2025-12-25 upload
This commit is contained in:
133
venv/Lib/site-packages/h2/windows.py
Normal file
133
venv/Lib/site-packages/h2/windows.py
Normal file
@@ -0,0 +1,133 @@
|
||||
"""
|
||||
h2/windows
|
||||
~~~~~~~~~~
|
||||
|
||||
Defines tools for managing HTTP/2 flow control windows.
|
||||
|
||||
The objects defined in this module are used to automatically manage HTTP/2
|
||||
flow control windows. Specifically, they keep track of what the size of the
|
||||
window is, how much data has been consumed from that window, and how much data
|
||||
the user has already used. It then implements a basic algorithm that attempts
|
||||
to manage the flow control window without user input, trying to ensure that it
|
||||
does not emit too many WINDOW_UPDATE frames.
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from .exceptions import FlowControlError
|
||||
|
||||
# The largest acceptable value for a HTTP/2 flow control window.
|
||||
LARGEST_FLOW_CONTROL_WINDOW = 2**31 - 1
|
||||
|
||||
|
||||
class WindowManager:
|
||||
"""
|
||||
A basic HTTP/2 window manager.
|
||||
|
||||
:param max_window_size: The maximum size of the flow control window.
|
||||
:type max_window_size: ``int``
|
||||
"""
|
||||
|
||||
def __init__(self, max_window_size: int) -> None:
|
||||
assert max_window_size <= LARGEST_FLOW_CONTROL_WINDOW
|
||||
self.max_window_size = max_window_size
|
||||
self.current_window_size = max_window_size
|
||||
self._bytes_processed = 0
|
||||
|
||||
def window_consumed(self, size: int) -> None:
|
||||
"""
|
||||
We have received a certain number of bytes from the remote peer. This
|
||||
necessarily shrinks the flow control window!
|
||||
|
||||
:param size: The number of flow controlled bytes we received from the
|
||||
remote peer.
|
||||
:type size: ``int``
|
||||
:returns: Nothing.
|
||||
:rtype: ``None``
|
||||
"""
|
||||
self.current_window_size -= size
|
||||
if self.current_window_size < 0:
|
||||
msg = "Flow control window shrunk below 0"
|
||||
raise FlowControlError(msg)
|
||||
|
||||
def window_opened(self, size: int) -> None:
|
||||
"""
|
||||
The flow control window has been incremented, either because of manual
|
||||
flow control management or because of the user changing the flow
|
||||
control settings. This can have the effect of increasing what we
|
||||
consider to be the "maximum" flow control window size.
|
||||
|
||||
This does not increase our view of how many bytes have been processed,
|
||||
only of how much space is in the window.
|
||||
|
||||
:param size: The increment to the flow control window we received.
|
||||
:type size: ``int``
|
||||
:returns: Nothing
|
||||
:rtype: ``None``
|
||||
"""
|
||||
self.current_window_size += size
|
||||
|
||||
if self.current_window_size > LARGEST_FLOW_CONTROL_WINDOW:
|
||||
msg = f"Flow control window mustn't exceed {LARGEST_FLOW_CONTROL_WINDOW}"
|
||||
raise FlowControlError(msg)
|
||||
|
||||
self.max_window_size = max(self.current_window_size, self.max_window_size)
|
||||
|
||||
def process_bytes(self, size: int) -> int | None:
|
||||
"""
|
||||
The application has informed us that it has processed a certain number
|
||||
of bytes. This may cause us to want to emit a window update frame. If
|
||||
we do want to emit a window update frame, this method will return the
|
||||
number of bytes that we should increment the window by.
|
||||
|
||||
:param size: The number of flow controlled bytes that the application
|
||||
has processed.
|
||||
:type size: ``int``
|
||||
:returns: The number of bytes to increment the flow control window by,
|
||||
or ``None``.
|
||||
:rtype: ``int`` or ``None``
|
||||
"""
|
||||
self._bytes_processed += size
|
||||
return self._maybe_update_window()
|
||||
|
||||
def _maybe_update_window(self) -> int | None:
|
||||
"""
|
||||
Run the algorithm.
|
||||
|
||||
Our current algorithm can be described like this.
|
||||
|
||||
1. If no bytes have been processed, we immediately return 0. There is
|
||||
no meaningful way for us to hand space in the window back to the
|
||||
remote peer, so let's not even try.
|
||||
2. If there is no space in the flow control window, and we have
|
||||
processed at least 1024 bytes (or 1/4 of the window, if the window
|
||||
is smaller), we will emit a window update frame. This is to avoid
|
||||
the risk of blocking a stream altogether.
|
||||
3. If there is space in the flow control window, and we have processed
|
||||
at least 1/2 of the window worth of bytes, we will emit a window
|
||||
update frame. This is to minimise the number of window update frames
|
||||
we have to emit.
|
||||
|
||||
In a healthy system with large flow control windows, this will
|
||||
irregularly emit WINDOW_UPDATE frames. This prevents us starving the
|
||||
connection by emitting eleventy bajillion WINDOW_UPDATE frames,
|
||||
especially in situations where the remote peer is sending a lot of very
|
||||
small DATA frames.
|
||||
"""
|
||||
# TODO: Can the window be smaller than 1024 bytes? If not, we can
|
||||
# streamline this algorithm.
|
||||
if not self._bytes_processed:
|
||||
return None
|
||||
|
||||
max_increment = (self.max_window_size - self.current_window_size)
|
||||
increment = 0
|
||||
|
||||
# Note that, even though we may increment less than _bytes_processed,
|
||||
# we still want to set it to zero whenever we emit an increment. This
|
||||
# is because we'll always increment up to the maximum we can.
|
||||
if ((self.current_window_size == 0) and (
|
||||
self._bytes_processed > min(1024, self.max_window_size // 4))) or self._bytes_processed >= (self.max_window_size // 2):
|
||||
increment = min(self._bytes_processed, max_increment)
|
||||
self._bytes_processed = 0
|
||||
|
||||
self.current_window_size += increment
|
||||
return increment
|
||||
Reference in New Issue
Block a user