-
Notifications
You must be signed in to change notification settings - Fork 191
/
nvme_metrics.py
executable file
·251 lines (226 loc) · 9.77 KB
/
nvme_metrics.py
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
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
#!/usr/bin/env python3
"""
NVMe device metrics textfile collector.
Requires nvme-cli package.
Formatted with Black:
$ black -l 100 nvme_metrics.py
"""
import json
import os
import re
import sys
import subprocess
# Disable automatic addition of _created series. Must be set before importing prometheus_client.
os.environ["PROMETHEUS_DISABLE_CREATED_SERIES"] = "true"
from prometheus_client import CollectorRegistry, Counter, Gauge, Info, generate_latest # noqa: E402
registry = CollectorRegistry()
namespace = "nvme"
metrics = {
# fmt: off
"avail_spare": Gauge(
"available_spare_ratio",
"Device available spare ratio",
["device"], namespace=namespace, registry=registry,
),
"controller_busy_time": Counter(
"controller_busy_time_seconds",
"Device controller busy time in seconds",
["device"], namespace=namespace, registry=registry,
),
"critical_warning": Gauge(
"critical_warning",
"Device critical warning bitmap field",
["device"], namespace=namespace, registry=registry,
),
"data_units_read": Counter(
"data_units_read_total",
"Number of 512-byte data units read by host, reported in thousands",
["device"], namespace=namespace, registry=registry,
),
"data_units_written": Counter(
"data_units_written_total",
"Number of 512-byte data units written by host, reported in thousands",
["device"], namespace=namespace, registry=registry,
),
"device_info": Info(
"device",
"Device information",
["device", "model", "firmware", "serial"], namespace=namespace, registry=registry,
),
"host_read_commands": Counter(
"host_read_commands_total",
"Device read commands from host",
["device"], namespace=namespace, registry=registry,
),
"host_write_commands": Counter(
"host_write_commands_total",
"Device write commands from host",
["device"], namespace=namespace, registry=registry,
),
"media_errors": Counter(
"media_errors_total",
"Device media errors total",
["device"], namespace=namespace, registry=registry,
),
"num_err_log_entries": Counter(
"num_err_log_entries_total",
"Device error log entry count",
["device"], namespace=namespace, registry=registry,
),
# FIXME: The "nvmecli" metric ought to be an Info type, not a Gauge. However, making this change
# will result in the metric having a "_info" suffix automatically appended, which is arguably
# a breaking change.
"nvmecli": Gauge(
"nvmecli",
"nvme-cli tool information",
["version"], namespace=namespace, registry=registry,
),
"percent_used": Gauge(
"percentage_used_ratio",
"Device percentage used ratio",
["device"], namespace=namespace, registry=registry,
),
"physical_size": Gauge(
"physical_size_bytes",
"Device size in bytes",
["device"], namespace=namespace, registry=registry,
),
"power_cycles": Counter(
"power_cycles_total",
"Device number of power cycles",
["device"], namespace=namespace, registry=registry,
),
"power_on_hours": Counter(
"power_on_hours_total",
"Device power-on hours",
["device"], namespace=namespace, registry=registry,
),
"sector_size": Gauge(
"sector_size_bytes",
"Device sector size in bytes",
["device"], namespace=namespace, registry=registry,
),
"spare_thresh": Gauge(
"available_spare_threshold_ratio",
"Device available spare threshold ratio",
["device"], namespace=namespace, registry=registry,
),
"temperature": Gauge(
"temperature_celsius",
"Device temperature in degrees Celsius",
["device"], namespace=namespace, registry=registry,
),
"unsafe_shutdowns": Counter(
"unsafe_shutdowns_total",
"Device number of unsafe shutdowns",
["device"], namespace=namespace, registry=registry,
),
"used_bytes": Gauge(
"used_bytes",
"Device used size in bytes",
["device"], namespace=namespace, registry=registry,
),
# fmt: on
}
def exec_nvme(*args):
"""
Execute nvme CLI tool with specified arguments and return captured stdout result. Set LC_ALL=C
in child process environment so that the nvme tool does not perform any locale-specific number
or date formatting, etc.
"""
cmd = ["nvme", *args]
return subprocess.check_output(cmd, stderr=subprocess.PIPE, env=dict(os.environ, LC_ALL="C"))
def exec_nvme_json(*args):
"""
Execute nvme CLI tool with specified arguments and return parsed JSON output.
"""
# Note: nvme-cli v2.11 effectively introduced a breaking change by forcing JSON output to always
# be verbose. Older versions of nvme-cli optionally produced verbose output if the --verbose
# flag was specified. In order to avoid having to handle two different JSON schemas, always
# add the --verbose flag.
output = exec_nvme(*args, "--output-format", "json", "--verbose")
return json.loads(output)
def main():
match = re.match(r"^nvme version (\S+)", exec_nvme("version").decode())
if match:
cli_version = match.group(1)
else:
cli_version = "unknown"
metrics["nvmecli"].labels(cli_version).set(1)
device_list = exec_nvme_json("list")
for device in device_list["Devices"]:
for subsys in device["Subsystems"]:
for ctrl in subsys["Controllers"]:
for ns in ctrl["Namespaces"]:
device_name = ns["NameSpace"]
# FIXME: This metric ought to be refactored into a "controller_info" metric,
# since it contains information that is not unique to the namespace. However,
# previous versions of this collector erroneously referred to namespaces, e.g.
# "nvme0n1", as devices, so preserve the former behaviour for now.
metrics["device_info"].labels(
device_name,
ctrl["ModelNumber"],
ctrl["Firmware"],
ctrl["SerialNumber"].strip(),
)
metrics["sector_size"].labels(device_name).set(ns["SectorSize"])
metrics["physical_size"].labels(device_name).set(ns["PhysicalSize"])
metrics["used_bytes"].labels(device_name).set(ns["UsedBytes"])
# FIXME: The smart-log should only need to be fetched once per controller, not
# per namespace. However, in order to preserve legacy metric labels, fetch it
# per namespace anyway. Most consumer grade SSDs will only have one namespace.
smart_log = exec_nvme_json("smart-log", os.path.join("/dev", device_name))
# Various counters in the NVMe specification are 128-bit, which would have to
# discard resolution if converted to a JSON number (i.e., float64_t). Instead,
# nvme-cli marshals them as strings. As such, they need to be explicitly cast
# to int or float when using them in Counter metrics.
metrics["data_units_read"].labels(device_name).inc(
int(smart_log["data_units_read"])
)
metrics["data_units_written"].labels(device_name).inc(
int(smart_log["data_units_written"])
)
metrics["host_read_commands"].labels(device_name).inc(
int(smart_log["host_read_commands"])
)
metrics["host_write_commands"].labels(device_name).inc(
int(smart_log["host_write_commands"])
)
metrics["avail_spare"].labels(device_name).set(smart_log["avail_spare"] / 100)
metrics["spare_thresh"].labels(device_name).set(smart_log["spare_thresh"] / 100)
metrics["percent_used"].labels(device_name).set(smart_log["percent_used"] / 100)
metrics["critical_warning"].labels(device_name).set(
smart_log["critical_warning"]["value"]
)
metrics["media_errors"].labels(device_name).inc(int(smart_log["media_errors"]))
metrics["num_err_log_entries"].labels(device_name).inc(
int(smart_log["num_err_log_entries"])
)
metrics["power_cycles"].labels(device_name).inc(int(smart_log["power_cycles"]))
metrics["power_on_hours"].labels(device_name).inc(
int(smart_log["power_on_hours"])
)
metrics["controller_busy_time"].labels(device_name).inc(
int(smart_log["controller_busy_time"])
)
metrics["unsafe_shutdowns"].labels(device_name).inc(
int(smart_log["unsafe_shutdowns"])
)
# NVMe reports temperature in kelvins; convert it to degrees Celsius.
metrics["temperature"].labels(device_name).set(smart_log["temperature"] - 273)
if __name__ == "__main__":
if os.geteuid() != 0:
print("ERROR: script requires root privileges", file=sys.stderr)
sys.exit(1)
# Check if nvme-cli is installed
try:
exec_nvme()
except FileNotFoundError:
print("ERROR: nvme-cli is not installed. Aborting.", file=sys.stderr)
sys.exit(1)
try:
main()
except Exception as e:
print("ERROR: {}".format(e), file=sys.stderr)
sys.exit(1)
print(generate_latest(registry).decode(), end="")