The CLI client was quite heavy in loading in a lot of libraries and
modules during runtime, which slowed it down quite a bit, especially on
slower systems.
This commit makes several major changes to help improve the situation.
1. Don't use pkg_resources to get our version, just hardcode it.
2. Reimplement our entire API call to use a custom http.client-based
system that prevents importing any unnecessary libraries (with a
custom User-Agent too).
3. Implement a lazy-loading method for some of the heavier modules, so
that they are only loaded if absolutely necessary.
A missing snapshot could cause an exception here which would break the
entire autobackup run. Catch the exception and continue on as this
should never be a fatal situation.
used_bytes is a broken value since it does not take object replicas into
account, thus throwing off these calculations. Use stored_bytes instead
which properly represents this value.
Moves the age human conversion logic to the client so that this value
can be used by API consumers programatically.
Rounding ensures recent snapshots are not showing as "older" than they
actually are, which is important for accuracy especially with
automirror snapshots and ages above half the next rounding value.
Allows shipping snapshots automatically to remote clusters on a cron,
identically to how autobackup handles local snapshot exports.
VMs are selected based on configured tags, and individual destination
clusters can be specified based on a colon-separated suffix to the
tag(s).
Automirror snapshots use the prefix "am" (analogous to "ab" for
autobackups) to differentiate them from normal "mr" mirrors.
1. Ensure the local connection is actually always present if it exists,
and stored in the store file.
2. Remove any invalid "local" store entries if present (i.e.
pvcapid.yaml entries from legacy versions).
3. Order the connection lists such that "local" is always first.
4. Improve pretty list output format such that all fields are wider if
needed
1. The destination state on an error was invalid; should be "stop".
2. If a lock was listed but removing it fails (because it was already
cleared somehow, this would error. In turn this would cause the VM to
not migrate and be left in an undefined state. Fix that when unlocking
is forced.
