On linux it is convention for users to have a group with the same GID as the users UID. On macOS this is not the case. This broke the test which lead to discovering this bug.
This is an initial implementation of the mgmt language. It is a
declarative (immutable) functional, reactive, domain specific
programming language. It is intended to be a language that is:
* safe
* powerful
* easy to reason about
With these properties, we hope this language, and the mgmt engine will
allow you to model the real-time systems that you'd like to automate.
This also includes a number of other associated changes. Sorry for the
large size of this patch.
If awschan closes, longpollWatch and snsWatch return nil
instead of an error. This will prevent the engine from
shutting down in case we choose to close the channel
early or from other struct methods.
In longpollWatch it was no longer sufficient to use only
WaitUntilInstanceStopped as it would block if the instance was
terminated. This patch launches two goroutines in its place, one
waits until the instance stops and the other waits until it
terminates. When either one returns, it cancels their context,
and execution continues.
The waiters now return the AwsErr error "ResourceNotReady: exceeded wait
attempts" when the instance state does not converge after 40 retries.
During longpollWatch() we need to detect this error and continue to
the top of the loop so we can restart the waiters and keep watching for
events.
If CheckApply was called when the instance was pending or stopping, it
would return an error. This patch supresses these errors and tells the
engine that the state can't yet be changed.
This patch makes sure that closeChan is closed as soon as the main loop
returns, so any channel operations are unblocked before we run shutdown.
This ensures that the server's goroutine can return before shutdown
completes and we don't panic by trying to serve the client after
shutdown returns.
This patch creates the cloudwatch rule that detects ec2 instance
state changes, and targets the rule to publish on our sns topic
which, in turn, pushes those event notifications to our endpoint.
This patch replaces the call to Server.ListenAndServe() with
Server.Serve(listener) in order to make sure the listener is up
and running before we subscribe to the topic in a future patch.
Builds started failing due to go-libvirt-xml 6d97448. In that patch,
the DomainChannelTarget struct was changed from having a single type
field, to having an individual field for each virtualization type.
This patch updates the connection check in Init to reflect the changes
to go-libvirt-xml, so that builds no longer fail.
If we get an error in the Watch loop, it will send this on awsChan,
which will cause Watch to loop. However, in this scenario it will never
cause closeChan to close, and we will deadlock because we have a
waitGroup in a helper goroutine which is waiting on this channel to
close the context.
Normally this wouldn't be an issue, but since we have more than one
goroutine (with associated waitGroup) it is. It's also good practice to
close all the channels to help avoid this kind of bug.
This patch also moves the waitGroup Wait into a more logical place for
visibility.
This patch adds autoedges between users and groups, and extends
users with additional fields for supplementary groups and a named
primary group. Also, some small fixes to log and error messages.
