core.py

# Copyright 2020 H2O.ai, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from io import BufferedReader
import asyncio
import ipaddress
import json
import platform
import secrets
import shutil
import subprocess
from urllib.parse import urlparse
from uuid import uuid4
import warnings
import logging
import os
import os.path
import sys
from typing import List, Dict, Union, Tuple, Any, Optional, IO

import httpx

logger = logging.getLogger(__name__)

Primitive = Union[bool, str, int, float, None]
PrimitiveCollection = Union[Tuple[Primitive], List[Primitive]]
FileContent = Union[IO[str], IO[bytes], str, bytes]

UNICAST = 'unicast'
MULTICAST = 'multicast'
BROADCAST = 'broadcast'


def _get_env(key: str, value: Any):
    return os.environ.get(f'H2O_WAVE_{key}', value)


_default_internal_address = 'http://127.0.0.1:8000'
_base_url = _get_env('BASE_URL', '/')


class _Config:
    def __init__(self):
        self.internal_address = _get_env('INTERNAL_ADDRESS', _default_internal_address)
        self.app_address = _get_env('APP_ADDRESS', _get_env('EXTERNAL_ADDRESS', self.internal_address))
        self.app_mode = _get_env('APP_MODE', UNICAST)
        self.hub_base_url = _get_env('BASE_URL', '/')
        self.hub_host_address = _get_env('ADDRESS', 'http://127.0.0.1:10101')
        self.hub_address = self.hub_host_address + self.hub_base_url
        self.hub_connection_timeout: int = int(_get_env('CONNECTION_TIMEOUT', '0'))
        self.hub_access_key_id: str = _get_env('ACCESS_KEY_ID', 'access_key_id')
        self.hub_access_key_secret: str = _get_env('ACCESS_KEY_SECRET', 'access_key_secret')
        self.app_access_key_id: str = _get_env('APP_ACCESS_KEY_ID', None) or secrets.token_urlsafe(16)
        self.app_access_key_secret: str = _get_env('APP_ACCESS_KEY_SECRET', None) or secrets.token_urlsafe(16)


_config = _Config()

_key_sep = ' '
_content_type_json = {'Content-type': 'application/json'}


def _is_int(x: Any) -> bool: return isinstance(x, int)


def _is_str(x: Any) -> bool: return isinstance(x, str)


def _is_list(x: Any) -> bool: return isinstance(x, (list, tuple))


def _is_primitive(x: Any) -> bool: return x is None or isinstance(x, (bool, str, int, float))


def _guard_primitive(x: Any):
    if not _is_primitive(x):
        raise ValueError('value must be a primitive')


def _are_primitives(xs: Any) -> bool:
    if xs is None:
        return True
    if not _is_list(xs):
        return False
    for x in xs:
        if not _is_primitive(x):
            return False
    return True


def _guard_primitive_list(xs: Any):
    if not _are_primitives(xs):
        raise ValueError('value must be a primitive list or tuple')


def _guard_primitive_dict_values(d: Dict[str, Any]):
    if d:
        for x in d.values():
            _guard_primitive(x)


def _guard_str_key(key: str):
    if not _is_str(key):
        raise KeyError('key must be str')
    if ' ' in key:
        raise KeyError('keys cannot contain spaces')


def _guard_key(key: str):
    if _is_str(key):
        _guard_str_key(key)
    else:
        if not _is_int(key):
            raise KeyError('invalid key type: want str or int')


def _rebase(host: str, path: str):
    return f"{host}{path.lstrip('/')}"


class ServiceError(Exception):
    pass


DICT = '__kv'


class Expando:
    """
    Represents an object whose members (attributes) can be dynamically added and removed at run time.

    Args:
        args: An optional ``dict`` of attribute-value pairs to initialize the expando instance with.
    """

    def __init__(self, args: Optional[Dict] = None):
        self.__dict__[DICT] = args if isinstance(args, dict) else dict()

    def __getattr__(self, k): return self.__dict__[DICT].get(k)

    def __getitem__(self, k): return self.__dict__[DICT].get(k)

    def __setattr__(self, k, v): self.__dict__[DICT][k] = v

    def __setitem__(self, k, v): self.__dict__[DICT][k] = v

    def __contains__(self, k): return k in self.__dict__[DICT]

    def __delattr__(self, k): del self.__dict__[DICT][k]

    def __delitem__(self, k): del self.__dict__[DICT][k]

    def __repr__(self): return repr(self.__dict__[DICT])

    def __str__(self): return ', '.join([f'{k}:{repr(v)}' for k, v in self.__dict__[DICT].items()])


def expando_to_dict(e: Expando) -> dict:
    """
    Extract an expando's underlying dictionary.
    Any modifications to the dictionary also affect the original expando.

    Args:
        e: The expando instance.

    Returns:
        The expando's dictionary.

    """
    return e.__dict__[DICT]


def clone_expando(source: Expando, exclude_keys: Optional[Union[list, tuple]] = None,
                  include_keys: Optional[Union[list, tuple]] = None) -> Expando:
    """
    Clone an expando instance. Creates a shallow clone.

    Args:
        source: The expando to clone.
        exclude_keys: Keys to exclude while cloning.
        include_keys: Keys to include while cloning.
    Returns:
        The expando clone.
    """
    return copy_expando(source, Expando(), exclude_keys, include_keys)


def copy_expando(source: Expando, target: Expando, exclude_keys: Optional[Union[list, tuple]] = None,
                 include_keys: Optional[Union[list, tuple]] = None) -> Expando:
    """
    Copy all entries from the source expando instance to the target expando instance.

    Args:
        source: The expando to copy from.
        target: The expando to copy to.
        exclude_keys: Keys to exclude while copying.
        include_keys: Keys to include while copying.
    Returns:
        The target expando.
    """
    if include_keys:
        if exclude_keys:
            for k in include_keys:
                if k not in exclude_keys:
                    target[k] = source[k]
        else:
            for k in include_keys:
                target[k] = source[k]
    else:
        d = expando_to_dict(source)
        if exclude_keys:
            for k, v in d.items():
                if k not in exclude_keys:
                    target[k] = v
        else:
            for k, v in d.items():
                target[k] = v

    return target


PAGE = '__page__'
KEY = '__key__'


def _set_op(o, k, v):
    _guard_key(k)
    k = getattr(o, KEY) + _key_sep + str(k)
    if isinstance(v, Data):
        op = v.dump()
        op['k'] = k
    else:
        op = dict(k=k, v=v)
    return op


def _can_dump(x: Any):
    return hasattr(x, 'dump') and callable(x.dump)


def _is_numpy_obj(x: Any) -> bool:
    if 'numpy' in sys.modules:
        np = sys.modules['numpy']
        if isinstance(x, (np.ndarray, np.dtype, np.integer, np.floating)):
            return True
    return False


def _dump(xs: Any):
    if _is_numpy_obj(xs):
        raise ValueError('NumPy objects are not serializable by Wave')

    if isinstance(xs, (list, tuple)):
        return [_dump(x) for x in xs]
    elif isinstance(xs, dict):
        return {k: _dump(v) for k, v in xs.items()}
    elif _can_dump(xs):
        return xs.dump()
    else:
        return xs


class Ref:
    """
    Represents a local reference to an element on a `h2o_wave.core.Page`.
    Any changes made to this local reference are tracked and sent to the remote Wave server when the page is saved.
    """

    def __init__(self, page: 'PageBase', key: str):
        self.__dict__[PAGE] = page
        self.__dict__[KEY] = key

    def __getattr__(self, key):
        _guard_key(key)
        return Ref(getattr(self, PAGE), getattr(self, KEY) + _key_sep + key)

    def __getitem__(self, key):
        _guard_key(key)
        return Ref(getattr(self, PAGE), getattr(self, KEY) + _key_sep + str(key))

    def __setattr__(self, key, value):
        if isinstance(value, Data):
            raise ValueError('Data instances cannot be used in assignments.')
        getattr(self, PAGE)._track(_set_op(self, key, _dump(value)))

    def __setitem__(self, key, value):
        if isinstance(value, Data):
            raise ValueError('Data instances cannot be used in assignments.')
        getattr(self, PAGE)._track(_set_op(self, key, _dump(value)))


class Data:
    """
    Represents a data placeholder. A data placeholder is used to allocate memory on the Wave server to store data.

    Args:
        fields: The names of the fields (columns names) in the data, either a list or tuple or string containing space-separated names.
        size: The number of rows to allocate memory for. Positive for fixed buffers, negative for circular buffers and zero for variable length buffers.
        data: Initial data. Must be either a key-row ``dict`` for variable-length buffers OR a row ``list`` for fixed-size and circular buffers.
    """

    def __init__(self, fields: Union[str, tuple, list], size: int = 0, data: Optional[Union[dict, list]] = None):
        self.fields = fields
        self.data = data
        self.size = size

    def dump(self):
        f = self.fields
        d = self.data
        n = self.size
        if d:
            if isinstance(d, dict):
                return dict(m=dict(f=f, d=d))
            else:
                if n < 0:
                    return dict(c=dict(f=f, d=d))
                else:
                    return dict(f=dict(f=f, d=d))
        else:
            if n == 0:
                return dict(m=dict(f=f))
            else:
                if n < 0:
                    return dict(c=dict(f=f, n=-n))
                else:
                    return dict(f=dict(f=f, n=n))


def data(
        fields: Union[str, tuple, list],
        size: int = 0,
        rows: Optional[Union[dict, list]] = None,
        columns: Optional[Union[dict, list]] = None,
        pack=False,
) -> Union[Data, str]:
    """
    Create a `h2o_wave.core.Data` instance for associating data with cards.

    ``data(fields, size)`` creates a placeholder for data and allocates memory on the Wave server.

    ``data(fields, size, rows)`` creates a placeholder and initializes it with the provided rows.

    If ``pack`` is ``True``, the ``size`` parameter is ignored, and the function returns a packed string representing the data.

    Args:
        fields: The names of the fields (columns names) in the data, either a list or tuple or string containing space-separated names.
        size: The number of rows to allocate memory for. Positive for fixed buffers, negative for circular buffers and zero for variable length buffers.
        rows: The rows in this data.
        columns: The columns in this data.
        pack: True to return a packed string representing the data instead of a `h2o_wave.core.Data` placeholder.

    Returns:
        Either a `h2o_wave.core.Data` placeholder or a packed string representing the data.
    """
    if _is_str(fields):
        fields = fields.strip()
        if fields == '':
            raise ValueError('fields is empty')
        fields = fields.split()
    if not _is_list(fields):
        raise ValueError('fields must be tuple or list')
    if len(fields) == 0:
        raise ValueError('fields is empty')
    for field in fields:
        if not _is_str(field):
            raise ValueError('field must be str')
        if field == '':
            raise ValueError('field cannot be empty str')

    if pack:
        if rows:
            if not isinstance(rows, list):
                # TODO validate if 2d
                raise ValueError('rows must be a list')
            return 'rows:' + marshal((fields, rows))
        if columns:
            if not isinstance(columns, list):
                # TODO validate if 2d
                raise ValueError('columns must be a list')
            return 'cols:' + marshal((fields, columns))
        raise ValueError('either rows or columns must be provided if pack=True')

    if rows:
        if not isinstance(rows, (list, dict)):
            raise ValueError('rows must be list or dict')
    elif columns:  # transpose to rows
        # TODO issue warning: better for caller to use pack=True
        n = len(columns[0])
        rows = []
        for i in range(n):
            rows.append([c[i] for c in columns])

    if not _is_int(size):
        raise ValueError('size must be int')

    return Data(fields, size, rows)


class _ServerCacheBase:
    def _keys(self, text: str) -> List[str]:
        content = text.strip()
        if len(content):
            return content.split('\n')
        return []


class _AsyncServerCache(_ServerCacheBase):
    def __init__(self, http: httpx.AsyncClient):
        self._http = http

    async def get(self, shard: str, key: str, default=None) -> Any:
        res = await self._http.get(f'{_config.hub_address}_c/{shard}/{key}')
        if res.status_code == 200:
            return unmarshal(res.text)
        return default

    async def keys(self, shard: str) -> List[str]:
        res = await self._http.get(f'{_config.hub_address}_c/{shard}')
        return self._keys(res.text) if res.status_code == 200 else []

    async def set(self, shard: str, key: str, value: Any):
        content = marshal(value)
        res = await self._http.put(f'{_config.hub_address}_c/{shard}/{key}', content=content)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')


class _ServerCache(_ServerCacheBase):
    def __init__(self, http: httpx.Client):
        self._http = http

    def get(self, shard: str, key: str, default=None):
        res = self._http.get(f'{_config.hub_address}_c/{shard}/{key}')
        if res.status_code == 200:
            return unmarshal(res.text)
        return default

    def keys(self, shard: str) -> List[str]:
        res = self._http.get(f'{_config.hub_address}_c/{shard}')
        return self._keys(res.text) if res.status_code == 200 else []

    def set(self, shard: str, key: str, value: Any):
        content = marshal(value)
        res = self._http.put(f'{_config.hub_address}_c/{shard}/{key}', content=content)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')


class PageBase:
    """
    Represents a remote page.

    Args:
        url: The URL of the remote page.
    """

    def __init__(self, url: str):
        self.url = url
        self._changes = []

    def add(self, key: str, card: Any) -> Ref:
        """
        Add a card to this page.

        Args:
            key: The card's key. Must uniquely identify the card on the page. Overwrites any other card with the same key.
            card: A card. Use one of the ``ui.*_card()`` to create cards.

        Returns:
            A reference to the added card.
        """
        _guard_str_key(key)

        props: Optional[dict] = None

        if isinstance(card, dict):
            props = card
        elif _can_dump(card):
            props = _dump(card)
        if not isinstance(props, dict):
            raise ValueError('card must be dict or implement .dump() -> dict')

        data = []
        bufs = []
        for k, v in props.items():
            if isinstance(v, Data):
                data.append((k, len(bufs)))
                bufs.append(v.dump())

        for k, v in data:
            del props[k]
            props[f'~{k}'] = v

        if len(bufs) > 0:
            self._track(dict(k=key, d=props, b=bufs))
        else:
            self._track(dict(k=key, d=props))

        return Ref(self, key)

    def _track(self, op: dict):
        self._changes.append(op)

    def _diff(self):
        if len(self._changes) == 0:
            return None
        d = marshal(dict(d=self._changes))
        self._changes.clear()
        return d

    def drop(self):
        """
        Delete this page from the remote site. Same as ``del site[url]``.
        """
        self._track({})

    def __setitem__(self, key, card):
        self.add(key, card)

    def __getitem__(self, key: str) -> Ref:
        _guard_str_key(key)
        return Ref(self, key)

    def __delitem__(self, key: str):
        _guard_str_key(key)
        self._track(dict(k=key))


class Page(PageBase):
    """
    Represents a reference to a remote Wave page.

    Args:
        site: The parent site.
        url: The URL of this page.
    """

    def __init__(self, site: 'Site', url: str):
        self.site = site
        super().__init__(url)

    def load(self) -> dict:
        """
        Retrieve the serialized form of this page from the remote site.

        Returns:
            The serialized form of this page
        """
        return self.site.load(self.url)

    def sync(self):
        """
        DEPRECATED: Use `h2o_wave.core.Page.save` instead.
        """
        warnings.warn('page.sync() is deprecated. Please use page.save() instead.', DeprecationWarning)
        self.save()

    def save(self):
        """
        Save the page. Sends all local changes made to this page to the remote site.
        """
        p = self._diff()
        if p:
            logger.debug(data)
            self.site._save(self.url, p)


class AsyncPage(PageBase):
    """
    Represents a reference to a remote Wave page. Similar to `h2o_wave.core.Page` except that this class exposes ``async`` methods.


    Args:
        site: The parent site.
        url: The URL of this page.
    """

    def __init__(self, site: 'AsyncSite', url: str):
        self.site = site
        super().__init__(url)

    async def load(self) -> dict:
        """
        Retrieve the serialized form of this page from the remote site.

        Returns:
            The serialized form of this page
        """
        return await self.site.load(self.url)

    async def push(self):
        """
        DEPRECATED: Use `h2o_wave.core.AsyncPage.save` instead.
        """
        warnings.warn('page.push() is deprecated. Please use page.save() instead.', DeprecationWarning)
        await self.save()

    async def save(self):
        """
        Save the page. Sends all local changes made to this page to the remote site.
        """
        p = self._diff()
        if p:
            logger.debug(p)
            await self.site._save(self.url, p)


class Site:
    """
    Represents a reference to the remote Wave site. A Site instance is used to obtain references to the site's pages.
    """

    def __init__(self):
        self._http = httpx.Client(
            auth=(_config.hub_access_key_id, _config.hub_access_key_secret),
            verify=False,
        )
        self.cache = _ServerCache(self._http)

    def __getitem__(self, url) -> Page:
        return Page(self, url)

    def __delitem__(self, key: str):
        page = self[key]
        page.drop()

    def _save(self, url: str, patch: str):
        res = self._http.patch(_rebase(_config.hub_address, url), content=patch)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')

    def load(self, url) -> dict:
        """
        Retrieve data at the given URL, typically the serialized form of a page.

        Args:
            url: The URL to read.

        Returns:
            The serialized page.
        """
        res = self._http.get(_rebase(_config.hub_address, url), headers=_content_type_json)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')
        return res.json()

    def upload(self, files: List[str]) -> List[str]:
        """
        Upload local files to the site.

        Args:
            files: A list of file paths of the files to be uploaded..

        Returns:
            A list of remote URLs for the uploaded files, in order.
        """
        for f in files:
            if not os.path.isfile(f):
                raise ValueError(f'{f} is not a file.')

        waved_dir = _get_env('WAVED_DIR', None)
        data_dir = _get_env('DATA_DIR', 'data')
        skip_local_upload = _get_env('NO_COPY_UPLOAD', 'false').lower() in ['true', '1', 't']

        # If we know the path of waved and running app on the same machine,
        # we can simply copy the files instead of making an HTTP request.
        if _is_loopback_address() and not skip_local_upload and waved_dir and data_dir:
            try:
                uploaded_files = []
                for f in files:
                    uuid = str(uuid4())
                    dst = os.path.join(waved_dir, data_dir, 'f', uuid)
                    os.makedirs(dst, exist_ok=True)

                    if 'Windows' in platform.system():
                        src = os.path.dirname(f) or os.getcwd()
                        args = ['robocopy', src, dst, os.path.basename(f), '/J', '/W:0']
                    else:
                        args = ['cp', f, dst]

                    _, err = subprocess.Popen(args, stderr=subprocess.PIPE, stdout=subprocess.DEVNULL).communicate()
                    if err:
                        raise ValueError(err.decode())

                    uploaded_files.append(f'{_base_url}_f/{uuid}/{os.path.basename(f)}')
                return uploaded_files
            except:
                pass

        uploaded_files = []
        for f in files:
            uploaded_files.append(('files', (os.path.basename(f), open(f, 'rb'))))

        res = self._http.post(f'{_config.hub_address}_f/', files=uploaded_files)

        for _, f in uploaded_files:
            f[1].close()

        if res.status_code == 200:
            return json.loads(res.text)['files']
        raise ServiceError(f'Upload failed (code={res.status_code}): {res.text}')

    def upload_dir(self, directory: str) -> str:
        """
        WARNING: Experimental and subject to change.
        Upload whole directory to the site with directory structure preserved.

        Args:
            directory: Folder to be uploaded.

        Returns:
            A list of remote URLs for the uploaded directory (always size of 1).
        """
        if not os.path.isdir(directory):
            raise ValueError(f'{directory} is not a directory.')

        waved_dir = _get_env('WAVED_DIR', None)
        data_dir = _get_env('DATA_DIR', 'data')
        skip_local_upload = _get_env('NO_COPY_UPLOAD', 'false').lower() in ['true', '1', 't']

        # If we know the path of waved and running app on the same machine,
        # we can simply copy the files instead of making an HTTP request.
        if _is_loopback_address() and not skip_local_upload and waved_dir and data_dir:
            try:
                uuid = str(uuid4())
                dst = os.path.join(waved_dir, data_dir, 'f', uuid)
                os.makedirs(dst, exist_ok=True)

                if 'Windows' in platform.system():
                    args = ['robocopy', directory, dst, '/S', '/J', '/W:0', '*.*']
                else:
                    args = ['rsync', '-a', os.path.join(directory, '.'), dst]

                _, err = subprocess.Popen(args, stderr=subprocess.PIPE).communicate()
                if err:
                    raise ValueError(err.decode())

                return [f'{_base_url}_f/{uuid}']
            except:
                pass

        upload_files = []
        for f in _get_files_in_directory(directory, []):
            upload_files.append(('files', (os.path.relpath(f, directory), open(f, 'rb'))))

        res = self._http.post(f'{_config.hub_address}_f/', headers={'Wave-Directory-Upload': "True"}, files=upload_files)

        for _, f in upload_files:
            f[1].close()

        if res.status_code == 200:
            return json.loads(res.text)['files']
        raise ServiceError(f'Upload failed (code={res.status_code}): {res.text}')


    def download(self, url: str, path: str) -> str:
        """
        Download a file from the site.

        Args:
            url: The URL of the file.
            path: The local directory or file path to download to. If a directory is provided, the original name of the file is retained.

        Returns:
            The path to the downloaded file.
        """
        path = os.path.abspath(path)
        # If path is a directory, get basename from url
        filepath = os.path.join(path, os.path.basename(url)) if os.path.isdir(path) else path

        with self._http.stream('GET', f'{_config.hub_host_address}{url}') as res:
            if res.status_code != 200:
                res.read()
                raise ServiceError(f'Download failed (code={res.status_code}): {res.text}')
            with open(filepath, 'wb') as f:
                for chunk in res.iter_bytes():
                    f.write(chunk)

        return filepath

    def unload(self, url: str):
        """
        Delete an uploaded file from the site.

        Args:
            url: The URL of the file to delete.
        """
        res = self._http.delete(f'{_config.hub_host_address}{url}')
        if res.status_code == 200:
            return
        raise ServiceError(f'Unload failed (code={res.status_code}): {res.text}')

    def uplink(self, path: str, content_type: str, file: FileContent) -> str:
        """
        Create or update a stream of images.

        The typical use of this function is to transmit a stream of images to a web page, providing the appearance of a
         video. The path returned by this function can be passed to ui.image() or ui.image_card() (or even a custom HTML
         img element). The stream is displayed in browsers using multipart/x-mixed-replace content. Callers must call
         the unlink() function to signal end-of-stream.

        Args:
            path: a unique path or name for the stream (e.g. 'foo/bar/qux.png'). Must be a valid URL path.
            content_type: The MIME type of the streamed content (e.g. 'image/jpeg', 'image/png', etc.).
            file: A file or file-like object (on-disk file, standard I/O, in-memory buffers, sockets or pipes).

        Returns:
            The stream endpoint, typically used as an image path.
        """
        endpoint = f'_m/{path}'
        res = self._http.post(f'{_config.hub_address}{endpoint}', files={'f': ('f', file, content_type)})
        if res.status_code == 200:
            return _config.hub_base_url + endpoint
        raise ServiceError(f'Uplink failed (code={res.status_code}): {res.text}')

    def unlink(self, path: str):
        """
        Signal the end of a stream.

        Args:
            path: The path of the stream
        """
        endpoint = f'_m/{path}'
        res = self._http.delete(f'{_config.hub_address}{endpoint}')
        if res.status_code == 200:
            return
        raise ServiceError(f'Unlink failed (code={res.status_code}): {res.text}')

    def proxy(self, method: str, url: str, headers: Optional[Dict[str, List[str]]] = None, body: Optional[str] = None):
        req = dict(method=method, url=url, headers=headers, body=body)
        res = self._http.post(f'{_config.hub_address}_p/', content=marshal(req))
        if res.status_code == 200:
            return res.json()
        raise ServiceError(f'Proxy request failed (code={res.status_code}): {res.text}')


site = Site()


class AsyncSite:
    """
    Represents a reference to the remote Wave site. Similar to `h2o_wave.core.Site` except that this class exposes `async` methods.
    """

    def __init__(self):
        self._http = httpx.AsyncClient(
            auth=(_config.hub_access_key_id, _config.hub_access_key_secret),
            verify=False,
        )
        self.cache = _AsyncServerCache(self._http)

    def __getitem__(self, url) -> AsyncPage:
        return AsyncPage(self, url)

    def __delitem__(self, key: str):
        page = self[key]
        page.drop()

    async def _save(self, url: str, patch: str):
        res = await self._http.patch(_rebase(_config.hub_address, url), content=patch)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')

    async def load(self, url) -> dict:
        """
        Retrieve data at the given URL, typically the serialized form of a page.

        Args:
            url: The URL to read.

        Returns:
            The serialized page.
        """
        res = await self._http.get(_rebase(_config.hub_address, url), headers=_content_type_json)
        if res.status_code != 200:
            raise ServiceError(f'Request failed (code={res.status_code}): {res.text}')
        return res.json()

    async def upload_dir(self, directory: str) -> str:
        """
        WARNING: Experimental and subject to change.
        Upload whole directory to the site with directory structure preserved.

        Args:
            directory: Folder to be uploaded.

        Returns:
            A list of remote URLs for the uploaded directory (always size of 1).
        """
        if not os.path.isdir(directory):
            raise ValueError(f'{directory} is not a directory.')

        waved_dir = _get_env('WAVED_DIR', None)
        data_dir = _get_env('DATA_DIR', 'data')
        skip_local_upload = _get_env('NO_COPY_UPLOAD', 'false').lower() in ['true', '1', 't']

        # If we know the path of waved and running app on the same machine,
        # we can simply copy the files instead of making an HTTP request.
        if _is_loopback_address() and not skip_local_upload and waved_dir and data_dir:
            try:
                uuid = str(uuid4())
                dst = os.path.join(waved_dir, data_dir, 'f', uuid)
                os.makedirs(dst, exist_ok=True)

                if 'Windows' in platform.system():
                    args = ['robocopy', directory, dst, '/S', '/J', '/W:0', '*.*']
                else:
                    args = ['rsync', '-a', os.path.join(directory, '.'), dst]

                return [await _copy_in_subprocess(args, uuid)]
            except:
                pass

        upload_files = []
        for f in _get_files_in_directory(directory, []):
            upload_files.append(('files', (os.path.relpath(f, directory), open(f, 'rb'))))

        res = await self._http.post(f'{_config.hub_address}_f/', headers={'Wave-Directory-Upload': "True"}, files=upload_files)

        for _, f in upload_files:
            f[1].close()

        if res.status_code == 200:
            return json.loads(res.text)['files']
        raise ServiceError(f'Upload failed (code={res.status_code}): {res.text}')

    async def upload(self, files: List[str]) -> List[str]:
        """
        Upload local files to the site.

        Args:
            files: A list of file paths of the files to be uploaded.

        Returns:
            A list of remote URLs for the uploaded files, in order.
        """
        for f in files:
            if not os.path.isfile(f):
                raise ValueError(f'{f} is not a file.')

        waved_dir = _get_env('WAVED_DIR', None)
        data_dir = _get_env('DATA_DIR', 'data')
        skip_local_upload = _get_env('NO_COPY_UPLOAD', 'false').lower() in ['true', '1', 't']

        # If we know the path of waved and running app on the same machine,
        # we can simply copy the files instead of making an HTTP request.
        if _is_loopback_address() and not skip_local_upload and waved_dir and data_dir:
            try:
                tasks = []
                for f in files:
                    uuid = str(uuid4())
                    dst = os.path.join(waved_dir, data_dir, 'f', uuid)
                    os.makedirs(dst, exist_ok=True)

                    if 'Windows' in platform.system():
                        src = os.path.dirname(f) or os.getcwd()
                        args = ['robocopy', src, dst, os.path.basename(f), '/J', '/W:0']
                    else:
                        args = ['cp', f, dst]

                    tasks.append(asyncio.create_task(_copy_in_subprocess(args, uuid, f)))

                return await asyncio.gather(*tasks)
            except:
                pass

        upload_files = []
        file_handles: List[BufferedReader] = []
        for f in files:
            file_handle = open(f, 'rb')
            upload_files.append(('files', (os.path.basename(f), file_handle)))
            file_handles.append(file_handle)

        res = await self._http.post(f'{_config.hub_address}_f/', files=upload_files)

        for h in file_handles:
            h.close()

        if res.status_code == 200:
            return json.loads(res.text)['files']
        raise ServiceError(f'Upload failed (code={res.status_code}): {res.text}')

    async def download(self, url: str, path: str) -> str:
        """
        Download a file from the site.

        Args:
            url: The URL of the file.
            path: The local directory or file path to download to. If a directory is provided, the original name of the file is retained.
        Returns:
            The path to the downloaded file.
        """
        path = os.path.abspath(path)
        # If path is a directory, get basename from url
        filepath = os.path.join(path, os.path.basename(url)) if os.path.isdir(path) else path

        async with self._http.stream('GET', f'{_config.hub_host_address}{url}') as res:
            if res.status_code != 200:
                await res.aread()
                raise ServiceError(f'Download failed (code={res.status_code}): {res.text}')
            with open(filepath, 'wb') as f:
                async for chunk in res.aiter_bytes():
                    f.write(chunk)

        return filepath

    async def unload(self, url: str):
        """
        Delete an uploaded file from the site.

        Args:
            url: The URL of the file to delete.
        """
        res = await self._http.delete(f'{_config.hub_host_address}{url}')
        if res.status_code == 200:
            return
        raise ServiceError(f'Unload failed (code={res.status_code}): {res.text}')

    async def uplink(self, path: str, content_type: str, file: FileContent) -> str:
        """
        Create or update a stream of images.

        The typical use of this function is to transmit a stream of images to a web page, providing the appearance of a
         video. The path returned by this function can be passed to ui.image() or ui.image_card() (or even a custom HTML
         img element). The stream is displayed in browsers using multipart/x-mixed-replace content. Callers must call
         the unlink() function to signal end-of-stream.

        Args:
            path: a unique path or name for the stream (e.g. 'foo/bar/qux.png'). Must be a valid URL path.
            content_type: The MIME type of the streamed content (e.g. 'image/jpeg', 'image/png', etc.).
            file: A file or file-like object (on-disk file, standard I/O, in-memory buffers, sockets or pipes).

        Returns:
            The stream endpoint, typically used as an image path.
        """
        endpoint = f'_m/{path}'
        res = await self._http.post(f'{_config.hub_address}{endpoint}', files={'f': ('f', file, content_type)})
        if res.status_code == 200:
            return _config.hub_base_url + endpoint
        raise ServiceError(f'Uplink failed (code={res.status_code}): {res.text}')

    async def unlink(self, path: str):
        """
        Signal the end of a stream.

        Args:
            path: The path of the stream
        """
        endpoint = f'_m/{path}'
        res = await self._http.delete(f'{_config.hub_address}{endpoint}')
        if res.status_code == 200:
            return
        raise ServiceError(f'Unlink failed (code={res.status_code}): {res.text}')

    async def proxy(self, method: str, url: str, headers: Optional[Dict[str, List[str]]] = None,
                    body: Optional[str] = None):
        req = dict(method=method, url=url, headers=headers, body=body)
        res = await self._http.post(f'{_config.hub_address}_p/', content=marshal(req))
        if res.status_code == 200:
            return res.json()
        raise ServiceError(f'Proxy request failed (code={res.status_code}): {res.text}')


async def _copy_in_subprocess(args: List[str], uuid: str, f='') -> str:
    p = await asyncio.create_subprocess_exec(*args, stderr=subprocess.PIPE, stdout=subprocess.DEVNULL)
    _, err = await p.communicate()

    if err:
        raise ValueError(err.decode())

    if f:
        return f'{_base_url}_f/{uuid}/{os.path.basename(f)}'
    else:
        return f'{_base_url}_f/{uuid}'


def _get_files_in_directory(directory: str, files: List[str]) -> List[str]:
    for f in os.listdir(directory):
        path = os.path.join(directory, f)
        if os.path.isfile(path):
            files.append(path)
        else:
            _get_files_in_directory(path, files)
    return files


def marshal(d: Any) -> str:
    """
    Marshal to JSON.

    Args:
        d: Any object or value.

    Returns:
        A string containing the JSON-serialized form.
    """
    return json.dumps(d, allow_nan=False, separators=(',', ':'))


def unmarshal(s: str) -> Any:
    """
    Unmarshal a JSON string.

    Args:
        s: A string containing JSON-serialized data.

    Returns:
        The deserialized object or value.
    """
    return json.loads(s)


def pack(data: Any) -> str:
    """
    Pack (compress) the provided value.

    Args:
        data: Any object or value.

    The object or value compressed into a string.
    """
    return 'data:' + marshal(_dump(data))


def _is_loopback_address() -> bool:
    try:
        hostname = urlparse(_config.hub_address).hostname
        return ipaddress.ip_address(hostname).is_loopback
    except ValueError:
        return False