While the TLS handshakes are a fairly small overhead compared to
downloading the providers, clients in some situation are failing to
complete the TLS handshake in a timely manner. It's unclear if this is
because of heavily constrained clients are stalling while doing the
major crpto operations, or the edge servers are throttling repeated
requests from the same IPs.
This should allow reusing the open TLS connection to the release edge
servers during init.
In order to parse provider, resource and data source configuration from
HCL2 config files, we need to know the relevant configuration schema.
This new method allows Terraform Core to request these from a provider.
This is a breaking change to this interface, so all of its implementers
in this package are updated too. This includes concrete implementations
of the new method in helper/schema that use the schema conversion code
added in an earlier commit to produce a configschema.Block automatically.
Plugins compiled against prior versions of helper/schema will not have
support for this method, and so calls to them will fail. Callers of
this new method will therefore need to sniff for support using the
SchemaAvailable field added to both ResourceType and DataSource.
This careful handling will need to persist until next time we increment
the plugin protocol version, at which point we can make the breaking
change of requiring this information to be available.
For users that have metered or slow internet connections it is annoying
to have Terraform constantly re-downloading the same files when they
initialize many separate directories.
To help such users, here we add an opt-in mechanism to use a local
directory as a read-through cache. When enabled, any plugin download will
be skipped if a suitable file already exists in the cache directory. If
the desired plugin isn't in the cache, it will be downloaded into the
cache for use next time.
This mechanism also serves to reduce total disk usage by allowing
plugin files to be shared between many configurations, as long as the
target system isn't Windows and supports either hardlinks or symlinks.
If we encounter something that isn't a file -- for example, a dangling
symlink whose referent has been deleted -- we'll ignore it so that we
can either later produce a "no such plugin" error or auto-install a plugin
that will actually work.
The go-plugin package now uses hclog. The default Logger has a level set
to Info, but all plugin output is relayed via Debug. Create a new named
logger for plugins with the level set to Trace so that all output comes
through.
On Windows systems the plugin binaries use a .exe suffix, which we were
misparsing as part of either the "v" or "x" parts of the filename.
This fixes#15578.
If the release site is missing the "x-terraform-protocol-version"
header, we should fetch the latest spec'ed release. Downloading the
wrong protocol version can't do any damage, and the version present is
more than likely compatible.
Previously we forced only installing for the current GOOS and GOARCH. Now
we allow this to be optionally overridden, which allows building tools
that can, for example, populate a directory with plugins to run on a Linux
server while working on a Mac.
Some errors from Get are essentially user error, so we want to be able to
recognize them and give the user good feedback on how to proceed.
Although sentinel values are not an ideal solution to this, it's something
reasonably simple we can do to get this done without lots of refactoring.
Fetch the SHA256SUMS file and verify it's signature before downloading
any plugins.
This embeds the hashicorp public key in the binary. If the publickey is
replaced, new releases will need to be cut anyway. A
--verify-plugin=false flag will be added to skip signature verification
in these cases.
Since the command package also needs to know about the specific OS_ARCH
directories, remove the logic fom the discovery package.
This doesn't completely remove the knowledge of the path from discovery,
in order to maintain the current behavior of skipping legacy plugin
names within a new-style path.
Previously we had a "getProvider" function type used to implement plugin
fetching. Here we replace that with an interface type, initially with
just a "Get" function.
For now this just simplifies the interface by allowing the target
directory and protocol version to be members of the struct rather than
passed as arguments.
A later change will extend this interface to also include a method to
purge unused plugins, so that upgrading frequently doesn't leave behind
a trail of unused executable files.
When running "terraform init" with providers that are unconstrained, we
will now produce information to help the user update configuration to
constrain for the particular providers that were chosen, to prevent
inadvertently drifting onto a newer major release that might contain
breaking changes.
A ~> constraint is used here because pinning to a single specific version
is expected to create dependency hell when using child modules. By using
this constraint mode, which allows minor version upgrades, we avoid the
need for users to constantly adjust version constraints across many
modules, but make major version upgrades still be opt-in.
Any constraint at all in the configuration will prevent the display of
these suggestions, so users are free to use stronger or weaker constraints
if desired, ignoring the recommendation.
We can filter the allowed versions and sort them before checking the
protocol version, that way we can just return the first one found
reducing network requests.
Extend the test reslease server to return the protocol version header
and a dummy zip file for the provider.
Test filtering the plugins by plugin protocol version and add a full
GetProvder test.
Get provider needs to be provided with the plugin protocol version,
because it can't be imported here directly.
The plugin url types and methods were confusing; replace them with a few
functions to format the urls.
This is used to mark the plugin protocol version. Currently we actually
just ignore this entirely, since only one protocol version exists anyway.
Later we will need to add checks here to ensure that we only pay attention
to plugins of the right version.
As well as constraining plugins by version number, we also want to be
able to pin plugins to use specific executables so that we can detect
drift in available plugins between commands.
This commit allows such requirements to be specified, but doesn't yet
specify any such requirements, nor validate them.
Add discovery.GetProviders to fetch plugins from the relases site.
This is an early version, with no tests, that only (probably) fetches
plugins from the default location. The URLs are still subject to change,
and since there are no plugin releases, it doesn't work at all yet.
Instead of providing the a path in BackendOpts, provide a loaded
*config.Config instead. This reduces the number of places where
configuration is loaded.
The semver library we were using doesn't have support for a "pessimistic
constraint" where e.g. the user wants to accept only minor or patch
version upgrades. This is important for providers since users will
generally want to pin their dependencies to not inadvertantly accept
breaking changes.
So here we switch to hashicorp's home-grown go-version library, which
has the ~> constraint operator for this sort of constraint.
Given how much the old version object was already intruding into the
interface and creating dependency noise in callers, this also now wraps
the "raw" go-version objects in package-local structs, thus keeping the
details encapsulated and allowing callers to deal just with this package's
own types.
Having this as a method of PluginMeta felt most natural, but unfortunately
that means that discovery must depend on plugin and plugin in turn
depends on core Terraform, thus making the discovery package hard to use
without creating dependency cycles.
To resolve this, we invert the dependency and make the plugin package be
responsible for instantiating clients given a meta, using a top-level
function.
The .terraformrc file allows the user to override the executable location
for certain plugins. This mechanism allows us to retain that behavior for
a deprecation period by treating such executables as an unversioned
plugin for the given name and excluding all other candidates for that
name, thus ensuring that the override will "win".
Users must eventually transition away from using this mechanism and use
vendor directories instead, because these unversioned overrides will never
match for a provider referenced with non-zero version constraints.
These new methods ClientConfig and Client provide the bridge into the
main plugin infrastructure by configuring and instantiating (respectively)
a client object for the referenced plugin.
This stops short of getting the proxy object from the client since that
then requires referencing the interface for the plugin kind, which would
then create a dependency on the main terraform package which we'd rather
avoid here. It'll be the responsibility of the caller in the command
package to do the final wiring to get a provider instance out of a
provider plugin client.
For now this supports both our old and new directory layouts, so we can
preserve compatibility with existing configurations until a future major
release where support for the old paths will be removed.
Currently this allows both styles in all given directories, which means we
support more variants than we actually intend to but this is accepted to
keep the interface simple such that we can easily remove the legacy
exception later. The documentation will reflect only the subset of
path layouts that we actually intend to support.
With forthcoming support for versioned plugins we need to be able to
answer questions like what versions of plugins are currently installed,
what's the newest version of a given plugin available, etc.
PluginMetaSet gives us a building block for this sort of plugin version
wrangling.
This is necessary since the TypeUnknown HIL handling in helper/schema
makes providers compiled WITHOUT TypeUnknown incompatible with the way
core handles unknown values.
This reverts commit c3a4cff133, reversing
changes made to 791a02e6e4.
This change requires plugin recompilation and we should hold off until a
minor release for that.
This is a breaking change to the ResourceProvider interface that adds the
new operations relating to data sources.
DataSources, ValidateDataSource, ReadDataDiff and ReadDataApply are the
data source equivalents of Resources, Validate, Diff and Apply (respectively)
for managed resources.
The diff/apply model seems at first glance a rather strange workflow for
read-only resources, but implementing data resources in this way allows them
to fit cleanly into the standard plan/apply lifecycle in cases where the
configuration contains computed arguments and thus the read must be deferred
until apply time.
Along with breaking the interface, we also fix up the plugin client/server
and helper/schema implementations of it, which are all of the callers
used when provider plugins use helper/schema. This would be a breaking
change for any provider plugin that directly implements the provider
interface, but no known plugins do this and it is not recommended.
At the helper/schema layer the implementer sees ReadDataApply as a "Read",
as opposed to "Create" or "Update" as in the managed resource Apply
implementation. The planning mechanics are handled entirely within
helper/schema, so that complexity is hidden from the provider implementation
itself.
This replaces this plugin system with the extracted hashicorp/go-plugin
library.
This doesn't introduce any new features such as binary flattening but
opens us up to that a lot more easily and removes a lot of code from TF
in favor of the upstream lib.
This will introduce a protocol change that will cause all existing
plugins to have to be recompiled to work properly. There is no actual
API changes so they just have to recompile, but it is technically
backwards incompatible.
Now that we support log line filtering (as of 0090c063) it's good to be
a bit more fussy about what log levels are assigned to different things.
Here we make a few things that are implementation details log as DEBUG,
and prevent spurious errors from EvalValidateCount where it was returning
an empty EvalValidateError rather than nil when everything was okay.
This is needed as preperation for adding WinRM support. There is still
one error in the tests which needs another look, but other than that it
seems like were now ready to start working on the WinRM part…
Sean Chittenden