InvokeAI/invokeai/app/services/latent_storage.py

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

120 lines
3.9 KiB
Python
Raw Normal View History

# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654)
from abc import ABC, abstractmethod
from pathlib import Path
from queue import Queue
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
from typing import Callable, Dict, Optional, Union
import torch
2023-07-27 14:54:01 +00:00
class LatentsStorageBase(ABC):
"""Responsible for storing and retrieving latents."""
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
_on_changed_callbacks: list[Callable[[torch.Tensor], None]]
_on_deleted_callbacks: list[Callable[[str], None]]
def __init__(self) -> None:
self._on_changed_callbacks = list()
self._on_deleted_callbacks = list()
@abstractmethod
def get(self, name: str) -> torch.Tensor:
pass
@abstractmethod
def save(self, name: str, data: torch.Tensor) -> None:
pass
@abstractmethod
def delete(self, name: str) -> None:
pass
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
def on_changed(self, on_changed: Callable[[torch.Tensor], None]) -> None:
"""Register a callback for when an item is changed"""
self._on_changed_callbacks.append(on_changed)
def on_deleted(self, on_deleted: Callable[[str], None]) -> None:
"""Register a callback for when an item is deleted"""
self._on_deleted_callbacks.append(on_deleted)
def _on_changed(self, item: torch.Tensor) -> None:
for callback in self._on_changed_callbacks:
callback(item)
def _on_deleted(self, item_id: str) -> None:
for callback in self._on_deleted_callbacks:
callback(item_id)
class ForwardCacheLatentsStorage(LatentsStorageBase):
"""Caches the latest N latents in memory, writing-thorugh to and reading from underlying storage"""
2023-07-27 14:54:01 +00:00
__cache: Dict[str, torch.Tensor]
__cache_ids: Queue
__max_cache_size: int
__underlying_storage: LatentsStorageBase
def __init__(self, underlying_storage: LatentsStorageBase, max_cache_size: int = 20):
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
super().__init__()
self.__underlying_storage = underlying_storage
self.__cache = dict()
self.__cache_ids = Queue()
self.__max_cache_size = max_cache_size
def get(self, name: str) -> torch.Tensor:
cache_item = self.__get_cache(name)
if cache_item is not None:
return cache_item
latent = self.__underlying_storage.get(name)
self.__set_cache(name, latent)
return latent
def save(self, name: str, data: torch.Tensor) -> None:
self.__underlying_storage.save(name, data)
self.__set_cache(name, data)
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
self._on_changed(data)
def delete(self, name: str) -> None:
self.__underlying_storage.delete(name)
if name in self.__cache:
del self.__cache[name]
feat(backend): selective invalidation for invocation cache This change enhances the invocation cache logic to delete cache entries when the resources to which they refer are deleted. For example, a cached output may refer to "some_image.png". If that image is deleted, and this particular cache entry is later retrieved by a node, that node's successors will receive references to the now non-existent "some_image.png". When they attempt to use that image, they will fail. To resolve this, we need to invalidate the cache when the resources to which it refers are deleted. Two options: - Invalidate the whole cache on every image/latents/etc delete - Selectively invalidate cache entries when their resources are deleted Node outputs can be any shape, with any number of resource references in arbitrarily nested pydantic models. Traversing that structure to identify resources is not trivial. But invalidating the whole cache is a bit heavy-handed. It would be nice to be more selective. Simple solution: - Invocation outputs' resource references are always string identifiers - like the image's or latents' name - Invocation outputs can be stringified, which includes said identifiers - When the invocation is cached, we store the stringified output alongside the "live" output classes - When a resource is deleted, pass its identifier to the cache service, which can then invalidate any cache entries that refer to it The images and latents storage services have been outfitted with `on_deleted()` callbacks, and the cache service registers itself to handle those events. This logic was copied from `ItemStorageABC`. `on_changed()` callback are also added to the images and latents services, though these are not currently used. Just following the existing pattern.
2023-09-20 08:26:47 +00:00
self._on_deleted(name)
2023-07-03 16:17:45 +00:00
def __get_cache(self, name: str) -> Optional[torch.Tensor]:
return None if name not in self.__cache else self.__cache[name]
def __set_cache(self, name: str, data: torch.Tensor):
2023-08-17 22:45:25 +00:00
if name not in self.__cache:
self.__cache[name] = data
self.__cache_ids.put(name)
if self.__cache_ids.qsize() > self.__max_cache_size:
self.__cache.pop(self.__cache_ids.get())
class DiskLatentsStorage(LatentsStorageBase):
"""Stores latents in a folder on disk without caching"""
__output_folder: Union[str, Path]
def __init__(self, output_folder: Union[str, Path]):
2023-06-14 14:16:09 +00:00
self.__output_folder = output_folder if isinstance(output_folder, Path) else Path(output_folder)
self.__output_folder.mkdir(parents=True, exist_ok=True)
def get(self, name: str) -> torch.Tensor:
latent_path = self.get_path(name)
return torch.load(latent_path)
def save(self, name: str, data: torch.Tensor) -> None:
2023-06-14 14:16:09 +00:00
self.__output_folder.mkdir(parents=True, exist_ok=True)
latent_path = self.get_path(name)
torch.save(data, latent_path)
def delete(self, name: str) -> None:
latent_path = self.get_path(name)
2023-06-14 14:16:09 +00:00
latent_path.unlink()
2023-06-14 14:16:09 +00:00
def get_path(self, name: str) -> Path:
return self.__output_folder / name