File: agent.rst

package info (click to toggle)
pytorch-cuda 2.6.0%2Bdfsg-7
  • links: PTS, VCS
  • area: contrib
  • in suites: forky, sid, trixie
  • size: 161,620 kB
  • sloc: python: 1,278,832; cpp: 900,322; ansic: 82,710; asm: 7,754; java: 3,363; sh: 2,811; javascript: 2,443; makefile: 597; ruby: 195; xml: 84; objc: 68
file content (94 lines) | stat: -rw-r--r-- 2,810 bytes parent folder | download | duplicates (3)
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
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
Elastic Agent
==============

.. automodule:: torch.distributed.elastic.agent
.. currentmodule:: torch.distributed.elastic.agent

Server
--------

.. automodule:: torch.distributed.elastic.agent.server

Below is a diagram of an agent that manages a local group of workers.

.. image:: agent_diagram.jpg

Concepts
--------

This section describes the high-level classes and concepts that
are relevant to understanding the role of the ``agent`` in torchelastic.

.. currentmodule:: torch.distributed.elastic.agent.server

.. autoclass:: ElasticAgent
   :members:

.. autoclass:: WorkerSpec
   :members:

.. autoclass:: WorkerState
   :members:

.. autoclass:: Worker
   :members:

.. autoclass:: WorkerGroup
   :members:

Implementations
-------------------

Below are the agent implementations provided by torchelastic.

.. currentmodule:: torch.distributed.elastic.agent.server.local_elastic_agent
.. autoclass:: LocalElasticAgent


Extending the Agent
---------------------

To extend the agent you can implement ```ElasticAgent`` directly, however
we recommend you extend ``SimpleElasticAgent`` instead, which provides
most of the scaffolding and leaves you with a few specific abstract methods
to implement.

.. currentmodule:: torch.distributed.elastic.agent.server
.. autoclass:: SimpleElasticAgent
   :members:
   :private-members:

.. autoclass:: torch.distributed.elastic.agent.server.api.RunResult


Watchdog in the Agent
---------------------

A named pipe based watchdog can be enabled in ```LocalElasticAgent``` if an
environment variable ``TORCHELASTIC_ENABLE_FILE_TIMER`` with value 1 has
been defined in the ```LocalElasticAgent``` process.
Optionally, another environment variable ```TORCHELASTIC_TIMER_FILE```
can be set with a unique file name for the named pipe. If the environment
variable ```TORCHELASTIC_TIMER_FILE``` is not set, ```LocalElasticAgent```
will internally create a unique file name and set it to the environment
variable ```TORCHELASTIC_TIMER_FILE```, and this environment variable will
be propagated to the worker processes to allow them to connect to the same
named pipe that ```LocalElasticAgent``` uses.


Health Check Server
-------------------

A health check monitoring server can be enabled in ```LocalElasticAgent```
if an environment variable ``TORCHELASTIC_HEALTH_CHECK_PORT`` has been defined
in the ```LocalElasticAgent``` process.
Adding interface for health check server which can be extended by starting tcp/http
server on the specified port number.
Additionally, health check server will have callback to check watchdog is alive.

.. automodule:: torch.distributed.elastic.agent.server.health_check_server

.. autoclass:: HealthCheckServer
   :members:

.. autofunction:: torch.distributed.elastic.agent.server.health_check_server.create_healthcheck_server