Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Background work to prepare move to v2 storage API in the client #2017

Open
wants to merge 4 commits into
base: main
Choose a base branch
from

Conversation

ogayot
Copy link
Member

@ogayot ogayot commented Jun 19, 2024

Going forward, the client storage/filesystem controller should perform all of the storage actions by making async calls to the API. These calls will be implemented using coroutines functions that will call aiohttp.

At the start, these new coroutine functions will override the methods from the manipulator. However, the functions defined by the manipulator are not coroutines functions so there is a mismatch.

Move the needed functions to coroutine functions so that we can properly override them with coroutine functions.

This PR introduces:

  • a way to connect a signal to a coroutine connect_async_signal. This is similar in essence to what urwid.connect_signal except that it creates a background task to execute the coroutine
  • a way to declare storage actions as coroutine functions.

NOTE: This may feel like a wasted effort to move manipulator functions to coroutines functions since eventually, the client will not inherit from the manipulator anymore. However, it is necessary to implement the client using async functions so this should allow us to make a simpler transition.

@ogayot ogayot force-pushed the storage-v2-manipulator-async branch 2 times, most recently from 4440207 to 587cc90 Compare June 19, 2024 14:50
Copy link
Collaborator

@mwhudson mwhudson left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A very interesting PR! I have a bunch of quibbles and questions I'm afraid but hopefully they are interesting to think about :-)

subiquitycore/async_helpers.py Show resolved Hide resolved
@@ -45,6 +48,14 @@ def schedule_task(coro, propagate_errors=True):
background_tasks = set()


def connect_async_signal(obj, name, callback, **kwargs):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is it worth typing the callback argument here?

Copy link
Member Author

@ogayot ogayot Jun 20, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think we can do much better than Callable[..., Awaitable[Any]] since the handler can take a variadic number of arguments. At least, it "forces" the use of a coroutine function rather than a function.

connect_signal(obj, name, partial(async_signal_callback, callback), **kwargs)


def async_signal_callback(callback, *args, **kwargs):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe this should have a leading underscore? I think I would be tempted to use a closure rather than partial here but its not very important I guess.

@@ -64,7 +64,7 @@ class MakeBootDevicePlan(abc.ABC):
"""
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Modifying this file feels like it's pushing things too far? It's API is not going to be part of the boundary between the client and server, surely?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed... I haven't found a good way around it yet though.

disk = make_disk(self.manipulator.model, ptable=None)
self.manipulator.reformat(disk, "msdos")
await self.manipulator.reformat(disk, "msdos")
self.assertEqual("msdos", disk.ptable)


Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I hope you didn't do all this by hand!

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I actually did 🙃

device = gap.device
part_align = device.alignment_data().part_align
bootfs_size = align_up(sizes.get_bootfs_size(gap.size), part_align)
gap_boot, gap_rest = gap.split(bootfs_size)
spec = dict(fstype="ext4", mount="/boot")
self.create_partition(device, gap_boot, spec)
part = self.create_partition(device, gap_rest, dict(fstype=None))
await self.create_partition(device, gap_boot, spec)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Now I wonder about things like this. What do we expect the code to look like in the end? Will we still have the FilesystemManipulator methods like create_partition? Will they be async? I realize this is asking after you've done a bunch of work but did you consider the approach of having the client FilesystemController inherit from an async wrapper of FilesystemManipulator full of methods like:

    async def create_partition(self, ...):
        return self._fsm.create_partition(...)

?

(why did I have both FilesystemControllers inherit from FilesystemManipulator rather than given them both an instance of the latter? that was lazy of me)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My intention was that at the end, the client would not use (and not inherit anymore from) the FilesystemManipulator. And if only the server uses the manipulator, then there should be no need for all the async stuff. create_partition would still be implemented in the manipulator but used by the server only.

But unless I can replace all the calls from the client to the manipulator with API calls in one go, I will run into problems with code that is not async.

I did consider a few ways to avoid making everything suddenly async but couldn't come up with something that would theoretically work.

I'm happy to try other solutions. It feels like the wrapper will not be enough to make it possible to call async code from methods of the manipulator ; which is what we'd need in practice AFAICT.

@@ -168,7 +170,9 @@ def refresh_model_inputs(self):
]
actions = [(_("Unmount"), mi.mount.can_delete(), "unmount")]
menu = ActionMenu(actions)
connect_signal(menu, "action", self._mount_action, mi.mount)
connect_async_signal(
menu, "action", self._mount_action, user_args=[mi.mount]
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This sort of thing is interesting. Are the semantics of async callbacks what we want here? In that the task of updating the model happens in the background to some extent and then the UI is redrawn some (in theory) arbitrary time later. Is that what we want? What do we want? Do we want to do TuiApplication._wait_with_indication games to block the UI for MAX_BLOCK_TIME and then do something like grey out the widget if the API call takes longer than that?

Am I overthinking this?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I think we should not update the UI until the API call returns. Whether we should _wait_with_indication is a good point to consider implementing at some point (so far, I don't have any specific use case in mind where the API would be that slow to finish).

I think redrawing the UI after the background task finishes would be ideal. We'd need access to view (or the urwid handler) to request a redraw though ; so I need to think about what that means.

self.parent.refresh_model_inputs()
self.parent.remove_overlay()
# This should probably be moved somewhere else
self.parent.request_redraw_if_visible()
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is the same sort of concern as my last comment I guess. "Doing an API call when a dialog is closed" seems likely to be a common enough operation to erect some infrastructure around!

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed, I'll give this more thoughts.

subiquity/ui/views/filesystem/tests/test_partition.py Outdated Show resolved Hide resolved
@ogayot ogayot force-pushed the storage-v2-manipulator-async branch from 587cc90 to 5ecaba6 Compare June 20, 2024 08:42
@mwhudson
Copy link
Collaborator

I was going to respond to each comment but I think perhaps it's best for me to wait until you ping again?

@ogayot ogayot force-pushed the storage-v2-manipulator-async branch from 5ecaba6 to c33fdc5 Compare June 24, 2024 09:18
self.parent.refresh_model_inputs()
self.parent.remove_overlay()
async def async_confirm() -> None:
await self.parent.controller.reformat(self.obj)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

minor: these two confirms are almost the same, suggest a refactor to unify them

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Later, a call to request_redraw_if_visible was added, does it make sense here? Do all cleanup operations need that? Should that just be part of remove_overlay?

subiquitycore/async_helpers.py Outdated Show resolved Hide resolved
@ogayot ogayot force-pushed the storage-v2-manipulator-async branch 2 times, most recently from 60d3bd1 to 7ca47b2 Compare June 27, 2024 07:24
Comment on lines +283 to +287
await self.parent.controller.remove_boot_disk(disk)
self.parent.refresh_model_inputs()
else:
self.parent.controller.add_boot_disk(disk)
self.parent.refresh_model_inputs()
await self.parent.controller.add_boot_disk(disk)
self.parent.refresh_model_inputs()
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: is there any reason for adding self.parent.refresh_model_inputs() to both clauses?

@ogayot ogayot force-pushed the storage-v2-manipulator-async branch from 7ca47b2 to 0d7640c Compare July 3, 2024 08:48
asyncio.create_task only accepts one positional argument and passing
more than one results in the following exception:

TypeError: create_task() takes 1 positional argument but 2 were given

Therefore, it does not make sense for run_bg_task to forward *args to
it. We now use explicit keyword argument names.

Signed-off-by: Olivier Gayot <[email protected]>
Going forward, the client storage/filesystem controller should perform
all of the storage actions by making async calls to the API. These calls
will be implemented using coroutines functions that will call aiohttp.

At the start, these new coroutine functions will override the methods
from the manipulator. However, the functions defined by the manipulator
are not coroutines functions so there is a mismatch.

Move the needed functions to coroutine functions so that we can properly
override them with coroutine functions.

Signed-off-by: Olivier Gayot <[email protected]>
@ogayot ogayot force-pushed the storage-v2-manipulator-async branch from 0d7640c to 6df6d68 Compare July 25, 2024 08:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants