blob: 1b1326d6ddd36dc248b62ed4b856b340a83fcd82 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
|
.. _pipeline-chapter:
=========
Pipelines
=========
The :py:class:`Pipeline` class is a subclass of :py:class:`StatsClient` that
batches together several stats before sending. It implements the entire client
interface, plus a :py:meth:`send() <Pipeline.send()>` method.
:py:class:`Pipeline` objects should be created with
:py:meth:`StatsClient.pipeline()`:
.. code-block:: python
client = StatsClient()
pipe = client.pipeline()
pipe.incr('foo')
pipe.decr('bar')
pipe.timing('baz', 520)
pipe.send()
No stats will be sent until :py:meth:`send() <Pipeline.send()>` is called, at
which point they will be packed into as few UDP packets as possible.
As a Context Manager
====================
:py:class:`Pipeline` objects can also be used as context managers:
.. code-block:: python
with StatsClient().pipeline() as pipe:
pipe.incr('foo')
pipe.decr('bar')
:py:meth:`Pipeline.send()` will be called automatically when the managed block
exits.
Thread Safety
=============
While :py:class:`StatsClient` instances are considered thread-safe (or at least
as thread-safe as the standard library's ``socket.send`` is),
:py:class:`Pipeline` instances **are not thread-safe**. Storing stats for later
creates at least two important race conditions in a multi-threaded environment.
You should create one :py:class:`Pipeline` per-thread, if necessary.
|
