1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
# |
3
|
|
|
# This file is part of Glances. |
4
|
|
|
# |
5
|
|
|
# Copyright (C) 2020 Kirby Banman <[email protected]> |
6
|
|
|
# |
7
|
|
|
# SPDX-License-Identifier: LGPL-3.0-only |
8
|
|
|
# |
9
|
|
|
|
10
|
|
|
"""GPU plugin (limited to NVIDIA chipsets).""" |
11
|
|
|
|
12
|
|
|
from glances.globals import nativestr, to_fahrenheit |
13
|
|
|
from glances.logger import logger |
14
|
|
|
from glances.plugins.plugin.model import GlancesPluginModel |
15
|
|
|
|
16
|
|
|
# In Glances 3.1.4 or higher, we use the py3nvml lib (see issue #1523) |
17
|
|
|
try: |
18
|
|
|
import py3nvml.py3nvml as pynvml |
19
|
|
|
except Exception as e: |
20
|
|
|
import_error_tag = True |
21
|
|
|
# Display debug message if import KeyError |
22
|
|
|
logger.warning("Missing Python Lib ({}), Nvidia GPU plugin is disabled".format(e)) |
23
|
|
|
else: |
24
|
|
|
import_error_tag = False |
25
|
|
|
|
26
|
|
|
# Define the history items list |
27
|
|
|
# All items in this list will be historised if the --enable-history tag is set |
28
|
|
|
items_history_list = [ |
29
|
|
|
{'name': 'proc', 'description': 'GPU processor', 'y_unit': '%'}, |
30
|
|
|
{'name': 'mem', 'description': 'Memory consumption', 'y_unit': '%'}, |
31
|
|
|
] |
32
|
|
|
|
33
|
|
|
|
34
|
|
|
class PluginModel(GlancesPluginModel): |
35
|
|
|
"""Glances GPU plugin (limited to NVIDIA chipsets). |
36
|
|
|
|
37
|
|
|
stats is a list of dictionaries with one entry per GPU |
38
|
|
|
""" |
39
|
|
|
|
40
|
|
|
def __init__(self, args=None, config=None): |
41
|
|
|
"""Init the plugin.""" |
42
|
|
|
super(PluginModel, self).__init__(args=args, |
43
|
|
|
config=config, |
44
|
|
|
items_history_list=items_history_list, |
45
|
|
|
stats_init_value=[]) |
46
|
|
|
|
47
|
|
|
# Init the Nvidia API |
48
|
|
|
self.init_nvidia() |
49
|
|
|
|
50
|
|
|
# We want to display the stat in the curse interface |
51
|
|
|
self.display_curse = True |
52
|
|
|
|
53
|
|
|
def init_nvidia(self): |
54
|
|
|
"""Init the NVIDIA API.""" |
55
|
|
|
if import_error_tag: |
56
|
|
|
self.nvml_ready = False |
57
|
|
|
|
58
|
|
|
try: |
59
|
|
|
pynvml.nvmlInit() |
60
|
|
|
self.device_handles = get_device_handles() |
61
|
|
|
self.nvml_ready = True |
62
|
|
|
except Exception: |
63
|
|
|
logger.debug("pynvml could not be initialized.") |
64
|
|
|
self.nvml_ready = False |
65
|
|
|
|
66
|
|
|
return self.nvml_ready |
67
|
|
|
|
68
|
|
|
def get_key(self): |
69
|
|
|
"""Return the key of the list.""" |
70
|
|
|
return 'gpu_id' |
71
|
|
|
|
72
|
|
|
@GlancesPluginModel._check_decorator |
73
|
|
|
@GlancesPluginModel._log_result_decorator |
74
|
|
|
def update(self): |
75
|
|
|
"""Update the GPU stats.""" |
76
|
|
|
# Init new stats |
77
|
|
|
stats = self.get_init_value() |
78
|
|
|
|
79
|
|
|
if not self.nvml_ready: |
80
|
|
|
# !!! |
81
|
|
|
# Uncomment to test on computer without GPU |
82
|
|
|
# One GPU sample: |
83
|
|
|
# self.stats = [ |
84
|
|
|
# { |
85
|
|
|
# "key": "gpu_id", |
86
|
|
|
# "gpu_id": 0, |
87
|
|
|
# "name": "Fake GeForce GTX", |
88
|
|
|
# "mem": 5.792331695556641, |
89
|
|
|
# "proc": 4, |
90
|
|
|
# "temperature": 26, |
91
|
|
|
# "fan_speed": 30 |
92
|
|
|
# } |
93
|
|
|
# ] |
94
|
|
|
# Two GPU sample: |
95
|
|
|
# self.stats = [ |
96
|
|
|
# { |
97
|
|
|
# "key": "gpu_id", |
98
|
|
|
# "gpu_id": 0, |
99
|
|
|
# "name": "Fake GeForce GTX1", |
100
|
|
|
# "mem": 5.792331695556641, |
101
|
|
|
# "proc": 4, |
102
|
|
|
# "temperature": 26, |
103
|
|
|
# "fan_speed": 30 |
104
|
|
|
# }, |
105
|
|
|
# { |
106
|
|
|
# "key": "gpu_id", |
107
|
|
|
# "gpu_id": 1, |
108
|
|
|
# "name": "Fake GeForce GTX2", |
109
|
|
|
# "mem": 15, |
110
|
|
|
# "proc": 8, |
111
|
|
|
# "temperature": 65, |
112
|
|
|
# "fan_speed": 75 |
113
|
|
|
# } |
114
|
|
|
# ] |
115
|
|
|
return self.stats |
116
|
|
|
|
117
|
|
|
if self.input_method == 'local': |
118
|
|
|
stats = self.get_device_stats() |
119
|
|
|
elif self.input_method == 'snmp': |
120
|
|
|
# not available |
121
|
|
|
pass |
122
|
|
|
|
123
|
|
|
# Update the stats |
124
|
|
|
self.stats = stats |
125
|
|
|
|
126
|
|
|
return self.stats |
127
|
|
|
|
128
|
|
|
def update_views(self): |
129
|
|
|
"""Update stats views.""" |
130
|
|
|
# Call the father's method |
131
|
|
|
super(PluginModel, self).update_views() |
132
|
|
|
|
133
|
|
|
# Add specifics information |
134
|
|
|
# Alert |
135
|
|
|
for i in self.stats: |
136
|
|
|
# Init the views for the current GPU |
137
|
|
|
self.views[i[self.get_key()]] = {'proc': {}, 'mem': {}, 'temperature': {}} |
138
|
|
|
# Processor alert |
139
|
|
|
if 'proc' in i: |
140
|
|
|
alert = self.get_alert(i['proc'], header='proc') |
141
|
|
|
self.views[i[self.get_key()]]['proc']['decoration'] = alert |
142
|
|
|
# Memory alert |
143
|
|
|
if 'mem' in i: |
144
|
|
|
alert = self.get_alert(i['mem'], header='mem') |
145
|
|
|
self.views[i[self.get_key()]]['mem']['decoration'] = alert |
146
|
|
|
# Temperature alert |
147
|
|
|
if 'temperature' in i: |
148
|
|
|
alert = self.get_alert(i['temperature'], header='temperature') |
149
|
|
|
self.views[i[self.get_key()]]['temperature']['decoration'] = alert |
150
|
|
|
|
151
|
|
|
return True |
152
|
|
|
|
153
|
|
|
def msg_curse(self, args=None, max_width=None): |
154
|
|
|
"""Return the dict to display in the curse interface.""" |
155
|
|
|
# Init the return message |
156
|
|
|
ret = [] |
157
|
|
|
|
158
|
|
|
# Only process if stats exist, not empty (issue #871) and plugin not disabled |
159
|
|
|
if not self.stats or (self.stats == []) or self.is_disabled(): |
160
|
|
|
return ret |
161
|
|
|
|
162
|
|
|
# Check if all GPU have the same name |
163
|
|
|
same_name = all(s['name'] == self.stats[0]['name'] for s in self.stats) |
164
|
|
|
|
165
|
|
|
# gpu_stats contain the first GPU in the list |
166
|
|
|
gpu_stats = self.stats[0] |
167
|
|
|
|
168
|
|
|
# Header |
169
|
|
|
header = '' |
170
|
|
|
if len(self.stats) > 1: |
171
|
|
|
header += '{} '.format(len(self.stats)) |
172
|
|
|
if same_name: |
173
|
|
|
header += '{} {}'.format('GPU', gpu_stats['name']) |
174
|
|
|
else: |
175
|
|
|
header += '{}'.format('GPU') |
176
|
|
|
msg = header[:17] |
177
|
|
|
ret.append(self.curse_add_line(msg, "TITLE")) |
178
|
|
|
|
179
|
|
|
# Build the string message |
180
|
|
|
if len(self.stats) == 1 or args.meangpu: |
181
|
|
|
# GPU stat summary or mono GPU |
182
|
|
|
# New line |
183
|
|
|
ret.append(self.curse_new_line()) |
184
|
|
|
# GPU PROC |
185
|
|
|
try: |
186
|
|
|
mean_proc = sum(s['proc'] for s in self.stats if s is not None) / len(self.stats) |
187
|
|
|
except TypeError: |
188
|
|
|
mean_proc_msg = '{:>4}'.format('N/A') |
189
|
|
|
else: |
190
|
|
|
mean_proc_msg = '{:>3.0f}%'.format(mean_proc) |
191
|
|
|
if len(self.stats) > 1: |
192
|
|
|
msg = '{:13}'.format('proc mean:') |
193
|
|
|
else: |
194
|
|
|
msg = '{:13}'.format('proc:') |
195
|
|
|
ret.append(self.curse_add_line(msg)) |
196
|
|
|
ret.append( |
197
|
|
|
self.curse_add_line( |
198
|
|
|
mean_proc_msg, self.get_views(item=gpu_stats[self.get_key()], key='proc', option='decoration') |
199
|
|
|
) |
200
|
|
|
) |
201
|
|
|
# New line |
202
|
|
|
ret.append(self.curse_new_line()) |
203
|
|
|
# GPU MEM |
204
|
|
|
try: |
205
|
|
|
mean_mem = sum(s['mem'] for s in self.stats if s is not None) / len(self.stats) |
206
|
|
|
except TypeError: |
207
|
|
|
mean_mem_msg = '{:>4}'.format('N/A') |
208
|
|
|
else: |
209
|
|
|
mean_mem_msg = '{:>3.0f}%'.format(mean_mem) |
210
|
|
|
if len(self.stats) > 1: |
211
|
|
|
msg = '{:13}'.format('mem mean:') |
212
|
|
|
else: |
213
|
|
|
msg = '{:13}'.format('mem:') |
214
|
|
|
ret.append(self.curse_add_line(msg)) |
215
|
|
|
ret.append( |
216
|
|
|
self.curse_add_line( |
217
|
|
|
mean_mem_msg, self.get_views(item=gpu_stats[self.get_key()], key='mem', option='decoration') |
218
|
|
|
) |
219
|
|
|
) |
220
|
|
|
# New line |
221
|
|
|
ret.append(self.curse_new_line()) |
222
|
|
|
# GPU TEMPERATURE |
223
|
|
|
try: |
224
|
|
|
mean_temperature = sum(s['temperature'] for s in self.stats if s is not None) / len(self.stats) |
225
|
|
|
except TypeError: |
226
|
|
|
mean_temperature_msg = '{:>4}'.format('N/A') |
227
|
|
|
else: |
228
|
|
|
unit = 'C' |
229
|
|
|
if args.fahrenheit: |
230
|
|
|
mean_temperature = to_fahrenheit(mean_temperature) |
231
|
|
|
unit = 'F' |
232
|
|
|
mean_temperature_msg = '{:>3.0f}{}'.format(mean_temperature, unit) |
233
|
|
|
if len(self.stats) > 1: |
234
|
|
|
msg = '{:13}'.format('temp mean:') |
235
|
|
|
else: |
236
|
|
|
msg = '{:13}'.format('temperature:') |
237
|
|
|
ret.append(self.curse_add_line(msg)) |
238
|
|
|
ret.append( |
239
|
|
|
self.curse_add_line( |
240
|
|
|
mean_temperature_msg, |
241
|
|
|
self.get_views(item=gpu_stats[self.get_key()], key='temperature', option='decoration'), |
242
|
|
|
) |
243
|
|
|
) |
244
|
|
|
else: |
245
|
|
|
# Multi GPU |
246
|
|
|
# Temperature is not displayed in this mode... |
247
|
|
|
for gpu_stats in self.stats: |
248
|
|
|
# New line |
249
|
|
|
ret.append(self.curse_new_line()) |
250
|
|
|
# GPU ID + PROC + MEM + TEMPERATURE |
251
|
|
|
id_msg = '{}'.format(gpu_stats['gpu_id']) |
252
|
|
|
try: |
253
|
|
|
proc_msg = '{:>3.0f}%'.format(gpu_stats['proc']) |
254
|
|
|
except (ValueError, TypeError): |
255
|
|
|
proc_msg = '{:>4}'.format('N/A') |
256
|
|
|
try: |
257
|
|
|
mem_msg = '{:>3.0f}%'.format(gpu_stats['mem']) |
258
|
|
|
except (ValueError, TypeError): |
259
|
|
|
mem_msg = '{:>4}'.format('N/A') |
260
|
|
|
msg = '{}: {} mem: {}'.format(id_msg, proc_msg, mem_msg) |
261
|
|
|
ret.append(self.curse_add_line(msg)) |
262
|
|
|
|
263
|
|
|
return ret |
264
|
|
|
|
265
|
|
|
def get_device_stats(self): |
266
|
|
|
"""Get GPU stats.""" |
267
|
|
|
stats = [] |
268
|
|
|
|
269
|
|
|
for index, device_handle in enumerate(self.device_handles): |
270
|
|
|
device_stats = dict() |
271
|
|
|
# Dictionary key is the GPU_ID |
272
|
|
|
device_stats['key'] = self.get_key() |
273
|
|
|
# GPU id (for multiple GPU, start at 0) |
274
|
|
|
device_stats['gpu_id'] = index |
275
|
|
|
# GPU name |
276
|
|
|
device_stats['name'] = get_device_name(device_handle) |
277
|
|
|
# Memory consumption in % (not available on all GPU) |
278
|
|
|
device_stats['mem'] = get_mem(device_handle) |
279
|
|
|
# Processor consumption in % |
280
|
|
|
device_stats['proc'] = get_proc(device_handle) |
281
|
|
|
# Processor temperature in °C |
282
|
|
|
device_stats['temperature'] = get_temperature(device_handle) |
283
|
|
|
# Fan speed in % |
284
|
|
|
device_stats['fan_speed'] = get_fan_speed(device_handle) |
285
|
|
|
stats.append(device_stats) |
286
|
|
|
|
287
|
|
|
return stats |
288
|
|
|
|
289
|
|
|
def exit(self): |
290
|
|
|
"""Overwrite the exit method to close the GPU API.""" |
291
|
|
|
if self.nvml_ready: |
292
|
|
|
try: |
293
|
|
|
pynvml.nvmlShutdown() |
294
|
|
|
except Exception as e: |
295
|
|
|
logger.debug("pynvml failed to shutdown correctly ({})".format(e)) |
296
|
|
|
|
297
|
|
|
# Call the father exit method |
298
|
|
|
super(PluginModel, self).exit() |
299
|
|
|
|
300
|
|
|
|
301
|
|
|
def get_device_handles(): |
302
|
|
|
"""Get a list of NVML device handles, one per device. |
303
|
|
|
|
304
|
|
|
Can throw NVMLError. |
305
|
|
|
""" |
306
|
|
|
return [pynvml.nvmlDeviceGetHandleByIndex(i) for i in range(pynvml.nvmlDeviceGetCount())] |
307
|
|
|
|
308
|
|
|
|
309
|
|
|
def get_device_name(device_handle): |
310
|
|
|
"""Get GPU device name.""" |
311
|
|
|
try: |
312
|
|
|
return nativestr(pynvml.nvmlDeviceGetName(device_handle)) |
313
|
|
|
except pynvml.NVMLError: |
314
|
|
|
return "NVIDIA" |
315
|
|
|
|
316
|
|
|
|
317
|
|
|
def get_mem(device_handle): |
318
|
|
|
"""Get GPU device memory consumption in percent.""" |
319
|
|
|
try: |
320
|
|
|
memory_info = pynvml.nvmlDeviceGetMemoryInfo(device_handle) |
321
|
|
|
return memory_info.used * 100.0 / memory_info.total |
322
|
|
|
except pynvml.NVMLError: |
323
|
|
|
return None |
324
|
|
|
|
325
|
|
|
|
326
|
|
|
def get_proc(device_handle): |
327
|
|
|
"""Get GPU device CPU consumption in percent.""" |
328
|
|
|
try: |
329
|
|
|
return pynvml.nvmlDeviceGetUtilizationRates(device_handle).gpu |
330
|
|
|
except pynvml.NVMLError: |
331
|
|
|
return None |
332
|
|
|
|
333
|
|
|
|
334
|
|
|
def get_temperature(device_handle): |
335
|
|
|
"""Get GPU device CPU temperature in Celsius.""" |
336
|
|
|
try: |
337
|
|
|
return pynvml.nvmlDeviceGetTemperature(device_handle, pynvml.NVML_TEMPERATURE_GPU) |
338
|
|
|
except pynvml.NVMLError: |
339
|
|
|
return None |
340
|
|
|
|
341
|
|
|
|
342
|
|
|
def get_fan_speed(device_handle): |
343
|
|
|
"""Get GPU device fan speed in percent.""" |
344
|
|
|
try: |
345
|
|
|
return pynvml.nvmlDeviceGetFanSpeed(device_handle) |
346
|
|
|
except pynvml.NVMLError: |
347
|
|
|
return None |
348
|
|
|
|