These changes restore a more-correct version of the behavior that was
disabled with #3014. The original implementation of this behavior was
done in the SDKs, which do not have access to the complete inputs for a
resource (in particular, default values filled in by the provider during
`Check` are not exposed to the SDK). This lack of information meant that
the resolved output values could disagree with the typings present in
a provider SDK. Exacerbating this problem was the fact that unknown
values were dropped entirely, causing `undefined` values to appear in
unexpected places.
By doing this in the engine and allowing unknown values to be
represented in a first-class manner in the SDK, we can attack both of
these issues.
Although this behavior is not _strictly_ consistent with respect to the
resource model--in an update, a resource's output properties will come
from its provider and may differ from its input properties--this
behavior was present in the product for a fairly long time without
significant issues. In the future, we may be able to improve the
accuracy of resource outputs during a preview by allowing the provider
to dry-run CRUD operations and return partially-known values where
possible.
These changes also introduce new APIs in the Node and Python SDKs
that work with unknown values in a first-class fashion:
- A new parameter to the `apply` function that indicates that the
callback should be run even if the result of the apply contains
unknown values
- `containsUnknowns` and `isUnknown`, which return true if a value
either contains nested unknown values or is exactly an unknown value
- The `Unknown` type, which represents unknown values
The primary use case for these APIs is to allow nested, properties with
known values to be accessed via the lifted property accessor even when
the containing property is not fully know. A common example of this
pattern is the `metadata.name` property of a Kubernetes `Namespace`
object: while other properties of the `metadata` bag may be unknown,
`name` is often known. These APIs allow `ns.metadata.name` to return a
known value in this case.
In order to avoid exposing downlevel SDKs to unknown values--a change
which could break user code by exposing it to unexpected values--a
language SDK must indicate whether or not it supports first-class
unknown values as part of each `RegisterResourceRequest`.
These changes also allow us to avoid breaking user code with the new
behavior introduced by the prior commit.
Fixes#3190.
With these changes, a user may explicitly set `deleteBeforeReplace` to
`false` in order to disable DBR behavior for a particular resource. This
is the SDK + CLI escape hatch for cases where the changes in
https://github.com/pulumi/pulumi-terraform/pull/465 cause undesirable
behavior.
* Allow resource IDs to change on reresh steps
This is a requirement for us to be able to move forward with
versions of the Terraform Azurerm provider. In v1.32.1, there was
a state migration that changed the ID format of the azure table
storage resource
We used to have a check in place for old ID being equal to new ID.
This has been changed now and we allow the change of ID to happen
in the RefreshStep
* Update pkg/resource/deploy/step.go
Co-Authored-By: Pat Gavlin <pat@pulumi.com>
These changes add support for passing `ignoreChanges` paths to resource
providers. This is intended to accommodate providers that perform diffs
between resource inputs and resource state (e.g. all Terraform-based
providers, the k8s provider when using API server dry-runs). These paths
are specified using the same syntax as the paths used in detailed diffs.
In addition to passing these paths to providers, the existing support
for `ignoreChanges` in inputs has been extended to accept paths rather
than top-level keys. It is an error to specify a path that is missing
one or more component in the old or new inputs.
Fixes#2936, #2663.
Most of these options are typically left unset. In order to make it
easier to update the lifecycle test when adding new options, collect
them in a bag s.t. most callsites can go without being updated.
If we encounter a provider with old inputs but no old outputs when reading
a checkpoint file, use the old inputs as the old outputs. This handles the
scenario where the CLI is being upgraded from a version that did not
reflect provider inputs to provider outputs, and a provider is being
upgraded from a version that did not implement `DiffConfig` to a version
that does.
Fixes https://github.com/pulumi/pulumi-kubernetes/issues/645.
* Plumbing the custom timeouts from the engine to the providers
* Plumbing the CustomTimeouts through to the engine and adding test to show this
* Change the provider proto to include individual timeouts
* Plumbing the CustomTimeouts from the engine through to the Provider RPC interface
* Change how the CustomTimeouts are sent across RPC
These errors were spotted in testing. We can now see that the timeout
information is arriving in the RegisterResourceRequest
```
req=&pulumirpc.RegisterResourceRequest{
Type: "aws:s3/bucket:Bucket",
Name: "my-bucket",
Parent: "urn:pulumi:dev::aws-vpc::pulumi:pulumi:Stack::aws-vpc-dev",
Custom: true,
Object: &structpb.Struct{},
Protect: false,
Dependencies: nil,
Provider: "",
PropertyDependencies: {},
DeleteBeforeReplace: false,
Version: "",
IgnoreChanges: nil,
AcceptSecrets: true,
AdditionalSecretOutputs: nil,
Aliases: nil,
CustomTimeouts: &pulumirpc.RegisterResourceRequest_CustomTimeouts{
Create: 300,
Update: 400,
Delete: 500,
XXX_NoUnkeyedLiteral: struct {}{},
XXX_unrecognized: nil,
XXX_sizecache: 0,
},
XXX_NoUnkeyedLiteral: struct {}{},
XXX_unrecognized: nil,
XXX_sizecache: 0,
}
```
* Changing the design to use strings
* CHANGELOG entry to include the CustomTimeouts work
* Changing custom timeouts to be passed around the engine as converted value
We don't want to pass around strings - the user can provide it but we want
to make the engine aware of the timeout in seconds as a float64
A resource can be imported by setting the `import` property in the
resource options bag when instantiating a resource. In order to
successfully import a resource, its desired configuration (i.e. its
inputs) must not differ from its actual configuration (i.e. its state)
as calculated by the resource's provider.
There are a few interesting state transitions hiding here when importing
a resource:
1. No prior resource exists in the checkpoint file. In this case, the
resource is simply imported.
2. An external resource exists in the checkpoint file. In this case, the
resource is imported and the old external state is discarded.
3. A non-external resource exists in the checkpoint file and its ID is
different from the ID to import. In this case, the new resource is
imported and the old resource is deleted.
4. A non-external resource exists in the checkpoint file, but the ID is
the same as the ID to import. In this case, the import ID is ignored
and the resource is treated as it would be in all cases except for
changes that would replace the resource. In that case, the step
generator issues an error that indicates that the import ID should be
removed: were we to move forward with the replace, the new state of
the stack would fall under case (3), which is almost certainly not
what the user intends.
Fixes#1662.
Thse changes make a subtle but critical adjustment to the process the
Pulumi engine uses to determine whether or not a difference exists
between a resource's actual and desired states, and adjusts the way this
difference is calculated and displayed accordingly.
Today, the Pulumi engine get the first chance to decide whether or not
there is a difference between a resource's actual and desired states. It
does this by comparing the current set of inputs for a resource (i.e.
the inputs from the running Pulumi program) with the last set of inputs
used to update the resource. If there is no difference between the old
and new inputs, the engine decides that no change is necessary without
consulting the resource's provider. Only if there are changes does the
engine consult the resource's provider for more information about the
difference. This can be problematic for a number of reasons:
- Not all providers do input-input comparison; some do input-state
comparison
- Not all providers are able to update the last deployed set of inputs
when performing a refresh
- Some providers--either intentionally or due to bugs--may see changes
in resources whose inputs have not changed
All of these situations are confusing at the very least, and the first
is problematic with respect to correctness. Furthermore, the display
code only renders diffs it observes rather than rendering the diffs
observed by the provider, which can obscure the actual changes detected
at runtime.
These changes address both of these issues:
- Rather than comparing the current inputs against the last inputs
before calling a resource provider's Diff function, the engine calls
the Diff function in all cases.
- Providers may now return a list of properties that differ between the
requested and actual state and the way in which they differ. This
information will then be used by the CLI to render the diff
appropriately. A provider may also indicate that a particular diff is
between old and new inputs rather than old state and new inputs.
Fixes#2453.
This commit touches an intersection of a few different provider-oriented
features that combined to cause a particularly severe bug that made it
impossible for users to upgrade provider versions without seeing
replacements with their resources.
For some context, Pulumi models all providers as resources and places
them in the snapshot like any other resource. Every resource has a
reference to the provider that created it. If a Pulumi program does not
specify a particular provider to use when performing a resource
operation, the Pulumi engine injects one automatically; these are called
"default providers" and are the most common ways that users end up with
providers in their snapshot. Default providers can be identified by
their name, which is always prefixed with "default".
Recently, in an effort to make the Pulumi engine more flexible with
provider versions, it was made possible for the engine to have multiple
default providers active for a provider of a particular type, which was
previously not possible. Because a provider is identified as a tuple of
package name and version, it was difficult to find a name for these
duplicate default providers that did not cause additional problems. The
provider versioning PR gave these default providers a name that was
derived from the version of the package. This proved to be a problem,
because when users upgraded from one version of a package to another,
this changed the name of their default provider which in turn caused all
of their resources created using that provider (read: everything) to be
replaced.
To combat this, this PR introduces a rule that the engine will apply
when diffing a resource to determine whether or not it needs to be
replaced: "If a resource's provider changes, and both old and new
providers are default providers whose properties do not require
replacement, proceed as if there were no diff." This allows the engine
to gracefully recognize and recover when a resource's default provider changes
names, as long as the provider's config has not changed.
Adds a new resource option `aliases` which can be used to rename a resource. When making a breaking change to the name or type of a resource or component, the old name can be added to the list of `aliases` for a resource to ensure that existing resources will be migrated to the new name instead of being deleted and replaced with the new named resource.
There are two key places this change is implemented.
The first is the step generator in the engine. When computing whether there is an old version of a registered resource, we now take into account the aliases specified on the registered resource. That is, we first look up the resource by its new URN in the old state, and then by any aliases provided (in order). This can allow the resource to be matched as a (potential) update to an existing resource with a different URN.
The second is the core `Resource` constructor in the JavaScript (and soon Python) SDKs. This change ensures that when a parent resource is aliased, that all children implicitly inherit corresponding aliases. It is similar to how many other resource options are "inherited" implicitly from the parent.
Four specific scenarios are explicitly tested as part of this PR:
1. Renaming a resource
2. Adopting a resource into a component (as the owner of both component and consumption codebases)
3. Renaming a component instance (as the owner of the consumption codebase without changes to the component)
4. Changing the type of a component (as the owner of the component codebase without changes to the consumption codebase)
4. Combining (1) and (3) to make both changes to a resource at the same time
For provider plugins, the gRPC interfaces expect that a URN would be
included as part of the DiffConfig/CheckConfig request, which means we
need to flow this value into our Provider interface.
This change does that.
We have many cases where we want to do the following:
deployment -> snapshot -> process snapshot -> deployment
We now retain information in the snapshot about the secrets manager
that was used to construct it, so in these round trip cases, we can
re-use the existing manager.
Fixes#2277.
Adds a new ignoreChanges resource option that allows specifying a list of property names whose values will be ignored during updates. The property values will be used for Create, but will be ignored for purposes of updates, and as a result also cannot trigger replacements.
This is a feature of the Pulumi engine, not of the resource providers, so no new logic is needed in providers to support this feature. Instead, the engine simply replaces the values of input properties in the goal state with old inputs for properties marked as ignoreChanges.
Currently, only top level properties may be specified in ignoreChanges. In the future, this could be extended to support paths to nested properties (including into array elements) with a JSONPath/JMESPath syntax.
* Load specific provider versions if requested
As part of pulumi/pulumi#2389, we need the ability for language hosts to
tell the engine that a particular resource registration, read, or invoke
needs to use a particular version of a resource provider. This was not
previously possible before; the engine prior to this commit loaded
plugins from a default provider map, which was inferred for every
resource provider based on the contents of a user's package.json, and
was itself prone to bugs.
This PR adds the engine support needed for language hosts to request a
particular version of a provider. If this occurs, the source evaluator
specifically records the intent to load a provider with a given version
and produces a "default" provider registration that requests exactly
that version. This allows the source evaluator to produce multiple
default providers for a signle package, which was previously not
possible.
This is accomplished by having the source evaluator deal in the
"ProviderRequest" type, which is a tuple of version and package. A
request to load a provider whose version matches the package of a
previously loaded provider will re-use the existing default provider. If
the version was not previously loaded, a new default provider is
injected.
* CR Feedback: raise error if semver is invalid
* CR: call String() if you want a hash key
* Update pkg/resource/deploy/providers/provider.go
Co-Authored-By: swgillespie <sean@pulumi.com>
Fixes#2633.
Currently when a user runs `refresh` and a resource is in a state of
error, the `refresh` will fail and the resource state will not be
persisted. This can make it vastly harder to incrementally fix
infrastructure. The issue mentioned above explains more of the
historical context, as well as some specific failure modes.
This commit resolves this issue by causing refresh to *not* report an
error in this case, and instead to simply log a warning that the
`refresh` has recognized that the resource is in an unhealthy state
during state sync.
These changes take advantage of the newly-added support for returning
inputs from Read to update a resource's inputs as part of a refresh.
As a consequence, the Pulumi engine will now properly detect drift
between the actual state of a resource and the desired state described
by the program and generate appropriate update or replace steps.
As part of these changes, a resource's old inputs are now passed to the
provider when performing a refresh. The provider can take advantage of
this to maintain the accuracy of any additional data or metadata in the
resource's inputs that may need to be updated during the refresh.
This is required for the complete implementation of
https://github.com/pulumi/pulumi-terraform/pull/349. Without access to
the old inputs for a resource, TF-based providers would lose all
information about default population during a refresh.
These changes add a new flag to the various `ResourceOptions` types that
indicates that a resource should be deleted before it is replaced, even
if the provider does not require this behavior. The usual
delete-before-replace cascade semantics apply.
Fixes#1620.
In general, a "delete" in Pulumi is destroying an actual physical
resource. In the case of a read resource, however, the delete is
merely removing the resource from the stack; the physical resource
is not affected. These changes attempt to clarify this situation by
using the term "discard" rather than "delete".
Fixes#2015.
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.
This test had been intermittently failing due to a race condition. Its
implementation of `plugin.Provider.Read` was intended to ensure that
the cancellation of a refresh operation occurred. As written, it was
only able to ensure that the cancellation was requested.
These changes ensure that cancellation has been acknowledged by the engine by
implementing providing an implementation for `plugin.Provider.Cancel`
that closes a channel on which the implementation of `Read` waits.
This ensures that the gRPC server is properly shut down. This fixes an
issue in which a resource plugin that is still configuring could report
log messages to the plugin host, which would in turn attempt to send
diagnostic packets over a closed channel, causing a panic.
Fixes#2170.
After #2088, we began calling `Diff` on providers that are not configured
due to unknown configuration values. This hit an assertion intended to
detect exactly this scenario, which was previously unexpected.
These changes adjust `Diff` to indicate that a Diff is unavailable and
return an error message that describes why. The step generator then
interprets the diff as indicating a normal update and issues the error
message to the diagnostic stream.
Fixes#2223.
These changes add a new resource to the Pulumi SDK,
`pulumi.StackReference`, that represents a reference to another stack.
This resource has an output property, `outputs`, that contains the
complete set of outputs for the referenced stack. The Pulumi account
performing the deployment that creates a `StackReference` must have
access to the referenced stack or the call will fail.
This resource is implemented by a builtin provider managed by the engine.
This provider will be used for any custom resources and invokes inside
the `pulumi:pulumi` module. Currently this provider supports only the
`pulumi:pulumi:StackReference` resource.
Fixes#109.
We run the same suite of changes that we did on gometalinter. This
ended up catching a few new issues, some of which were addressed and
some of which were baselined.
In preparation for some workspace restructuring, I decided to scratch a
few itches of my own in the code:
* Change project's RuntimeInfo field to just Runtime, to match the
serialized name in JSON/YAML.
* Eliminate the no-longer-used Context and NoDefaultIgnores fields on
project, and all of the associated legacy PPC-related code.
* Eliminate the no-longer-used IgnoreFile constant.
* Remove a bunch of "// nolint: lll" annotations, and simply format
the structures with comments on dedicated lines, to avoid overly
lengthy lines and lint suppressions.
* Mark Dependencies and InitErrors as `omitempty` in the JSON
serialization directives for CheckpointV2 files. This was done for
the YAML directives, but (presumably accidentally) omitted for JSON.
Whenever an update fails partially, it gives the engine back some state
bag of outputs that should be persisted to the snapshot. When saving
this state, we shouldn't save the inputs that triggered the update that
failed, since the resource that exists will never have been updated
successfully with those inputs.
Instead of saving the new inputs on partial failed updates, this commit
saves the old inputs and the new outputs. The new outputs will likely
need to be refreshed to be useful, but the old inputs will be correct
from the perpsective of the Pulumi program that generated the last
successful update.
Fixespulumi/pulumi#2011
The preview will proceed as if the operations had not been issued (i.e.
we will not speculate on a new state for the stack). This is consistent
with our behavior prior to the changes that added pending operations to
the checkpoint.
* Retire pending deletions at start of plan
Instead of letting pending deletions pile up to be retired at the end of
a plan, this commit eagerly disposes of any pending deletions that were
pending at the end of the previous plan. This is a nice usability win
and also reclaims an invariant that at most one resource with a given
URN is live and at most one is pending deletion at any point in time.
* Rebase against master
* Fix a test issue arising from shared snapshots
* CR feedback
* plan -> replacement
* Use ephemeral statuses to communicate deletions
We signal provider cancellation by hangning a goroutine off of the plan
executor's parent context. To ensure clean shutdown, this goroutine also
listens on a channel that closes once the plan has finished executing.
Unfortunately, we were closing this channel too early, and the close was
racing with the cancellation signal. These changes ensure that the
channel closes after the plan has fully completed.
Fixes#1906.
Fixespulumi/pulumi-kubernetes#185.
* Validate type tokens before using them
When registering or reading a resource, we take the type token given to
us from the language host and assume that it's valid, which resulted in
assertion failures in various places in the engine. This commit
validates the format of type tokens given to us from the language host
and issues an appropriate error if it's not valid.
Along the way, this commit also improves the way that fatal exceptions
are rendered in the Node language host.
* Pre-allocate an exception for ReadResource
* Fix integration test
* CR Feedback
This commit is a lower-impact change that fixes the bugs associated with
invalid types on component resources and only checks that a type is
valid on custom resources.
* CR Take 2: Fix up IsProviderType instead of fixing call sites
* Please gometalinter
This commit reverts most of #1853 and replaces it with functionally
identical logic, using the notion of status message-specific sinks.
In other words, where the original commit implemented ephemeral status
messages by adding an `isStatus` parameter to most of the logging
methdos in pulumi/pulumi, this implements ephemeral status messages as a
parallel logging sink, which emits _only_ ephemeral status messages.
The original commit message in that PR was:
> Allow log events to be marked "status" events
>
> This commit will introduce a field, IsStatus to LogRequest. A "status"
> logging event will be displayed in the Info column of the main
> display, but will not be printed out at the end, when resource
> operations complete.
>
> For example, for complex resource initialization, we'd like to display
> a series of intermediate results: [1/4] Service object created, for
> example. We'd like these to appear in the Info column, but not at the
> end, where they are not helpful to the user.
- Create all refresh steps before issuing any. This is important as the
state update loop expects all steps to exist.
- Check for cancellation later in the refresher.
This also fixes races in the SnapshotManager and the test journal that
could cause panics during cancellation.
Replace the Source-based implementation of refresh with a phase that
runs as the first part of plan execution and rewrites the snapshot in-memory.
In order to fit neatly within the existing framework for resource operations,
these changes introduce a new kind of step, RefreshStep, to represent
refreshes. RefreshSteps operate similar to ReadSteps but do not imply that
the resource being read is not managed by Pulumi.
In addition to the refresh reimplementation, these changes incorporate those
from #1394 to run refresh in the integration test framework.
Fixes#1598.
Fixespulumi/pulumi-terraform#165.
Contributes to #1449.
* Show a better error message when decrypting fails
It is most often the case that failing to decrypt a secret implies that
the secret was transferred from one stack to another via copying the
configuration. This commit introduces a better error message for this
case and instructs users to explicitly re-encrypt their encrypted keys
in the context of the new stack.
* Spelling
* CR: Grammar fixes
* Log errors coming from the language host
Similar to pulumi/pulumi#1762, fixespulumi/pulumi#1775. The language
host can fail without issuing any diagnostics and it is very unclear
what happens if the engine does not log the error.
* CR feedback
The plan executor assumed that the step generator was responsible for
logging its own diagnostics, which it sort-of is but also doesn't log a
majority of the diagnositcs that come out of it. This commit logs all
errors coming out of step generation so that we don't unintentionally
drop errors.
* Add a list of in-flight operations to the deployment
This commit augments 'DeploymentV2' with a list of operations that are
currently in flight. This information is used by the engine to keep
track of whether or not a particular deployment is in a valid state.
The SnapshotManager is responsible for inserting and removing operations
from the in-flight operation list. When the engine registers an intent
to perform an operation, SnapshotManager inserts an Operation into this
list and saves it to the snapshot. When an operation completes, the
SnapshotManager removes it from the snapshot. From this, the engine can
infer that if it ever sees a deployment with pending operations, the
Pulumi CLI must have crashed or otherwise abnormally terminated before
seeing whether or not an operation completed successfully.
To remedy this state, this commit also adds code to 'pulumi stack
import' that clears all pending operations from a deployment, as well as
code to plan generation that will reject any deployments that have
pending operations present.
At the CLI level, if we see that we are in a state where pending
operations were in-flight when the engine died, we'll issue a
human-friendly error message that indicates which resources are in a bad
state and how to recover their stack.
* CR: Multi-line string literals, renaming in-flight -> pending
* CR: Add enum to apitype for operation type, also name status -> type for clarity
* Fix the yaml type
* Fix missed renames
* Add implementation for lifecycle_test.go
* Rebase against master
Some time ago, we introduced the concept of the initialization error to
Pulumi (i.e., an error where the resource was successfully created but
failed to fully initialize). This was originally implemented in `Create`
and `Update` methods of the resource provider interface; when we
detected an initialization failure, we'd pack the live version of the
object into the error, and return that to the engine.
Omitted from this initial implementation was a similar semantics for
`Read`. There are many implications of this, but one of them is that a
`pulumi refresh` will erase any initialization errors that had
previously been observed, even if the initialization errors still exist
in the resource.
This commit will introduce the initialization error semantics to `Read`,
fixing this issue.
* Emit reads for external resources when refreshing
Fixespulumi/pulumi#1744. This commit educates the refresh source about
external resources. If a refresh source encounters a resource with the
External bit set, it'll send a Read event to the engine and the engine
will process it accordingly.
* CR: save last event channel instead of last event, style fixes
* Serialize SourceEvents coming from the refresh source
The engine requires that a source event coming from a source be "ready
to execute" at the moment that it is sent to the engine. Since the
refresh source sent all goal states eagerly through its source iterator,
the engine assumed that it was legal to execute them all in parallel and
did so. This is a problem for the snapshot, since the snapshot expects
to be in an order that is a legal topological ordering of the dependency
DAG.
This PR fixes the issue by sending refresh source events one-at-a-time
through the refresh source iterator, only unblocking to send the next
step as soon as the previous step completes.
* Fix deadlock in refresh test
* Fix an issue where the engine "completed" steps too early
By signalling that a step is done before committing the step's results
to the snapshot, the engine was left with a race where dependent
resources could find themselves completely executed and committed before
a resource that they depend on has been committed.
Fixespulumi/pulumi#1726
* Fix an issue with Replace steps at the end of a plan
If the last step that was executed successfully was a Replace, we could
end up in a situation where we unintentionally left the snapshot
invalid.
* Add a test
* CR: pass context.Context as first parameter to Iterate
* CR: null->nil
When calculating deletes, we will only issue a single delete step for a
particular URN. This is incorrect in the presence of pending deletes
that share URNs with a live resource if the pending deletes follow the
live resource in the checkpoint: instead of issuing a delete for
every resource with a particular URN, we will only issue deletes for
the pending deletes.
Before first-class providers, this was mostly benigin: any remaining
resources could be deleted by re-running the destroy. With the
first-class provider changes, however, the provider for the undeleted
resources will be deleted, leaving the checkpoint in an invalid state.
These changes fix this issue by allowing the step generator to issue
multiple deletes for a single URN and add a test for this scenario.
