pulumi/pkg/engine/update.go
Pat Gavlin 1ecdc83a33 Implement more precise delete-before-replace semantics. (#2369)
This implements the new algorithm for deciding which resources must be
deleted due to a delete-before-replace operation.

We need to compute the set of resources that may be replaced by a
change to the resource under consideration. We do this by taking the
complete set of transitive dependents on the resource under
consideration and removing any resources that would not be replaced by
changes to their dependencies. We determine whether or not a resource
may be replaced by substituting unknowns for input properties that may
change due to deletion of the resources their value depends on and
calling the resource provider's Diff method.

This is perhaps clearer when described by example. Consider the
following dependency graph:

  A
__|__
B   C
|  _|_
D  E F

In this graph, all of B, C, D, E, and F transitively depend on A. It may
be the case, however, that changes to the specific properties of any of
those resources R that would occur if a resource on the path to A were
deleted and recreated may not cause R to be replaced. For example, the
edge from B to A may be a simple dependsOn edge such that a change to
B does not actually influence any of B's input properties. In that case,
neither B nor D would need to be deleted before A could be deleted.

In order to make the above algorithm a reality, the resource monitor
interface has been updated to include a map that associates an input
property key with the list of resources that input property depends on.
Older clients of the resource monitor will leave this map empty, in
which case all input properties will be treated as depending on all
dependencies of the resource. This is probably overly conservative, but
it is less conservative than what we currently implement, and is
certainly correct.
2019-01-28 09:46:30 -08:00

327 lines
11 KiB
Go

// 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 engine
import (
"sync"
"time"
"github.com/blang/semver"
"github.com/pulumi/pulumi/pkg/diag"
"github.com/pulumi/pulumi/pkg/resource"
"github.com/pulumi/pulumi/pkg/resource/deploy"
"github.com/pulumi/pulumi/pkg/resource/plugin"
"github.com/pulumi/pulumi/pkg/tokens"
"github.com/pulumi/pulumi/pkg/util/contract"
"github.com/pulumi/pulumi/pkg/util/logging"
"github.com/pulumi/pulumi/pkg/workspace"
)
// UpdateOptions contains all the settings for customizing how an update (deploy, preview, or destroy) is performed.
//
// This structre is embedded in another which uses some of the unexported fields, which trips up the `structcheck`
// linter.
// nolint: structcheck
type UpdateOptions struct {
// an optional set of analyzers to run as part of this deployment.
Analyzers []string
// the degree of parallelism for resource operations (<=1 for serial).
Parallel int
// true if debugging output it enabled
Debug bool
// true if the plan should refresh before executing.
Refresh bool
// true if we should report events for steps that involve default providers.
reportDefaultProviderSteps bool
// the plugin host to use for this update
host plugin.Host
}
// ResourceChanges contains the aggregate resource changes by operation type.
type ResourceChanges map[deploy.StepOp]int
// HasChanges returns true if there are any non-same changes in the resulting summary.
func (changes ResourceChanges) HasChanges() bool {
var c int
for op, count := range changes {
if op != deploy.OpSame {
c += count
}
}
return c > 0
}
func Update(u UpdateInfo, ctx *Context, opts UpdateOptions, dryRun bool) (ResourceChanges, error) {
contract.Require(u != nil, "update")
contract.Require(ctx != nil, "ctx")
defer func() { ctx.Events <- cancelEvent() }()
info, err := newPlanContext(u, "update", ctx.ParentSpan)
if err != nil {
return nil, err
}
defer info.Close()
emitter, err := makeEventEmitter(ctx.Events, u)
if err != nil {
return nil, err
}
return update(ctx, info, planOptions{
UpdateOptions: opts,
SourceFunc: newUpdateSource,
Events: emitter,
Diag: newEventSink(emitter, false),
StatusDiag: newEventSink(emitter, true),
}, dryRun)
}
func newUpdateSource(
opts planOptions, proj *workspace.Project, pwd, main string,
target *deploy.Target, plugctx *plugin.Context, dryRun bool) (deploy.Source, error) {
// Figure out which plugins to load by inspecting the program contents.
plugins, err := plugctx.Host.GetRequiredPlugins(plugin.ProgInfo{
Proj: proj,
Pwd: pwd,
Program: main,
}, plugin.AllPlugins)
if err != nil {
return nil, err
}
// Now ensure that we have loaded up any plugins that the program will need in advance.
const kinds = plugin.AnalyzerPlugins | plugin.LanguagePlugins
if err = plugctx.Host.EnsurePlugins(plugins, kinds); err != nil {
return nil, err
}
// Collect the version information for default providers.
defaultProviderVersions := make(map[tokens.Package]*semver.Version)
for _, p := range plugins {
if p.Kind != workspace.ResourcePlugin {
continue
}
defaultProviderVersions[tokens.Package(p.Name)] = p.Version
}
// If that succeeded, create a new source that will perform interpretation of the compiled program.
// TODO[pulumi/pulumi#88]: we are passing `nil` as the arguments map; we need to allow a way to pass these.
return deploy.NewEvalSource(plugctx, &deploy.EvalRunInfo{
Proj: proj,
Pwd: pwd,
Program: main,
Target: target,
}, defaultProviderVersions, dryRun), nil
}
func update(ctx *Context, info *planContext, opts planOptions, dryRun bool) (ResourceChanges, error) {
result, err := plan(ctx, info, opts, dryRun)
if err != nil {
return nil, err
}
var resourceChanges ResourceChanges
if result != nil {
defer contract.IgnoreClose(result)
// Make the current working directory the same as the program's, and restore it upon exit.
done, chErr := result.Chdir()
if chErr != nil {
return nil, chErr
}
defer done()
if dryRun {
// If a dry run, just print the plan, don't actually carry out the deployment.
resourceChanges, err = printPlan(ctx, result, dryRun)
} else {
// Otherwise, we will actually deploy the latest bits.
opts.Events.preludeEvent(dryRun, result.Ctx.Update.GetTarget().Config)
// Walk the plan, reporting progress and executing the actual operations as we go.
start := time.Now()
actions := newUpdateActions(ctx, info.Update, opts)
err = result.Walk(ctx, actions, false)
resourceChanges = ResourceChanges(actions.Ops)
if len(resourceChanges) != 0 {
// Print out the total number of steps performed (and their kinds), the duration, and any summary info.
opts.Events.updateSummaryEvent(actions.MaybeCorrupt, time.Since(start), resourceChanges)
}
}
}
return resourceChanges, err
}
// pluginActions listens for plugin events and persists the set of loaded plugins
// to the snapshot.
type pluginActions struct {
Context *Context
}
func (p *pluginActions) OnPluginLoad(loadedPlug workspace.PluginInfo) error {
return p.Context.SnapshotManager.RecordPlugin(loadedPlug)
}
// updateActions pretty-prints the plan application process as it goes.
type updateActions struct {
Context *Context
Steps int
Ops map[deploy.StepOp]int
Seen map[resource.URN]deploy.Step
MapLock sync.Mutex
MaybeCorrupt bool
Update UpdateInfo
Opts planOptions
}
func newUpdateActions(context *Context, u UpdateInfo, opts planOptions) *updateActions {
return &updateActions{
Context: context,
Ops: make(map[deploy.StepOp]int),
Seen: make(map[resource.URN]deploy.Step),
Update: u,
Opts: opts,
}
}
func (acts *updateActions) OnResourceStepPre(step deploy.Step) (interface{}, error) {
// Ensure we've marked this step as observed.
acts.MapLock.Lock()
acts.Seen[step.URN()] = step
acts.MapLock.Unlock()
// Skip reporting if necessary.
if shouldReportStep(step, acts.Opts) {
acts.Opts.Events.resourcePreEvent(step, false /*planning*/, acts.Opts.Debug)
}
// Inform the snapshot service that we are about to perform a step.
return acts.Context.SnapshotManager.BeginMutation(step)
}
func (acts *updateActions) OnResourceStepPost(
ctx interface{}, step deploy.Step,
status resource.Status, err error) error {
acts.MapLock.Lock()
assertSeen(acts.Seen, step)
acts.MapLock.Unlock()
// If we've already been terminated, exit without writing the checkpoint. We explicitly want to leave the
// checkpoint in an inconsistent state in this event.
if acts.Context.Cancel.TerminateErr() != nil {
return nil
}
reportStep := shouldReportStep(step, acts.Opts)
// Report the result of the step.
if err != nil {
if status == resource.StatusUnknown {
acts.MaybeCorrupt = true
}
errorURN := resource.URN("")
if reportStep {
errorURN = step.URN()
}
// Issue a true, bonafide error.
acts.Opts.Diag.Errorf(diag.GetPlanApplyFailedError(errorURN), err)
if reportStep {
acts.Opts.Events.resourceOperationFailedEvent(step, status, acts.Steps, acts.Opts.Debug)
}
} else if reportStep {
op, record := step.Op(), step.Logical()
if acts.Opts.isRefresh && op == deploy.OpRefresh {
// Refreshes are handled specially.
op, record = step.(*deploy.RefreshStep).ResultOp(), true
}
if step.Op() == deploy.OpRead {
record = ShouldRecordReadStep(step)
}
if record {
// Increment the counters.
acts.MapLock.Lock()
acts.Steps++
acts.Ops[op]++
acts.MapLock.Unlock()
}
// Also show outputs here for custom resources, since there might be some from the initial registration. We do
// not show outputs for component resources at this point: any that exist must be from a previous execution of
// the Pulumi program, as component resources only report outputs via calls to RegisterResourceOutputs.
if step.Res().Custom || acts.Opts.Refresh && step.Op() == deploy.OpRefresh {
acts.Opts.Events.resourceOutputsEvent(op, step, false /*planning*/, acts.Opts.Debug)
}
}
// See pulumi/pulumi#2011 for details. Terraform always returns the existing state with the diff applied to it in
// the event of an update failure. It's appropriate that we save this new state in the output of the resource, but
// it is not appropriate to save the inputs, because the resource that exists was not created or updated
// successfully with those inputs.
//
// If we were doing an update and got a `StatusPartialFailure`, the resource that ultimately gets persisted in the
// snapshot should be old inputs and new outputs. We accomplish that here by clobbering the new resource's inputs
// with the old inputs.
//
// This is a little kludgy given that these resources are global state. However, given the way that we have
// implemented the snapshot manager and engine today, it's the easiest way to accomplish what we are trying to do.
if status == resource.StatusPartialFailure && step.Op() == deploy.OpUpdate {
logging.V(7).Infof(
"OnResourceStepPost(%s): Step is partially-failed update, saving old inputs instead of new inputs",
step.URN())
new := step.New()
old := step.Old()
contract.Assert(new != nil)
contract.Assert(old != nil)
new.Inputs = make(resource.PropertyMap)
for key, value := range old.Inputs {
new.Inputs[key] = value
}
}
// Write out the current snapshot. Note that even if a failure has occurred, we should still have a
// safe checkpoint. Note that any error that occurs when writing the checkpoint trumps the error
// reported above.
return ctx.(SnapshotMutation).End(step, err == nil || status == resource.StatusPartialFailure)
}
func (acts *updateActions) OnResourceOutputs(step deploy.Step) error {
acts.MapLock.Lock()
assertSeen(acts.Seen, step)
acts.MapLock.Unlock()
// Skip reporting if necessary.
if shouldReportStep(step, acts.Opts) {
acts.Opts.Events.resourceOutputsEvent(step.Op(), step, false /*planning*/, acts.Opts.Debug)
}
// There's a chance there are new outputs that weren't written out last time.
// We need to perform another snapshot write to ensure they get written out.
return acts.Context.SnapshotManager.RegisterResourceOutputs(step)
}