Dependencies = deps
}
// Add this resource to the resource list and mark it as referenceable.
resources = append(resources, new)
referenceable[new.URN] = true
// Do not record resources that are pending deletion in the "olds" lookup table.
if !new.Delete {
olds[new.URN] = new
}
}
pe.plan.prev.Resources = resources
pe.plan.olds, pe.plan.depGraph = olds, graph.NewDependencyGraph(resources)
// NOTE: we use the presence of an error in the caller context in order to distinguish caller-initiated
// cancellation from internally-initiated cancellation.
canceled := callerCtx.Err() != nil
if stepExec.Errored() {
return execError("failed", preview)
} else if canceled {
return execError("canceled", preview)
}
return nil
}