If a user forgets to specify the source address for a provider, Terraform
will assume they meant a provider in the registry.terraform.io/hashicorp/
namespace. If that ultimately doesn't exist, we'll now try to see if
there's some other provider source address recorded in the registry's
legacy provider lookup table, and suggest it if so.
The error message here is a terse one addressed primarily to folks who are
already somewhat familiar with provider source addresses and how to
specify them. Terraform v0.13 had a more elaborate version of this error
message which directed the user to try the v0.13 automatic upgrade tool,
but we no longer have that available in v0.14 and later so the user must
make the fix themselves.
The temporary directory on some systems (most notably MacOS) contains
symlinks, which would not be recorded by the installer. In order to make
these paths comparable in the tests we need to eval the symlinks in
the paths before giving them to the installer.
Previously we were only verifying locked hashes for local archive zip
files, but if we have non-ziphash hashes available then we can and should
also verify that a local directory matches at least one of them.
This does mean that folks using filesystem mirrors but yet also running
Terraform across multiple platforms will need to take some extra care to
ensure the hashes pass on all relevant platforms, which could mean using
"terraform providers lock" to pre-seed their lock files with hashes across
all platforms, or could mean using the "packed" directory layout for the
filesystem mirror so that Terraform will end up in the install-from-archive
codepath instead of this install-from-directory codepath, and can thus
verify ziphash too.
(There's no additional documentation about the above here because there's
already general information about this in the lock file documentation
due to some similar -- though not identical -- situations with network
mirrors.)
We previously had some tests for some happy paths and a few specific
failures into an empty directory with no existing locks, but we didn't
have tests for the installer respecting existing lock file entries.
This is a start on a more exhaustive set of tests for the installer,
aiming to visit as many of the possible codepaths as we can reasonably
test using this mocking strategy. (Some other codepaths require different
underlying source implementations, etc, so we'll have to visit those in
other tests separately.)
This probably isn't the best UI we could do here, but it's a placeholder
for now just to avoid making it seem like we're ignoring the lock file
and checking for new versions anyway.
This changes the approach used by the provider installer to remember
between runs which selections it has previously made, using the lock file
format implemented in internal/depsfile.
This means that version constraints in the configuration are considered
only for providers we've not seen before or when -upgrade mode is active.
helper/copy CopyDir was used heavily in tests. It differes from
internal/copydir in a few ways, the main one being that it creates the
dst directory while the internal version expected the dst to exist
(there are other differences, which is why I did not just switch tests
to using internal's CopyDir).
I moved the CopyDir func from helper/copy into command_test.go; I could
also have moved it into internal/copy and named it something like
CreateDirAndCopy so if that seems like a better option please let me
know.
helper/copy/CopyFile was used in a couple of spots so I moved it into
internal, at which point I thought it made more sense to rename the
package copy (instead of copydir).
There's also a `go mod tidy` included.
In earlier commits we started to make the installation codepath
context-aware so that it could be canceled in the event of a SIGINT, but
we didn't complete wiring that through the API of the getproviders
package.
Here we make the getproviders.Source interface methods, along with some
other functions that can make network requests, take a context.Context
argument and act appropriately if that context is cancelled.
The main providercache.Installer.EnsureProviderVersions method now also
has some context-awareness so that it can abort its work early if its
context reports any sort of error. That avoids waiting for the process
to wind through all of the remaining iterations of the various loops,
logging each request failure separately, and instead returns just
a single aggregate "canceled" error.
We can then set things up in the "terraform init" and
"terraform providers mirror" commands so that the context will be
cancelled if we get an interrupt signal, allowing provider installation
to abort early while still atomically completing any local-side effects
that may have started.
The logic for what constitutes a valid hash and how different hash schemes
are represented was starting to get sprawled over many different files and
packages.
Consistently with other cases where we've used named types to gather the
definition of a particular string into a single place and have the Go
compiler help us use it properly, this introduces both getproviders.Hash
representing a hash value and getproviders.HashScheme representing the
idea of a particular hash scheme.
Most of this changeset is updating existing uses of primitive strings to
uses of getproviders.Hash. The new type definitions are in
internal/getproviders/hash.go.
For reasons that are unclear, these two tests just started failing on
macOS very recently. The failure looked like:
PackageDir: strings.Join({
"/",
+ "private/",
"var/folders/3h/foobar/T/terraform-test-p",
"rovidercache655312854/registry.terraform.io/hashicorp/null/2.0.0",
"/windows_amd64",
},
Speculating that the macOS temporary directory moved into the /private
directory, I added a couple of EvalSymlinks calls and the tests pass
again.
No other unit tests appear to be affected by this at the moment.
The installFromHTTPURL function downloads a package to a temporary file,
then delegates to installFromLocalArchive to install it. We were
previously not deleting the temporary file afterwards. This commit fixes
that.
When installing a provider which is already cached, we attempt to create
a symlink from the install directory targeting the cache. If symlinking
fails due to missing OS/filesystem support, we instead want to copy the
cached provider.
The fallback code to do this would always fail, due to a missing target
directory. This commit fixes that. I was unable to find a way to add
automated tests around this, but I have manually verified the fix on
Windows 8.1.
At the end of the EnsureProviderVersions process, we generate a lockfile
of the selected and installed provider versions. This includes a hash of
the unpacked provider directory.
When calculating this hash and generating the lockfile, we now also
verify that the provider directory contains a valid executable file. If
not, we return an error for this provider and trigger the installer's
HashPackageFailure event. Note that this event is not yet processed by
terraform init; that comes in the next commit.
Instead of searching the installed provider package directory for a
binary as we install it, we can lazily detect the executable as it is
required. Doing so allows us to separately report an invalid unpacked
package, giving the user more actionable error messages.
* internal/getproviders: decode and return any registry warnings
The public registry may include a list of warnings in the "versions"
response for any given provider. This PR adds support for warnings from
the registry and an installer event to return those warnings to the
user.
We previously had this functionality available for cached packages in the
providercache package. This moves the main implementation of this over
to the getproviders package and then implements it also for PackageMeta,
allowing us to compute hashes in a consistent way across both of our
representations of a provider package.
The new methods on PackageMeta will only be effective for packages in the
local filesystem because we need direct access to the contents in order
to produce the hash. Hopefully in future the registry protocol will be
able to also provide hashes using this content-based (rather than
archive-based) algorithm and then we'll be able to make this work for
PackageMeta referring to a package obtained from a registry too, but
hashes for local packages only are still useful for some cases right now,
such as generating mirror directories in the "terraform providers mirror"
command.
This adds supports for "unmanaged" providers, or providers with process
lifecycles not controlled by Terraform. These providers are assumed to
be started before Terraform is launched, and are assumed to shut
themselves down after Terraform has finished running.
To do this, we must update the go-plugin dependency to v1.3.0, which
added support for the "test mode" plugin serving that powers all this.
As a side-effect of not needing to manage the process lifecycle anymore,
Terraform also no longer needs to worry about the provider's binary, as
it won't be used for anything anymore. Because of this, we can disable
the init behavior that concerns itself with downloading that provider's
binary, checking its version, and otherwise managing the binary.
This is all managed on a per-provider basis, so managed providers that
Terraform downloads, starts, and stops can be used in the same commands
as unmanaged providers. The TF_REATTACH_PROVIDERS environment variable
is added, and is a JSON encoding of the provider's address to the
information we need to connect to it.
This change enables two benefits: first, delve and other debuggers can
now be attached to provider server processes, and Terraform can connect.
This allows for attaching debuggers to provider processes, which before
was difficult to impossible. Second, it allows the SDK test framework to
host the provider in the same process as the test driver, while running
a production Terraform binary against the provider. This allows for Go's
built-in race detector and test coverage tooling to work as expected in
provider tests.
Unmanaged providers are expected to work in the exact same way as
managed providers, with one caveat: Terraform kills provider processes
and restarts them once per graph walk, meaning multiple times during
most Terraform CLI commands. As unmanaged providers can't be killed by
Terraform, and have no visibility into graph walks, unmanaged providers
are likely to have differences in how their global mutable state behaves
when compared to managed providers. Namely, unmanaged providers are
likely to retain global state when managed providers would have reset
it. Developers relying on global state should be aware of this.
* providercache: add logging for errors from getproviders.SearchLocalDirectory
providercache.fillMetaCache() was silently swallowing errors when
searching the cache directory. This commit logs the error without
changing the behavior otherwise.
* command/cliconfig: validate plugin cache dir exists
The plugin cache directory must exist for terraform to use it, so we
will add a check at the begining.
* internal/registry source: return error if requested provider version protocols are not supported
* getproviders: move responsibility for protocol compatibility checks into the registry client
The original implementation had the providercache checking the provider
metadata for protocol compatibility, but this is only relevant for the
registry source so it made more sense to move the logic into
getproviders.
This also addresses an issue where we were pulling the metadata for
every provider version until we found one that was supported. I've
extended the registry client to unmarshal the protocols in
`ProviderVersions` so we can filter through that list, instead of
pulling each version's metadata.
* internal/providercache: verify that the provider protocol version is
compatible
The public registry includes a list of supported provider protocol
versions for each provider version. This change adds verification of
support and adds a specific error message pointing users to the closest
matching version.
* tools/terraform-bundle: refactor to use new provider installer and
provider directory layouts
terraform-bundle now supports a "source" attribute for providers,
uses the new provider installer, and the archive it creates preserves
the new (required) directory hierarchy for providers, under a "plugins"
directory.
This is a breaking change in many ways: source is required for any
non-HashiCorp provider, locally-installed providers must be given a
source (can be arbitrary, see docs) and placed in the expected directory
hierarchy, and the unzipped archive is no longer flat; there is a new
"plugins" directory created with providers in the new directory layout.
This PR also extends the existing test to check the contents of the zip
file.
TODO: Re-enable e2e tests (currently suppressed with a t.Skip)
This commit includes an update to our travis configuration, so the terraform-bundle e2e tests run. It also turns off the e2e tests, which will fail until we have a terraform 0.13.* release under releases.hashicorp.com. We decided it was better to merge this now instead of waiting when we started seeing issues opened from users who built terraform-bundle from 0.13 and found it didn't work with 0.12 - better that they get an immediate error message from the binary directing them to build from the appropriate release.
Providers installed from the registry are accompanied by a list of
checksums (the "SHA256SUMS" file), which is cryptographically signed to
allow package authentication. The process of verifying this has multiple
steps:
- First we must verify that the SHA256 hash of the package archive
matches the expected hash. This could be done for local installations
too, in the future.
- Next we ensure that the expected hash returned as part of the registry
API response matches an entry in the checksum list.
- Finally we verify the cryptographic signature of the checksum list,
using the public keys provided by the registry.
Each of these steps is implemented as a separate PackageAuthentication
type. The local archive installation mechanism uses only the archive
checksum authenticator, and the HTTP installation uses all three in the
order given.
The package authentication system now also returns a result value, which
is used by command/init to display the result of the authentication
process.
There are three tiers of signature, each of which is presented
differently to the user:
- Signatures from the embedded HashiCorp public key indicate that the
provider is officially supported by HashiCorp;
- If the signing key is not from HashiCorp, it may have an associated
trust signature, which indicates that the provider is from one of
HashiCorp's trusted partners;
- Otherwise, if the signature is valid, this is a community provider.
Earlier on in the stubbing of this package we realized that it wasn't
going to be possible to populate the authentication-related bits for all
packages because the relevant metadata just isn't available for packages
that are already local.
However, we just moved ahead with that awkward design at the time because
we needed to get other work done, and so we've been mostly producing
PackageMeta values with all-zeros hashes and just ignoring them entirely
as a temporary workaround.
This is a first step towards what is hopefully a more intuitive model:
authentication is an optional thing in a PackageMeta that is currently
populated only for packages coming from a registry.
So far this still just models checking a SHA256 hash, which is not a
sufficient set of checks for a real release but hopefully the "real"
implementation is a natural iteration of this starting point, and if not
then at least this interim step is a bit more honest about the fact that
Authentication will not be populated on every PackageMeta.
Built-in providers are special providers that are distributed as part of
Terraform CLI itself, rather than being installed separately. They always
live in the terraform.io/builtin/... namespace so it's easier to see that
they are special, and currently there is only one built-in provider named
"terraform".
Previous commits established the addressing scheme for built-in providers.
This commit makes the installer aware of them to the extent that it knows
not to try to install them the usual way and it's able to report an error
if the user requests a built-in provider that doesn't exist or tries to
impose a particular version constraint for a built-in provider.
For the moment the tests for this are the ones in the "command" package
because that's where the existing testing infrastructure for this
functionality lives. A later commit should add some more focused unit
tests here in the internal/providercache package, too.
Due to some incomplete rework of this function in an earlier commit, the
safety check for using the same directory as both the target and the
cache was inverted and was raising an error _unless_ they matched, rather
than _if_ they matched.
This change is verified by the e2etest TestInitProviders_pluginCache,
which is also updated to use the new-style cache directory layout as part
of this commit.
We previously skipped this one because it wasn't strictly necessary for
replicating the old "terraform init" behavior, but we do need it to work
so that things like the -plugin-dir option can behave correctly.
Linking packages from other cache directories and installing from unpacked
directories are fundamentally the same operation because a cache directory
is really just a collection of unpacked packages, so here we refactor
the LinkFromOtherCache functionality to actually be in
installFromLocalDir, and LinkFromOtherCache becomes a wrapper for
the installFromLocalDir function that just calculates the source and
target directories automatically and invalidates the metaCache.
On Unix-derived systems a directory must be marked as "executable" in
order to be accessible, so our previous mode of 0660 here was unsufficient
and would cause a failure if it happened to be the installer that was
creating the plugins directory for the first time here.
Now we'll make it executable and readable for all but only writable by
the same user/group. For consistency, we also make the selections file
itself readable by everyone. In both cases, the umask we are run with may
further constrain these modes.
Just as with the old installer mechanism, our goal is that explicit
provider installation is the only way that new provider versions can be
selected.
To achieve that, we conclude each call to EnsureProviderVersions by
writing a selections lock file into the target directory. A later caller
can then recall the selections from that file by calling SelectedPackages,
which both ensures that it selects the same set of versions and also
verifies that the checksums recorded by the installer still match.
This new selections.json file has a different layout than our old
plugins.json lock file. Not only does it use a different hashing algorithm
than before, we also record explicitly which version of each provider
was selected. In the old model, we'd repeat normal discovery when
reloading the lock file and then fail with a confusing error message if
discovery happened to select a different version, but now we'll be able
to distinguish between a package that's gone missing since installation
(which could previously have then selected a different available version)
from a package that has been modified.
For the old-style provider cache directory model we hashed the individual
executable file for each provider. That's no longer appropriate because
we're giving each provider package a whole directory to itself where it
can potentially have many files.
This therefore introduces a new directory-oriented hashing algorithm, and
it's just using the Go Modules directory hashing algorithm directly
because that's already had its cross-platform quirks and other wrinkles
addressed during the Go Modules release process, and is now used
prolifically enough in Go codebases that breaking changes to the upstream
algorithm would be very expensive to the Go ecosystem.
This is also a bit of forward planning, anticipating that later we'll use
hashes in a top-level lock file intended to be checked in to user version
control, and then use those hashes also to verify packages _during_
installation, where we'd need to be able to hash unpacked zip files. The
Go Modules hashing algorithm is already implemented to consistently hash
both a zip file and an unpacked version of that zip file.
We've been using the models from the "moduledeps" package to represent our
provider dependencies everywhere since the idea of provider dependencies
was introduced in Terraform 0.10, but that model is not convenient to use
for any use-case other than the "terraform providers" command that needs
individual-module-level detail.
To make things easier for new codepaths working with the new-style
provider installer, here we introduce a new model type
getproviders.Requirements which is based on the type the new installer was
already taking as its input. We have new methods in the states, configs,
and earlyconfig packages to produce values of this type, and a helper
to merge Requirements together so we can combine config-derived and
state-derived requirements together during installation.
The advantage of this new model over the moduledeps one is that all of
recursive module walking is done up front and we produce a simple, flat
structure that is more convenient for the main use-cases of selecting
providers for installation and then finding providers in the local cache
to use them for other operations.
This new model is _not_ suitable for implementing "terraform providers"
because it does not retain module-specific requirement details. Therefore
we will likely keep using moduledeps for "terraform providers" for now,
and then possibly at a later time consider specializing the moduledeps
logic for only what "terraform providers" needs, because it seems to be
the only use-case that needs to retain that level of detail.
This was incorrectly removing the _source_ entry prior to creating the
symlink, therefore ending up with a dangling symlink and no source file.
This wasn't obvious before because the test case for LinkFromOtherCache
was also incorrectly named and therefore wasn't running. Fixing the name
of that test made this problem apparent.
The TestLinkFromOtherCache test case now ends up seeing the final resolved
directory rather than the symlink target, because of upstream changes
to the internal/getproviders filesystem scanning logic to handle symlinks
properly.
There's a lot going on in these functions that can be hard to follow from
the outside, so we'll add some additional trace logging so that we can
more easily understand why things are behaving the way they are.
When a provider source produces an HTTP URL location we'll expect it to
resolve to a zip file, which we'll first download to a temporary
directory and then treat it like a local archive.
When a provider source produces a local archive path we'll expect it to
be a zip file and extract it into the target directory.
This does not yet include an implementation of installing from an
already-unpacked local directory. That will follow in a subsequent commit,
likely following a similar principle as in Dir.LinkFromOtherCache.
These new functions allow command implementations to get hold of the
providercache objects and installation source object derived from the
current CLI configuration.
This is not tested yet, but it's a compilable strawman implementation of
the necessary sequence of events to coordinate all of the moving parts
of running a provider installation operation.
This will inevitably see more iteration in later commits as we complete
the surrounding parts and wire it up to be used by "terraform init". So
far, it's just dead code not called by any other package.
Pam Selle