1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
# |
3
|
|
|
# This file is part of Glances. |
4
|
|
|
# |
5
|
|
|
# Copyright (C) 2017 Kirby Banman <[email protected]> |
6
|
|
|
# |
7
|
|
|
# Glances is free software; you can redistribute it and/or modify |
8
|
|
|
# it under the terms of the GNU Lesser General Public License as published by |
9
|
|
|
# the Free Software Foundation, either version 3 of the License, or |
10
|
|
|
# (at your option) any later version. |
11
|
|
|
# |
12
|
|
|
# Glances is distributed in the hope that it will be useful, |
13
|
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of |
14
|
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
15
|
|
|
# GNU Lesser General Public License for more details. |
16
|
|
|
# |
17
|
|
|
# You should have received a copy of the GNU Lesser General Public License |
18
|
|
|
# along with this program. If not, see <http://www.gnu.org/licenses/>. |
19
|
|
|
|
20
|
|
|
"""GPU plugin (limited to NVIDIA chipsets)""" |
21
|
|
|
|
22
|
|
|
from glances.logger import logger |
23
|
|
|
from glances.plugins.glances_plugin import GlancesPlugin |
24
|
|
|
|
25
|
|
|
try: |
26
|
|
|
import pynvml |
27
|
|
|
except ImportError: |
28
|
|
|
logger.debug("Could not import pynvml. NVIDIA stats will not be collected.") |
29
|
|
|
gpu_nvidia_tag = False |
30
|
|
|
else: |
31
|
|
|
gpu_nvidia_tag = True |
32
|
|
|
|
33
|
|
|
|
34
|
|
|
class Plugin(GlancesPlugin): |
35
|
|
|
|
36
|
|
|
"""Glances GPU plugin (limited to NVIDIA chipsets). |
37
|
|
|
|
38
|
|
|
stats is a list of dictionaries with one entry per GPU |
39
|
|
|
""" |
40
|
|
|
|
41
|
|
|
def __init__(self, args=None): |
42
|
|
|
"""Init the plugin""" |
43
|
|
|
super(Plugin, self).__init__(args=args) |
44
|
|
|
|
45
|
|
|
# Init the NVidia API |
46
|
|
|
self.init_nvidia() |
47
|
|
|
|
48
|
|
|
# We want to display the stat in the curse interface |
49
|
|
|
self.display_curse = True |
50
|
|
|
|
51
|
|
|
# Init the stats |
52
|
|
|
self.reset() |
53
|
|
|
|
54
|
|
|
def reset(self): |
55
|
|
|
"""Reset/init the stats.""" |
56
|
|
|
self.stats = [] |
57
|
|
|
|
58
|
|
|
def init_nvidia(self): |
59
|
|
|
"""Init the NVIDIA API""" |
60
|
|
|
if not gpu_nvidia_tag: |
61
|
|
|
self.nvml_ready = False |
62
|
|
|
|
63
|
|
|
try: |
64
|
|
|
pynvml.nvmlInit() |
65
|
|
|
self.device_handles = self.get_device_handles() |
66
|
|
|
self.nvml_ready = True |
67
|
|
|
except Exception: |
68
|
|
|
logger.debug("pynvml could not be initialized.") |
69
|
|
|
self.nvml_ready = False |
70
|
|
|
|
71
|
|
|
return self.nvml_ready |
72
|
|
|
|
73
|
|
|
def get_key(self): |
74
|
|
|
"""Return the key of the list.""" |
75
|
|
|
return 'gpu_id' |
76
|
|
|
|
77
|
|
|
@GlancesPlugin._check_decorator |
78
|
|
|
@GlancesPlugin._log_result_decorator |
79
|
|
|
def update(self): |
80
|
|
|
"""Update the GPU stats""" |
81
|
|
|
|
82
|
|
|
self.reset() |
83
|
|
|
|
84
|
|
|
# !!! JUST FOR TEST |
85
|
|
|
# self.stats = [{"key": "gpu_id", "mem": None, "proc": 60, "gpu_id": 0, "name": "GeForce GTX 560 Ti"}] |
86
|
|
|
# self.stats = [{"key": "gpu_id", "mem": 10, "proc": 60, "gpu_id": 0, "name": "GeForce GTX 560 Ti"}] |
87
|
|
|
# self.stats = [{"key": "gpu_id", "mem": 48.64645, "proc": 60.73, "gpu_id": 0, "name": "GeForce GTX 560 Ti"}, |
88
|
|
|
# {"key": "gpu_id", "mem": 70.743, "proc": 80.28, "gpu_id": 1, "name": "GeForce GTX 560 Ti"}, |
89
|
|
|
# {"key": "gpu_id", "mem": 0, "proc": 0, "gpu_id": 2, "name": "GeForce GTX 560 Ti"}] |
90
|
|
|
# self.stats = [{"key": "gpu_id", "mem": 48.64645, "proc": 60.73, "gpu_id": 0, "name": "GeForce GTX 560 Ti"}, |
91
|
|
|
# {"key": "gpu_id", "mem": None, "proc": 80.28, "gpu_id": 1, "name": "GeForce GTX 560 Ti"}, |
92
|
|
|
# {"key": "gpu_id", "mem": 0, "proc": 0, "gpu_id": 2, "name": "ANOTHER GPU"}] |
93
|
|
|
# !!! TO BE COMMENTED |
94
|
|
|
|
95
|
|
|
if not self.nvml_ready: |
96
|
|
|
return self.stats |
97
|
|
|
|
98
|
|
|
if self.input_method == 'local': |
99
|
|
|
self.stats = self.get_device_stats() |
100
|
|
|
elif self.input_method == 'snmp': |
101
|
|
|
# not available |
102
|
|
|
pass |
103
|
|
|
|
104
|
|
|
return self.stats |
105
|
|
|
|
106
|
|
|
def update_views(self): |
107
|
|
|
"""Update stats views.""" |
108
|
|
|
# Call the father's method |
109
|
|
|
super(Plugin, self).update_views() |
110
|
|
|
|
111
|
|
|
# Add specifics informations |
112
|
|
|
# Alert |
113
|
|
|
for i in self.stats: |
114
|
|
|
# Init the views for the current GPU |
115
|
|
|
self.views[i[self.get_key()]] = {'proc': {}, 'mem': {}} |
116
|
|
|
# Processor alert |
117
|
|
|
if 'proc' in i: |
118
|
|
|
alert = self.get_alert(i['proc'], header='proc') |
119
|
|
|
self.views[i[self.get_key()]]['proc']['decoration'] = alert |
120
|
|
|
# Memory alert |
121
|
|
|
if 'mem' in i: |
122
|
|
|
alert = self.get_alert(i['mem'], header='mem') |
123
|
|
|
self.views[i[self.get_key()]]['mem']['decoration'] = alert |
124
|
|
|
|
125
|
|
|
return True |
126
|
|
|
|
127
|
|
|
def msg_curse(self, args=None): |
128
|
|
|
"""Return the dict to display in the curse interface.""" |
129
|
|
|
# Init the return message |
130
|
|
|
ret = [] |
131
|
|
|
|
132
|
|
|
# Only process if stats exist, not empty (issue #871) and plugin not disabled |
133
|
|
|
if not self.stats or (self.stats == []) or self.is_disable(): |
134
|
|
|
return ret |
135
|
|
|
|
136
|
|
|
# Check if all GPU have the same name |
137
|
|
|
same_name = all(s['name'] == self.stats[0]['name'] for s in self.stats) |
138
|
|
|
|
139
|
|
|
# gpu_stats contain the first GPU in the list |
140
|
|
|
gpu_stats = self.stats[0] |
141
|
|
|
|
142
|
|
|
# Header |
143
|
|
|
header = '' |
144
|
|
|
if len(self.stats) > 1: |
145
|
|
|
header += '{} '.format(len(self.stats)) |
146
|
|
|
if same_name: |
147
|
|
|
header += '{} {}'.format('GPU', gpu_stats['name']) |
148
|
|
|
else: |
149
|
|
|
header += '{}'.format('GPU') |
150
|
|
|
msg = header[:17] |
151
|
|
|
ret.append(self.curse_add_line(msg, "TITLE")) |
152
|
|
|
|
153
|
|
|
# Build the string message |
154
|
|
|
if len(self.stats) == 1 or args.meangpu: |
155
|
|
|
# GPU stat summary or mono GPU |
156
|
|
|
# New line |
157
|
|
|
ret.append(self.curse_new_line()) |
158
|
|
|
# GPU PROC |
159
|
|
|
try: |
160
|
|
|
mean_proc = sum(s['proc'] for s in self.stats if s is not None) / len(self.stats) |
161
|
|
|
except TypeError: |
162
|
|
|
mean_proc_msg = '{:>4}'.format('N/A') |
163
|
|
|
else: |
164
|
|
|
mean_proc_msg = '{:>3.0f}%'.format(mean_proc) |
165
|
|
|
if len(self.stats) > 1: |
166
|
|
|
msg = '{:13}'.format('proc mean:') |
167
|
|
|
else: |
168
|
|
|
msg = '{:13}'.format('proc:') |
169
|
|
|
ret.append(self.curse_add_line(msg)) |
170
|
|
|
ret.append(self.curse_add_line( |
171
|
|
|
mean_proc_msg, self.get_views(item=gpu_stats[self.get_key()], |
172
|
|
|
key='proc', |
173
|
|
|
option='decoration'))) |
174
|
|
|
# New line |
175
|
|
|
ret.append(self.curse_new_line()) |
176
|
|
|
# GPU MEM |
177
|
|
|
try: |
178
|
|
|
mean_mem = sum(s['mem'] for s in self.stats if s is not None) / len(self.stats) |
179
|
|
|
except TypeError: |
180
|
|
|
mean_mem_msg = '{:>4}'.format('N/A') |
181
|
|
|
else: |
182
|
|
|
mean_mem_msg = '{:>3.0f}%'.format(mean_mem) |
183
|
|
|
if len(self.stats) > 1: |
184
|
|
|
msg = '{:13}'.format('mem mean:') |
185
|
|
|
else: |
186
|
|
|
msg = '{:13}'.format('mem:') |
187
|
|
|
ret.append(self.curse_add_line(msg)) |
188
|
|
|
ret.append(self.curse_add_line( |
189
|
|
|
mean_mem_msg, self.get_views(item=gpu_stats[self.get_key()], |
190
|
|
|
key='mem', |
191
|
|
|
option='decoration'))) |
192
|
|
|
else: |
193
|
|
|
# Multi GPU |
194
|
|
|
for gpu_stats in self.stats: |
195
|
|
|
# New line |
196
|
|
|
ret.append(self.curse_new_line()) |
197
|
|
|
# GPU ID + PROC + MEM |
198
|
|
|
id_msg = '{}'.format(gpu_stats['gpu_id']) |
199
|
|
|
try: |
200
|
|
|
proc_msg = '{:>3.0f}%'.format(gpu_stats['proc']) |
201
|
|
|
except ValueError: |
202
|
|
|
proc_msg = '{:>4}'.format('N/A') |
203
|
|
|
try: |
204
|
|
|
mem_msg = '{:>3.0f}%'.format(gpu_stats['mem']) |
205
|
|
|
except ValueError: |
206
|
|
|
mem_msg = '{:>4}'.format('N/A') |
207
|
|
|
msg = '{}: {} mem: {}'.format(id_msg, proc_msg, mem_msg) |
208
|
|
|
ret.append(self.curse_add_line(msg)) |
209
|
|
|
|
210
|
|
|
return ret |
211
|
|
|
|
212
|
|
|
def get_device_handles(self): |
213
|
|
|
""" |
214
|
|
|
Returns a list of NVML device handles, one per device. Can throw NVMLError. |
215
|
|
|
""" |
216
|
|
|
return [pynvml.nvmlDeviceGetHandleByIndex(i) for i in range(pynvml.nvmlDeviceGetCount())] |
217
|
|
|
|
218
|
|
|
def get_device_stats(self): |
219
|
|
|
"""Get GPU stats""" |
220
|
|
|
stats = [] |
221
|
|
|
|
222
|
|
|
for index, device_handle in enumerate(self.device_handles): |
223
|
|
|
device_stats = {} |
224
|
|
|
# Dictionnary key is the GPU_ID |
225
|
|
|
device_stats['key'] = self.get_key() |
226
|
|
|
# GPU id (for multiple GPU, start at 0) |
227
|
|
|
device_stats['gpu_id'] = index |
228
|
|
|
# GPU name |
229
|
|
|
device_stats['name'] = self.get_device_name(device_handle) |
230
|
|
|
# Memory consumption in % (not available on all GPU) |
231
|
|
|
device_stats['mem'] = self.get_mem(device_handle) |
232
|
|
|
# Processor consumption in % |
233
|
|
|
device_stats['proc'] = self.get_proc(device_handle) |
234
|
|
|
stats.append(device_stats) |
235
|
|
|
|
236
|
|
|
return stats |
237
|
|
|
|
238
|
|
|
def get_device_name(self, device_handle): |
239
|
|
|
"""Get GPU device name""" |
240
|
|
|
try: |
241
|
|
|
return pynvml.nvmlDeviceGetName(device_handle) |
242
|
|
|
except pynvml.NVMlError: |
243
|
|
|
return "NVIDIA" |
244
|
|
|
|
245
|
|
|
def get_mem(self, device_handle): |
246
|
|
|
"""Get GPU device memory consumption in percent""" |
247
|
|
|
try: |
248
|
|
|
memory_info = pynvml.nvmlDeviceGetMemoryInfo(device_handle) |
249
|
|
|
return memory_info.used * 100.0 / memory_info.total |
250
|
|
|
except pynvml.NVMLError: |
251
|
|
|
return None |
252
|
|
|
|
253
|
|
|
def get_proc(self, device_handle): |
254
|
|
|
"""Get GPU device CPU consumption in percent""" |
255
|
|
|
try: |
256
|
|
|
return pynvml.nvmlDeviceGetUtilizationRates(device_handle).gpu |
257
|
|
|
except pynvml.NVMLError: |
258
|
|
|
return None |
259
|
|
|
|
260
|
|
|
def exit(self): |
261
|
|
|
"""Overwrite the exit method to close the GPU API""" |
262
|
|
|
if self.nvml_ready: |
263
|
|
|
try: |
264
|
|
|
pynvml.nvmlShutdown() |
265
|
|
|
except Exception as e: |
266
|
|
|
logger.debug("pynvml failed to shutdown correctly ({})".format(e)) |
267
|
|
|
|
268
|
|
|
# Call the father exit method |
269
|
|
|
super(Plugin, self).exit() |
270
|
|
|
|