1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
# |
3
|
|
|
# This file is part of Glances. |
4
|
|
|
# |
5
|
|
|
# Copyright (C) 2020 Kirby Banman <[email protected]> |
6
|
|
|
# Copyright (C) 2024 Nicolas Hennion <[email protected]> |
7
|
|
|
# |
8
|
|
|
# SPDX-License-Identifier: LGPL-3.0-only |
9
|
|
|
# |
10
|
|
|
|
11
|
|
|
"""GPU plugin for Glances. |
12
|
|
|
|
13
|
|
|
Currently supported: |
14
|
|
|
- NVIDIA GPU (need pynvml lib) |
15
|
|
|
- AMD GPU (no lib needed) |
16
|
|
|
""" |
17
|
|
|
|
18
|
|
|
from glances.globals import to_fahrenheit |
19
|
|
|
from glances.plugins.gpu.cards.nvidia import NvidiaGPU |
20
|
|
|
from glances.plugins.gpu.cards.amd import AmdGPU |
21
|
|
|
from glances.plugins.plugin.model import GlancesPluginModel |
22
|
|
|
|
23
|
|
|
# Fields description |
24
|
|
|
# description: human readable description |
25
|
|
|
# short_name: shortname to use un UI |
26
|
|
|
# unit: unit type |
27
|
|
|
# rate: is it a rate ? If yes, // by time_since_update when displayed, |
28
|
|
|
# min_symbol: Auto unit should be used if value > than 1 'X' (K, M, G)... |
29
|
|
|
fields_description = { |
30
|
|
|
'gpu_id': { |
31
|
|
|
'description': 'GPU identification', |
32
|
|
|
}, |
33
|
|
|
'name': { |
34
|
|
|
'description': 'GPU name', |
35
|
|
|
}, |
36
|
|
|
'mem': { |
37
|
|
|
'description': 'Memory consumption', |
38
|
|
|
'unit': 'percent', |
39
|
|
|
}, |
40
|
|
|
'proc': { |
41
|
|
|
'description': 'GPU processor consumption', |
42
|
|
|
'unit': 'percent', |
43
|
|
|
}, |
44
|
|
|
'temperature': { |
45
|
|
|
'description': 'GPU temperature', |
46
|
|
|
'unit': 'celsius', |
47
|
|
|
}, |
48
|
|
|
'fan_speed': { |
49
|
|
|
'description': 'GPU fan speed', |
50
|
|
|
'unit': 'roundperminute', |
51
|
|
|
}, |
52
|
|
|
} |
53
|
|
|
|
54
|
|
|
# Define the history items list |
55
|
|
|
# All items in this list will be historised if the --enable-history tag is set |
56
|
|
|
items_history_list = [ |
57
|
|
|
{'name': 'proc', 'description': 'GPU processor', 'y_unit': '%'}, |
58
|
|
|
{'name': 'mem', 'description': 'Memory consumption', 'y_unit': '%'}, |
59
|
|
|
] |
60
|
|
|
|
61
|
|
|
|
62
|
|
|
class PluginModel(GlancesPluginModel): |
63
|
|
|
"""Glances GPU plugin. |
64
|
|
|
|
65
|
|
|
stats is a list of dictionaries with one entry per GPU |
66
|
|
|
""" |
67
|
|
|
|
68
|
|
|
def __init__(self, args=None, config=None): |
69
|
|
|
"""Init the plugin.""" |
70
|
|
|
super(PluginModel, self).__init__( |
71
|
|
|
args=args, |
72
|
|
|
config=config, |
73
|
|
|
items_history_list=items_history_list, |
74
|
|
|
stats_init_value=[], |
75
|
|
|
fields_description=fields_description, |
76
|
|
|
) |
77
|
|
|
# Init the GPU API |
78
|
|
|
self.nvidia = NvidiaGPU() |
79
|
|
|
self.amd = AmdGPU() |
80
|
|
|
# Just for test purpose (uncomment to test on computer without AMD GPU) |
81
|
|
|
# self.amd = AmdGPU(drm_root_folder='./test-data/plugins/gpu/amd/sys/class/drm') |
82
|
|
|
|
83
|
|
|
# We want to display the stat in the curse interface |
84
|
|
|
self.display_curse = True |
85
|
|
|
|
86
|
|
|
def exit(self): |
87
|
|
|
"""Overwrite the exit method to close the GPU API.""" |
88
|
|
|
self.nvidia.exit() |
89
|
|
|
self.amd.exit() |
90
|
|
|
|
91
|
|
|
# Call the father exit method |
92
|
|
|
super(PluginModel, self).exit() |
93
|
|
|
|
94
|
|
|
def get_key(self): |
95
|
|
|
"""Return the key of the list.""" |
96
|
|
|
return 'gpu_id' |
97
|
|
|
|
98
|
|
|
@GlancesPluginModel._check_decorator |
99
|
|
|
@GlancesPluginModel._log_result_decorator |
100
|
|
|
def update(self): |
101
|
|
|
"""Update the GPU stats.""" |
102
|
|
|
# Init new stats |
103
|
|
|
stats = self.get_init_value() |
104
|
|
|
|
105
|
|
|
# Get the stats |
106
|
|
|
stats.extend(self.nvidia.get_device_stats()) |
107
|
|
|
stats.extend(self.amd.get_device_stats()) |
108
|
|
|
|
109
|
|
|
# !!! |
110
|
|
|
# Uncomment to test on computer without GPU |
111
|
|
|
# One GPU sample: |
112
|
|
|
# stats = [ |
113
|
|
|
# { |
114
|
|
|
# "key": "gpu_id", |
115
|
|
|
# "gpu_id": "nvidia0", |
116
|
|
|
# "name": "Fake GeForce GTX", |
117
|
|
|
# "mem": 5.792331695556641, |
118
|
|
|
# "proc": 4, |
119
|
|
|
# "temperature": 26, |
120
|
|
|
# "fan_speed": 30 |
121
|
|
|
# } |
122
|
|
|
# ] |
123
|
|
|
# Two GPU sample: |
124
|
|
|
# stats = [ |
125
|
|
|
# { |
126
|
|
|
# "key": "gpu_id", |
127
|
|
|
# "gpu_id": "nvidia0", |
128
|
|
|
# "name": "Fake GeForce GTX1", |
129
|
|
|
# "mem": 5.792331695556641, |
130
|
|
|
# "proc": 4, |
131
|
|
|
# "temperature": 26, |
132
|
|
|
# "fan_speed": 30 |
133
|
|
|
# }, |
134
|
|
|
# { |
135
|
|
|
# "key": "gpu_id", |
136
|
|
|
# "gpu_id": "nvidia1", |
137
|
|
|
# "name": "Fake GeForce GTX1", |
138
|
|
|
# "mem": 15, |
139
|
|
|
# "proc": 8, |
140
|
|
|
# "temperature": 65, |
141
|
|
|
# "fan_speed": 75 |
142
|
|
|
# } |
143
|
|
|
# ] |
144
|
|
|
|
145
|
|
|
# Update the stats |
146
|
|
|
self.stats = stats |
147
|
|
|
|
148
|
|
|
return self.stats |
149
|
|
|
|
150
|
|
|
def update_views(self): |
151
|
|
|
"""Update stats views.""" |
152
|
|
|
# Call the father's method |
153
|
|
|
super(PluginModel, self).update_views() |
154
|
|
|
|
155
|
|
|
# Add specifics information |
156
|
|
|
# Alert |
157
|
|
|
for i in self.stats: |
158
|
|
|
# Init the views for the current GPU |
159
|
|
|
self.views[i[self.get_key()]] = {'proc': {}, 'mem': {}, 'temperature': {}} |
160
|
|
|
# Processor alert |
161
|
|
|
if 'proc' in i: |
162
|
|
|
alert = self.get_alert(i['proc'], header='proc') |
163
|
|
|
self.views[i[self.get_key()]]['proc']['decoration'] = alert |
164
|
|
|
# Memory alert |
165
|
|
|
if 'mem' in i: |
166
|
|
|
alert = self.get_alert(i['mem'], header='mem') |
167
|
|
|
self.views[i[self.get_key()]]['mem']['decoration'] = alert |
168
|
|
|
# Temperature alert |
169
|
|
|
if 'temperature' in i: |
170
|
|
|
alert = self.get_alert(i['temperature'], header='temperature') |
171
|
|
|
self.views[i[self.get_key()]]['temperature']['decoration'] = alert |
172
|
|
|
|
173
|
|
|
return True |
174
|
|
|
|
175
|
|
|
def msg_curse(self, args=None, max_width=None): |
176
|
|
|
"""Return the dict to display in the curse interface.""" |
177
|
|
|
# Init the return message |
178
|
|
|
ret = [] |
179
|
|
|
|
180
|
|
|
# Only process if stats exist, not empty (issue #871) and plugin not disabled |
181
|
|
|
if not self.stats or (self.stats == []) or self.is_disabled(): |
182
|
|
|
return ret |
183
|
|
|
|
184
|
|
|
# Check if all GPU have the same name |
185
|
|
|
same_name = all(s['name'] == self.stats[0]['name'] for s in self.stats) |
186
|
|
|
|
187
|
|
|
# gpu_stats contain the first GPU in the list |
188
|
|
|
gpu_stats = self.stats[0] |
189
|
|
|
|
190
|
|
|
# Header |
191
|
|
|
header = '' |
192
|
|
|
if len(self.stats) > 1: |
193
|
|
|
header += '{}'.format(len(self.stats)) |
194
|
|
|
if same_name: |
195
|
|
|
header += ' {}'.format(gpu_stats['name']) |
196
|
|
|
else: |
197
|
|
|
header += ' GPUs' |
198
|
|
|
elif same_name: |
199
|
|
|
header += '{}'.format(gpu_stats['name']) |
200
|
|
|
else: |
201
|
|
|
header += 'GPU' |
202
|
|
|
msg = header[:17] |
203
|
|
|
ret.append(self.curse_add_line(msg, "TITLE")) |
204
|
|
|
|
205
|
|
|
# Build the string message |
206
|
|
|
if len(self.stats) == 1 or args.meangpu: |
207
|
|
|
# GPU stat summary or mono GPU |
208
|
|
|
# New line |
209
|
|
|
ret.append(self.curse_new_line()) |
210
|
|
|
# GPU PROC |
211
|
|
|
try: |
212
|
|
|
mean_proc = sum(s['proc'] for s in self.stats if s is not None) / len(self.stats) |
213
|
|
|
except TypeError: |
214
|
|
|
mean_proc_msg = '{:>4}'.format('N/A') |
215
|
|
|
else: |
216
|
|
|
mean_proc_msg = '{:>3.0f}%'.format(mean_proc) |
217
|
|
|
if len(self.stats) > 1: |
218
|
|
|
msg = '{:13}'.format('proc mean:') |
219
|
|
|
else: |
220
|
|
|
msg = '{:13}'.format('proc:') |
221
|
|
|
ret.append(self.curse_add_line(msg)) |
222
|
|
|
ret.append( |
223
|
|
|
self.curse_add_line( |
224
|
|
|
mean_proc_msg, self.get_views(item=gpu_stats[self.get_key()], key='proc', option='decoration') |
225
|
|
|
) |
226
|
|
|
) |
227
|
|
|
# New line |
228
|
|
|
ret.append(self.curse_new_line()) |
229
|
|
|
# GPU MEM |
230
|
|
|
try: |
231
|
|
|
mean_mem = sum(s['mem'] for s in self.stats if s is not None) / len(self.stats) |
232
|
|
|
except TypeError: |
233
|
|
|
mean_mem_msg = '{:>4}'.format('N/A') |
234
|
|
|
else: |
235
|
|
|
mean_mem_msg = '{:>3.0f}%'.format(mean_mem) |
236
|
|
|
if len(self.stats) > 1: |
237
|
|
|
msg = '{:13}'.format('mem mean:') |
238
|
|
|
else: |
239
|
|
|
msg = '{:13}'.format('mem:') |
240
|
|
|
ret.append(self.curse_add_line(msg)) |
241
|
|
|
ret.append( |
242
|
|
|
self.curse_add_line( |
243
|
|
|
mean_mem_msg, self.get_views(item=gpu_stats[self.get_key()], key='mem', option='decoration') |
244
|
|
|
) |
245
|
|
|
) |
246
|
|
|
# New line |
247
|
|
|
ret.append(self.curse_new_line()) |
248
|
|
|
# GPU TEMPERATURE |
249
|
|
|
try: |
250
|
|
|
mean_temperature = sum(s['temperature'] for s in self.stats if s is not None) / len(self.stats) |
251
|
|
|
except TypeError: |
252
|
|
|
mean_temperature_msg = '{:>4}'.format('N/A') |
253
|
|
|
else: |
254
|
|
|
unit = 'C' |
255
|
|
|
if args.fahrenheit: |
256
|
|
|
mean_temperature = to_fahrenheit(mean_temperature) |
257
|
|
|
unit = 'F' |
258
|
|
|
mean_temperature_msg = '{:>3.0f}{}'.format(mean_temperature, unit) |
259
|
|
|
if len(self.stats) > 1: |
260
|
|
|
msg = '{:13}'.format('temp mean:') |
261
|
|
|
else: |
262
|
|
|
msg = '{:13}'.format('temperature:') |
263
|
|
|
ret.append(self.curse_add_line(msg)) |
264
|
|
|
ret.append( |
265
|
|
|
self.curse_add_line( |
266
|
|
|
mean_temperature_msg, |
267
|
|
|
self.get_views(item=gpu_stats[self.get_key()], key='temperature', option='decoration'), |
268
|
|
|
) |
269
|
|
|
) |
270
|
|
|
else: |
271
|
|
|
# Multi GPU |
272
|
|
|
# Temperature is not displayed in this mode... |
273
|
|
|
for gpu_stats in self.stats: |
274
|
|
|
# New line |
275
|
|
|
ret.append(self.curse_new_line()) |
276
|
|
|
# GPU ID + PROC + MEM + TEMPERATURE |
277
|
|
|
id_msg = '{:>7}'.format(gpu_stats['gpu_id']) |
278
|
|
|
try: |
279
|
|
|
proc_msg = '{:>3.0f}%'.format(gpu_stats['proc']) |
280
|
|
|
except (ValueError, TypeError): |
281
|
|
|
proc_msg = '{:>4}'.format('N/A') |
282
|
|
|
try: |
283
|
|
|
mem_msg = '{:>3.0f}%'.format(gpu_stats['mem']) |
284
|
|
|
except (ValueError, TypeError): |
285
|
|
|
mem_msg = '{:>4}'.format('N/A') |
286
|
|
|
msg = '{} {} mem {}'.format(id_msg, proc_msg, mem_msg) |
287
|
|
|
ret.append(self.curse_add_line(msg)) |
288
|
|
|
|
289
|
|
|
return ret |
290
|
|
|
|