In an earlier commit we added the "Best Practices" situation to try to
clarify the intended uses of named workspaces and to warn against using
them as an alternative to system decomposition.
However, the prior statement was cautious in its recommendations in the
interests of being pragmatic, and as a result we've seen that users have
in some cases misunderstood or disregarded these recommendations.
The new "When to use Multiple Workspaces" section aims to be more explicit
that having multiple separate Terraform configurations is the preferred
solution for many use-cases, and that workspaces are intended for a
more limited set of use-cases around convenient development and testing.
It also emphasizes the analogy to version control branches that was just
a footnote in the prior text, to help the reader become familiar with the
concept by relating it to a concept and workflow they are hopefully
already familiar with.
This new section also attempts to provide a more elaborate description of
the proposed alternative when the goal is system decomposition. In the
long run some of this content would probably be better placed elsewhere
since it is useful advice even for users who never discover named
workspaces, but it can live here for the time being to limit the scope of
this change until we are ready to make more comprehensive revisions to
the docs in this area.
Finally, the introductory documentation here is adjusted slightly in
preparation for the intended future expansion of workspaces to include
stored variable values and, for more tailored backends like Terraform
Enterprise, a full log of prior operations. More revisions will be
required to cover the specifics of this later, but this new framing will
hopefully help users form a mental model of named workspaces that has
room for these future enhancements and the corresponding concept in
Terraform Enterprise, rather than our previous framing that workspaces
are fundamentally just named states.
Fixed a bug in which an example custom provider would be saved as "terraform-provider-custom-v0.0.1" instead of "terraform-provider-custom_v0.0.1". Not an issue when running the bundle on TFE, but a problem when trying to run a terraform bundle locally.
Return the global default timeout if the ResourceData timeouts are nil.
Set the timeouts from the Resource when calling Resource.Data, so that
the config values are always available.
The state locking improvements for the regular command had the side
effect of locking the state in the console, import, graph and push
commands. Those commands had been updated to get a state via the
Backend.Context method, which locks the state whenever possible, and now
need to call Unlock directly.
Add Unlock calls to all commands that call Context directly.
Add a test to remote-exec to make sure the proper timeout is honored
during apply.
TODO: we need some test helpers for provisioners, so they can all be
verified.
The timeout for a provisioner is expected to only apply to the initial
connection. Keep the context for the communicator.Retry separate from
the global cancellation context.
* Updates the chef provisioner to allow specifying a channel
This also updates the omnitruck url to the current url.
Signed-off-by: Scott Hain <shain@chef.io>
* Update omnitruck URL
Signed-off-by: Scott Hain <shain@chef.io>
filterPartialOutputs was not taking into account that some dependent
resources might yet be removed from the graph. Check that they are not
in the targeted set before declaring that the output remain.
Only open a new file descriptor for RefreshState if we haven't written a
state, and don't have the correct state open already. This prevents
windows from failing to refresh a locked state.
ReadState would hide any errors, assuming that it was an empty state.
This can mask errors on Windows, where the OS enforces read locks on the
state file.
Refreshing a locked state on windows could return nil if the read path
was locked, no state was yet written, and the read path is the same as
the write path.
Add a test that locks then refreshes a newly initialized state struct.
To make it easier to include third-party plugins in generated bundles, we'll now search a local directory for available plugins and skip attempting to install from releases.hashicorp.com if a dependency can be satisfied locally.
Combine the ExitStatus and Err values from remote.Cmd into an error
returned by Wait, better matching the behavior of the os/exec package.
Non-zero exit codes are returned from Wait as a remote.ExitError.
Communicator related errors are returned directly.
Clean up all the error handling in the provisioners using a
communicator. Also remove the extra copyOutput synchronization that was
copied from package to package.
Martin Atkins