Some YAML parsers don't correctly handle Byte-order marks,
so automatically strip it off during load.
Related to #423
Co-authored-by: Justin Van Patten <jvp@justinvp.com>
* [sdk/go] Cache loaded configuration files
Previously, the CLI did not cache configuration files, which
required a read from disk + unmarshalling + validation each
time a consumer needed to read one of these configurations.
This change introduces global caches for each type of Pulumi
configuration file (Project, ProjectStack, PolicyPackProject, and
PluginProject). The configuration is cached after the first request
and the cached value will be used for any subsequent operations.
Important note: The global configurations are not concurrency safe,
but this same problem exists using the previous method of
reading/writing config files on disk. Synchronization
will be added in a follow up change to allow for concurrency safe config
operations.
When using the filestate backend (local files and cloud buckets) there is no protection to prevent two processes from managing the same stack simultaneously.
This PR creates a locks directory in the management directory that stores lock files for a stack. Each backend implementation gets its own UUID that is joined with the stack name. The feature is currently available behind the `PULUMI_SELF_MANAGED_STATE_LOCKING=1` environment variable flag.
* Fix resource-ref-as-ID marshaling. (#6125)
This reapplies 2f0dba23ab.
* Fix malformed resource value bug
PR #6125 introduced a bug by marshaling resource
ids as PropertyValues, but not handling that case on
the unmarshaling side. The previous code assumed
that the id was a simple string value. This bug prevents
any stack update operations (preview, update, destroy,
refresh). Since this change was already
released, we must now handle both cases in the
unmarshaling code.
* Add resource ref unit tests for the Go SDK. (#6142)
This reapplies 3d505912b8.
Co-authored-by: Pat Gavlin <pat@pulumi.com>
When marshaling a resource reference as its ID (i.e. when
opts.KeepResources is false, as it will be in the case of downlevel SDKs
and resource providers), we must take care to marshal/unmarshal an empty
ID as the unknown property value.
This includes the following changes to the resource ref APIs:
- Bifurcate resource reference creation into two methods: one for
creating references to custom resources and one for creating
references to component resources.
- Store the ID in a resource reference as a PropertyValue s.t. it can be
computed.
- Add a helper method for retrieving the ID as a string + an indicator of
whether or not the reference has an ID.
Fixes#5939.
These changes are a combination of three commits, each of which
contributes to the testing and/or fixing of a problem with marshaling
unknowns in `plugin.provider.Update` when `preview` is true.
## deploytest: add support for gRPC adapters.
These changes add support for communicating with providers using the
gRPC adapters to the deploytest pacakage. This makes it easier to test
the gRPC adapters across typical lifecycle patterns.
Supporting these changes are two additions to the `resource/plugin`
package:
1. A type that bridges between the `plugin.Provider` interface and the
`pulumirpc.ResourceProviderServer`
2. A function to create a `plugin.Provider` given a
`pulumirpc.ResourceProviderClient`
The deploytest package uses these to wrap an in-process
`plugin.Provider` in a gRPC interface and connect to it without using
the default plugin host, respectively.
## pulumi_test: test provider preview over gRPC.
Add a test that runs the provider preview lifecycle, but using a
provider that communicates over gRPC.
## gRPC bridge: fix unknowns in `Update` previews
Set the `KeepUnknowns` and `RejectUnknowns` bits in the `MarshalOptions`
used when unmarshaling update results to preserve unknowns during a
preview and reject them otherwise.
These changes also set the `RejectUnknowns` bit in the `MarshalOptions`
used by `Create` if `preview` is false, and fix a bug in the array
unmarshaler that could cause out-of-bounds accesses.
Fixes https://github.com/pulumi/pulumi/issues/6004.
* Do not read TGZs into memory.
This runs a serious risk of exhausting the memory on lower-end machines
(e.g. certain CI VMs), especially given the potential size of some
plugins.
* CHANGELOG
* fixes
- Add component ref coverage to the existing test
- Add coverage for a downlevel SDK communicating with an engine that
supports resource refs
- Add coverage for a downlevel engine communicating with an SDK that
supports resource refs
As part of improving coverage, these changes add a knob to explicitly
disable resource refs in the engine without the use of the environment
variable. The environment variable is now only read by the CLI, and has
been restored to its prior polarity (i.e. `PULUMI_ENABLE_RESOURCE_REFERENCES`).
* Enable resource reference feature by default
Unless the PULUMI_DISABLE_RESOURCE_REFERENCES flag
is explicitly set to a truthy value, the resource reference feature is now
enabled by default.
* Set AcceptResources in the language SDKs
This can be disabled by setting the `PULUMI_DISABLE_RESOURCE_REFERENCES` environment variable to a truthy value.
Co-authored-by: Justin Van Patten <jvp@justinvp.com>
This re-applies the fix in 5857 to make credentials.json writes concurrency safe.
The original fix used `path.Dir` instead of `filepath.Dir` - which led to not placing the temp file in the same folder (and drive) as the renamed file target. This led to errors on Windows environments where the working directory was on a different drive than the `~/.pulumi` directory. The change to use `filepath.Dir` instead ensures that even on Windows, the true directory containing the credentials file is used for the temp file as well.
Fixes#3877.
* Properly resize arrays when adding
The current logic attempts to update the array but, because
append may need to allocate a new array with adequate space,
the code can currently leave dest referring to the old,
under-sized array. The solution is to use the set(dest)
logic that already exists and is used for the IsNull case.
Added a test case that would fail before this fix and now passes.
This fixespulumi/pulumi#5871.
* Add CHANGELOG entry
Implement GetRequiredPlugins for Python, which determines the plugins
required by the program.
Also, if the `virtualenv` runtime option is set, and the specified
virtual directory is missing or empty, automatically create it and
install dependencies into it.
Running `pulumi` operations in parallel could occasionally result in truncating the `~/.pulumi/credentials.json` file and reading that truncated file from another process before the content could be written.
Instead, use `os.Rename` to atomically replace the file contents.
Concurrent `pulumi` operations could still compete for who gets to write the file first, and could lead to surprising results in some extreme cases. But we should not see the corrupted file contents any longer.
Fixes#3877.
- Differentiate between resource references that have no ID (i.e. because
the referenced resource is not a CustomResource) and resource references
that have IDs that are not known. This is necessary for proper
backwards-compatible serialization of resource references.
- Fix the key that stores a resource reference's package version in the
.NET, NodeJS, and Python SDKs.
- Ensure that the resource monitor's marshalling/unmarshalling of inputs
and outputs to/from calls to `Construct` retain resource references as
appropriate.
- Fix serialization behavior for resources -> resource references in the
Go SDK: if a resource's ID is unknown, it should still be serialized
as a resource reference, albeit a reference with an unknown ID.
The PULUMI_BACKEND_URL env var allows specifying the backend to use instead of deferring to the project or the ~/.pulumi/credentials.json file to decide on the "current" backend. This allows for using Pulumi without a dependence on this piece of global filesystem state, so that each `pulumi` invocation can control the exact backend it want's to operate on, without having to do stateful `pulumi login`/`pulumi logout` operations.
This is especially useful for automation scenarios like Automation API generally (and effectively solves https://github.com/pulumi/pulumi/issues/5591), or https://github.com/pulumi/pulumi-kubernetes-operator/issues/83 specifically.
This also makes things like efe7a599e6/dist/actions/entrypoint.sh (L10) less necessary, and possible to accomplish for any containerized `pulumi` execution without the need for this logic to be embedded in bash scripts wrapping the CLI.
Two improvements:
1. Don't display "[resource plugin <foo>] installing" if the plugin is already installed.
2. Close the plugin download progress bar before displaying any subsequent output, and only show output of `npm install` when there is an error.
When installing a plugin, previous versions of Pulumi extracted the
plugin tarball to a temp directory and then renamed the temp directory
to the final plugin directory. This was done to prevent concurrent
installs: if a process fails to rename the temp dir because the final
dir already exists, it means another process already installed the
plugin. Unfortunately, on Windows the rename operation often fails due
to aggressive virus scanners opening files in the temp dir.
In order to provide reliable plugin installs on Windows, we now extract
the tarball directly into the final directory, and use file locks to
prevent concurrent installs from toppling over one another.
During install, a lock file is created in the plugin cache directory
with the same name as the plugin's final directory but suffixed with
`.lock`. The process that obtains the lock is responsible for extracting
the tarball. Before it does that, it cleans up any previous temp
directories of failed installs of previous versions of Pulumi. Then it
creates an empty `.partial` file next to the `.lock` file. The
`.partial` file indicates an installation is in-progress. The `.partial`
file is deleted when installation is complete, indicating the plugin was
successfully installed. If a failure occurs during installation, the
`.partial` file will remain indicating the plugin wasn't fully
installed. The next time the plugin is installed, the old installation
directory will be removed and replaced with a fresh install.
This is the same approach Go uses for installing modules in its
module cache.
Just what it says on the tin. This is implemented by changing the
`GetPackageConfig` method of `ConfigSource` to return a `PropertyMap`
and ensuring that any secret config is represented by a `Secret`.
This is necessary due to the way we've factored the libraries imported
by users into modules. The primary alternative is to ensure that each
child module imports the root module for a package and registers itself
with that package where necessary to prevent circular dependencies. This
simplifies the core SDKs slightly at the cost of greater complications
in the generated SDKs; the approach taken by these changes seems like a
more maintainable option.
Contributes to #2430.
Co-authored-by: Justin Van Patten <jvp@justinvp.com>
Resources are serialized as their URN, ID, and package version. Each
Pulumi package is expected to register itself with the SDK. The package
will be invoked to construct appropriate instances of rehydrated
resources. Packages are distinguished by their name and their version.
This is the foundation of cross-process resources.
Related to #2430.
Co-authored-by: Mikhail Shilkov <github@mikhail.io>
Co-authored-by: Luke Hoban <luke@pulumi.com>
Co-authored-by: Levi Blackstone <levi@pulumi.com>
These changes add support for provider-side previews of create and
update operations, which allows resource providers to supply output
property values for resources that are being created or updated during a
preview.
If a plugin supports provider-side preview, its create/update methods
will be invoked during previews with the `preview` property set to true.
It is the responsibility of the provider to fill in any output
properties that are known before returning. It is a best practice for
providers to only fill in property values that are guaranteed to be
identical if the preview were instead an update (i.e. only those output
properties whose values can be conclusively determined without
actually performing the create/update operation should be populated).
Providers that support previews must accept unknown values in their
create and update methods.
If a plugin does not support provider-side preview, the inputs to a
create or update operation will be propagated to the outputs as they are
today.
Fixes#4992.
The logic for validating prompted values in 'new' wasn't quite right,
leading to the possibility of creating Pulumi.yaml files with blank
project names.
This manifests in various ways and I've hit it a number of times
over the past few months because of the way we handle project/stack
name conflicts in 'new' -- which itself is a bit annoying too:
https://github.com/pulumi/pulumi/blob/master/pkg/cmd/pulumi/new.go#L206-L207
Because we substitue a default value of "", and because the prompting
logic assumed default values are always valid, we would skip validation
and therefore accept a blank Pulumi.yaml file.
This generates an invalid project which causes errors elsewhere, such as
error: failed to load Pulumi project located at ".../Pulumi.yaml":
project is missing a 'name' attribute
I hit this all the time with our getting started guide because I've
gone through it so many times and have leftover stacks from prior
run-throughs. I wouldn't be surprised if a lot of people hit this.
The solution here validates all values, including the default.
Note also that we failed to validate the value used by 'new --yes'
which meant you could bypass all validation by passing --yes, leading
to similar outcomes.
I've added a couple new tests for these cases. There is a risk we
depend on illegal default values somewhere which will now be rejected,
but that would seem strange, and assuming the tests pass, I would
assume that's not true. Let me know if that's wrong.
Fixespulumi/pulumi#3255.
* Revise host mode.
The current implementation of host mode uses a `pulumi host` command and
an ad-hoc communication protocol between the engine and client to
connect a language host after the host has begun listening. The most
significant disadvantages of this approach are the communication
protocol (which currently requires the use of stdout), the host-specific
command, and the difficulty of accommodating the typical program-bound
lifetime for an update.
These changes reimplement host mode by adding engine support for
connecting to an existing language runtime service rather than launching
a plugin. This capability is provided via an engine-specific language
runtime, `client`, which accepts the address of the existing languge
runtime service as a runtime option. The CLI exposes this runtime via
the `--client` flag to the `up` and `preview` commands, which similarly
accepts the address of an existing language runtime service as an
argument. These changes also adjust the automation API to consume the
new host mode implementation.
