Previously if the `daemon.yml` of a host was changed it would first
start up, load that new daemon.yml in, persist the new configuration for
the host to garage using `update-garage-host`, pull that config back
down and persist it to the bootstrap in `runDaemonPmuxOnce`, and restart
all child processes so they get the new config.
Now, once `daemon.yml` is loaded in we immediately merge it into and
persist this host's bootstrap file, prior to ever starting child
processes. This removes the necessity of restarting those process at
start.
This change also allows the bootstrap file to be the sole repository of
information required to pick a garage node to connect to, since it is
presumably always as up-to-date as it can possibly be. This allows for
removing some more logic from `Env`.
The `bootstrap/creator` package is gone, almost as quickly as it
arrived. The `Bootstrap` type is now able to write its own tgz file, and
the two places where bootstrap files are being created pull the data
down to do so and create the `Bootstrap` structs directly.
The structure of the bootstrap file itself has been changed, now there's
just a single `hosts` directory which contains files which are yaml
encodings of the `Host` type, rather than having it be split into
`nebula` and `garage` directories. This makes creating bootstrap files a
lot easier.
Host types have been moved within the `bootstrap` package.
Refactored how boostrap FS is interacted with. There is now a
`Bootstrap` struct which has pre-loaded all data within the bootstrap
FS. This helps centralize the logic around reading the data (though not
yet completely).
Choosing of the garage node to interact with no longer occurs in like
three different places. It occurs at the environment level now, and is
aided by the new `garage.Peer` type.
There has been over 1 year of commit history leading up to this point,
but almost all of that has had some kind network configuration or
secrets built into the code. As of today all of that has been removed,
and the codebase can finally be published!
I am keeping a private copy of the previous commit history, though it's
unclear if it will ever be able to be published.
