[feat]: send batches of updates in the subprocess

[kyteinsky]

Instead of processing one type of update in a subprocess (which is from forkserver so is fast but is still a subprocess), bucket the list and send the batches in the service functions so all of the updates are done with the overhead of one subprocess

context_chat_backend/context_chat_backend/task_fetcher.py

Lines 346 to 398 in 9c525e5

	for i, (db_id, action_item) in enumerate(q_items.actions.items()):
	try:
	match action_item.type:
	case ActionType.DELETE_SOURCE_IDS:
	exec_in_proc(target=delete_by_source, args=(vectordb_loader, action_item.payload.sourceIds))
	case ActionType.DELETE_PROVIDER_ID:
	exec_in_proc(target=delete_by_provider, args=(vectordb_loader, action_item.payload.providerId))
	case ActionType.DELETE_USER_ID:
	exec_in_proc(target=delete_user, args=(vectordb_loader, action_item.payload.userId))
	case ActionType.UPDATE_ACCESS_SOURCE_ID:
	exec_in_proc(
	target=update_access,
	args=(
	vectordb_loader,
	action_item.payload.op,
	action_item.payload.userIds,
	action_item.payload.sourceId,
	),
	)
	case ActionType.UPDATE_ACCESS_PROVIDER_ID:
	exec_in_proc(
	target=update_access_provider,
	args=(
	vectordb_loader,
	action_item.payload.op,
	action_item.payload.userIds,
	action_item.payload.providerId,
	),
	)
	case ActionType.UPDATE_ACCESS_DECL_SOURCE_ID:
	exec_in_proc(
	target=decl_update_access,
	args=(
	vectordb_loader,
	action_item.payload.userIds,
	action_item.payload.sourceId,
	),
	)
	case _:
	LOGGER.warning(
	f'Unknown action type {action_item.type} for action id {db_id},'
	f' type {action_item.type}, skipping and marking as processed',
	extra={ 'action_item': action_item },
	)
	continue
	processed_event_ids.append(db_id)

https://github.com/nextcloud/context_chat_backend/blob/master/context_chat_backend/vectordb/service.py

Breathing time between updates/writes to the DB may still be required.