Yury Selivanov | 7c7605f | 2018-09-11 09:54:40 -0700 | [diff] [blame^] | 1 | .. currentmodule:: asyncio |
| 2 | |
| 3 | |
| 4 | .. _asyncio-policies: |
| 5 | |
| 6 | ======== |
| 7 | Policies |
| 8 | ======== |
| 9 | |
| 10 | An event loop policy, a global per-process object, controls |
| 11 | management of the event loop. Each event loop has a default |
| 12 | policy, which can be changed and customized using the API. |
| 13 | |
| 14 | A policy defines the notion of context and manages a |
| 15 | separate event loop per context. The default policy |
| 16 | defines context to be the current thread. |
| 17 | |
| 18 | By using a custom event loop policy, the behavior of |
| 19 | :func:`get_event_loop`, :func:`set_event_loop`, and |
| 20 | :func:`new_event_loop` functions can be customized. |
| 21 | |
| 22 | Policy objects should implement the APIs defined |
| 23 | in the abstract base class :class:`AbstractEventLoopPolicy`. |
| 24 | |
| 25 | |
| 26 | Access the Policy |
| 27 | ================= |
| 28 | |
| 29 | The following functions can be used to get and set the policy |
| 30 | for the current process: |
| 31 | |
| 32 | .. function:: get_event_loop_policy() |
| 33 | |
| 34 | Return the current process-wide policy. |
| 35 | |
| 36 | .. function:: set_event_loop_policy(policy) |
| 37 | |
| 38 | Set the current process-wide policy to *policy*. |
| 39 | |
| 40 | If *policy* is set to ``None``, the default policy is restored. |
| 41 | |
| 42 | |
| 43 | Policy Objects |
| 44 | ============== |
| 45 | |
| 46 | The abstract event loop policy base class is defined as follows: |
| 47 | |
| 48 | .. class:: AbstractEventLoopPolicy |
| 49 | |
| 50 | An abstract base class for asyncio policies. |
| 51 | |
| 52 | .. method:: get_event_loop() |
| 53 | |
| 54 | Get the event loop for the current context. |
| 55 | |
| 56 | Return an event loop object implementing the |
| 57 | :class:`AbstractEventLoop` interface. |
| 58 | |
| 59 | This method should never return ``None``. |
| 60 | |
| 61 | .. versionchanged:: 3.6 |
| 62 | |
| 63 | .. method:: set_event_loop(loop) |
| 64 | |
| 65 | Set the event loop for the current context to *loop*. |
| 66 | |
| 67 | .. method:: new_event_loop() |
| 68 | |
| 69 | Create and return a new event loop object. |
| 70 | |
| 71 | This method should never return ``None``. |
| 72 | |
| 73 | .. method:: get_child_watcher() |
| 74 | |
| 75 | Get a child process watcher object. |
| 76 | |
| 77 | Return a watcher object implementing the |
| 78 | :class:`AbstractChildWatcher` interface. |
| 79 | |
| 80 | This function is Unix specific. |
| 81 | |
| 82 | .. method:: set_child_watcher(watcher) |
| 83 | |
| 84 | Get the current child process watcher to *watcher*. |
| 85 | |
| 86 | This function is Unix specific. |
| 87 | |
| 88 | |
| 89 | asyncio ships with the following built-in policies: |
| 90 | |
| 91 | |
| 92 | .. class:: DefaultEventLoopPolicy |
| 93 | |
| 94 | The default asyncio policy. Uses :class:`SelectorEventLoop` |
| 95 | on both Unix and Windows platforms. |
| 96 | |
| 97 | There is no need to install the default policy manually; asyncio |
| 98 | is configured to use it automatically. |
| 99 | |
| 100 | |
| 101 | .. class:: WindowsProactorEventLoopPolicy |
| 102 | |
| 103 | An alternative event loop policy that uses the |
| 104 | :class:`ProactorEventLoop` event loop implementation. |
| 105 | |
| 106 | Availability: Windows. |
| 107 | |
| 108 | |
| 109 | Process Watchers |
| 110 | ================ |
| 111 | |
| 112 | A process watcher allows customization of how an event loop monitors |
| 113 | child processes on Unix. Specifically, the event loop needs to know |
| 114 | when a child process has finished its execution. |
| 115 | |
| 116 | In asyncio, child processes are created with |
| 117 | :func:`create_subprocess_exec` and :meth:`loop.subprocess_exec` |
| 118 | functions. |
| 119 | |
| 120 | asyncio defines an abstract base class :class:`AbstractChildWatcher` |
| 121 | that child watchers should implement, and has two different |
| 122 | implementations: :class:`SafeChildWatcher` (configured to be used |
| 123 | by default) and :class:`FastChildWatcher`. |
| 124 | |
| 125 | See also the :ref:`Subprocess and Threads <asyncio-subprocess-threads>` |
| 126 | section. |
| 127 | |
| 128 | The following two functions can be used to customize the watcher |
| 129 | implementation used by the asyncio event loop: |
| 130 | |
| 131 | .. function:: get_child_watcher() |
| 132 | |
| 133 | Return the current child watcher for the current policy. |
| 134 | |
| 135 | .. function:: set_child_watcher(watcher) |
| 136 | |
| 137 | Set the current child watcher to *watcher* for the current |
| 138 | policy. *watcher* must implement methods defined in the |
| 139 | :class:`AbstractChildWatcher` base class. |
| 140 | |
| 141 | .. note:: |
| 142 | Third-party event loops implementations might not support |
| 143 | custom child watchers. For such event loops, using |
| 144 | :func:`set_child_watcher` might have no effect or even can |
| 145 | be prohibited. |
| 146 | |
| 147 | .. class:: AbstractChildWatcher |
| 148 | |
| 149 | .. method:: add_child_handler(pid, callback, \*args) |
| 150 | |
| 151 | Register a new child handler. |
| 152 | |
| 153 | Arrange for ``callback(pid, returncode, *args)`` to be called |
| 154 | when a process with PID equal to *pid* terminates. Specifying |
| 155 | another callback for the same process replaces the previous |
| 156 | handler. |
| 157 | |
| 158 | *callback* callable must be thread-safe. |
| 159 | |
| 160 | .. method:: remove_child_handler(pid) |
| 161 | |
| 162 | Removes the handler for process with PID equal to *pid*. |
| 163 | |
| 164 | The function returns ``True`` if the handler was successfully |
| 165 | removed, ``False`` if there was nothing to remove. |
| 166 | |
| 167 | .. method:: attach_loop(loop) |
| 168 | |
| 169 | Attach the watcher to an event loop. |
| 170 | |
| 171 | If the watcher was previously attached to an event loop, then |
| 172 | it is first detached before attaching to the new loop. |
| 173 | |
| 174 | Note: loop may be ``None``. |
| 175 | |
| 176 | .. method:: close() |
| 177 | |
| 178 | Close the watcher. |
| 179 | |
| 180 | This method has to be called to ensure that underlying |
| 181 | resources are cleaned-up. |
| 182 | |
| 183 | .. class:: SafeChildWatcher |
| 184 | |
| 185 | This implementation avoids disrupting other code spawning processes |
| 186 | by polling every process explicitly on a :py:data:`SIGCHLD` signal. |
| 187 | |
| 188 | This is a safe solution but it has a significant overhead when |
| 189 | handling a big number of processes (*O(n)* each time a |
| 190 | :py:data:`SIGCHLD` is received). |
| 191 | |
| 192 | asyncio uses this implementation by default. |
| 193 | |
| 194 | .. class:: FastChildWatcher |
| 195 | |
| 196 | This implementation reaps every terminated processes by calling |
| 197 | ``os.waitpid(-1)`` directly, possibly breaking other code spawning |
| 198 | processes and waiting for their termination. |
| 199 | |
| 200 | There is no noticeable overhead when handling a big number of |
| 201 | children (*O(1)* each time a child terminates). |
| 202 | |
| 203 | |
| 204 | Custom Policies |
| 205 | =============== |
| 206 | |
| 207 | To implement a new event loop policy, it is recommended to subclass |
| 208 | :class:`DefaultEventLoopPolicy` and override the methods for which |
| 209 | custom behavior is wanted, e.g.:: |
| 210 | |
| 211 | class MyEventLoopPolicy(asyncio.DefaultEventLoopPolicy): |
| 212 | |
| 213 | def get_event_loop(self): |
| 214 | """Get the event loop. |
| 215 | |
| 216 | This may be None or an instance of EventLoop. |
| 217 | """ |
| 218 | loop = super().get_event_loop() |
| 219 | # Do something with loop ... |
| 220 | return loop |
| 221 | |
| 222 | asyncio.set_event_loop_policy(MyEventLoopPolicy()) |