Multipass clone

[georgeliao]

Please read the spec first to get a sense of the context.

The code can be divided into several parts:

1. client module, the clone command line part.
2. daemon::clone method in daemon.cpp,
   2.1. generate destination name, this step processes the name options of the input, use the given name or generate a new one if no name is provided. This step also needs the name list of instances to check if the name exist, it also requires the source spec VMSpecs::clone_count field to generate a new name.
   2.2. Once the new destination name is produced, we can clone the vmspec object, this step requires copying the vmspec instance and updating the VMSpecs::default_mac_address and VMSpecs::extra_interfaces fields. Ideally, this method can be a data member method (called clone maybe) in VMSpecs class. However, given the fact that the dependency code (generate_unused_mac_address method and allocated_mac_addrs data) still resides in daemon.cpp, so clone_spec is a local lambda method in daemon.cpp for now.
   2.3. clone image record through the VMImageVault::clone method.
   2.4. calls create_vm_and_clone_instance_dir_data from VirtualMachineFactory.
3. remove_all_snapshots_from_the_image function is added to VirtualMachine class, it gets called in the factory class and it removes the snapshots from the image file.
4. qemu_virtual_machine_factory::create_vm_and_clone_instance_dir_data, it copies the whole instance folder with the exclusion of snapshot files and updates the unique identifiers in cloud-init-config.iso file. Besides that, it creates the new VirtualMachine instance and also removes the snapshots from the image file.
5. auto-completion script of multipass clone command line.

Functional testing should cover at least the following scenarios:

1. multipass clone command line part, --help, and options, explanation of these options. Error message in different cases.
2. Animated spinner works properly in a successful clone case.
3. In the case of cloning without a designated name, check if the name generator works right.
4. In the case of cloning a general vm instance that has mounts, and snapshots, check if the new instance can run properly, check whether the new instance disk data have the updated unique identifiers.

Some further thoughts on the code architecture (open for discussion):

1. Using VMSpecs::clone_count to count cloned instances and the BaseVirtualMachine::snapshot_count to count snapshots is cumbersome. We need a variable and a file or json item to track the value. Besides that, the current name-generation scheme also does not persist on trying. What I mean by that is if the next about-to-be-generated name already exists (if the user already created the snapshot/cloned instance by specifying that name), then the auto name-generation will fail. To fix these things, we can try a different name-generation scheme, that is always looping over the existing instances/snapshots to seek the next available name, this scheme will have to do a linear search but it no longer depends on the counter variable/file and will persist on name generation. The corresponding name-generation behavior will change when there is a gap in the integer sequence (when user deletes the snapshots/instances in between) , the new scheme will reuse that deleted name and the counter based scheme will not. I think this approach is better overall, but it requires some changes in already released behavior.
2. In the multipassd-vm-instances.json file, the metadata::arguments contains the image and clou-init file absolute path, the multipassd-instance-image-records.json file also has the path item points to the image file path. These paths in a way added more unnecessary instance name occurrences (the instance folder name in the path) which need to be updated during the clone and renaming process. Then maybe a better disk data layout and C++ code structure can remove these paths. For example, if we split these json files into one item per file and move that into the instance directory, and meanwhile we can move VMSpecs into VirtualMachine class. With this new data layout and code structure, the VMSpecs can access the instance directory and further access the multipassd-vm-instances.json file (with his own item), besides that, the data sync between the VMSpecs and VirtualMachine class becomes much easier. On top of these, there are more benefits like the rollback mechanism and thread-safety can be much easier.
3. Make a global Mac_address_generator class that can be accessed from everywhere and refactor the corresponding code. If this can be done, then the clone_spec does not have to be a local lambda in daemon.cpp.

[codecov]

Codecov Report

Attention: Patch coverage is 92.30769% with 21 lines in your changes missing coverage. Please review.

Project coverage is 88.93%. Comparing base (b698daa) to head (53c0fd0).
Report is 2 commits behind head on main.

Files	Patch %	Lines
src/daemon/daemon.cpp	91.66%	7 Missing ⚠️
...orm/backends/qemu/qemu_virtual_machine_factory.cpp	76.92%	6 Missing ⚠️
...orm/backends/shared/base_virtual_machine_factory.h	0.00%	4 Missing ⚠️
...rc/platform/backends/shared/base_virtual_machine.h	0.00%	2 Missing ⚠️
...tform/backends/qemu/qemu_virtual_machine_factory.h	0.00%	1 Missing ⚠️
src/utils/yaml_node_utils.cpp	95.23%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3264      +/-   ##
==========================================
+ Coverage   88.88%   88.93%   +0.04%     
==========================================
  Files         254      256       +2     
  Lines       14225    14457     +232     
==========================================
+ Hits        12644    12857     +213     
- Misses       1581     1600      +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sharder996]

In response to your architecture discussion points:

For No.1, I think the current name generation pattern is fine. No need to over complicate at this point as it isn't crucial towards the functionality of the feature.

For No.2, this looks like an extension of the already discussed refactor of the VMSpec, data structures in the Daemon, etc. I'm thinking that refactor is going to touch a lot of files and might need its own planning/technical spec. Regardless, I think it's outside the scope of this feature.

For No.3, cloning a VMSpec sounds like something that the VMSpec should be able to do. Again, this quickly spirals into having to change a bunch of other things as well; make VMSpec a class, maybe combine other data members of the VirtualMachine with the VMSpec, create a singleton MAC address generator, etc. Again, not something that I think is high priority at this point and could be made to be a part of refactoring VMSpec and the daemon as mentioned in the previous point.