forked from iovisor/bcc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
funclatency_example.txt
225 lines (194 loc) · 11.7 KB
/
funclatency_example.txt
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
Demonstrations of funclatency, the Linux eBPF/bcc version.
Timing the do_sys_open() kernel function until Ctrl-C:
# ./funclatency do_sys_open
Tracing do_sys_open... Hit Ctrl-C to end.
^C
nsecs : count distribution
0 -> 1 : 0 | |
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 0 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 0 | |
128 -> 255 : 0 | |
256 -> 511 : 0 | |
512 -> 1023 : 0 | |
1024 -> 2047 : 0 | |
2048 -> 4095 : 124 |**************** |
4096 -> 8191 : 291 |**************************************|
8192 -> 16383 : 36 |**** |
16384 -> 32767 : 16 |** |
32768 -> 65535 : 8 |* |
65536 -> 131071 : 0 | |
131072 -> 262143 : 0 | |
262144 -> 524287 : 0 | |
524288 -> 1048575 : 0 | |
1048576 -> 2097151 : 0 | |
2097152 -> 4194303 : 1 | |
Detaching...
The output shows a histogram of function latency (call time), measured from when
the function began executing (was called) to when it finished (returned).
This example output shows that most of the time, do_sys_open() took between
2048 and 65536 nanoseconds (2 to 65 microseconds). The peak of this distribution
shows 291 calls of between 4096 and 8191 nanoseconds. There was also one
occurrance, an outlier, in the 2 to 4 millisecond range.
How this works: the function entry and return are traced using the kernel kprobe
and kretprobe tracer. Timestamps are collected, the delta time calculated, which
is the bucketized and stored as an in-kernel histogram for efficiency. The
histgram is visible in the output: it's the "count" column; everything else is
decoration. Only the count column is copied to user-level on output. This is an
efficient way to time kernel functions and examine their latency distribution.
Now vfs_read() is traced, and a microseconds histogram printed:
# ./funclatency -u vfs_read
Tracing vfs_read... Hit Ctrl-C to end.
^C
usecs : count distribution
0 -> 1 : 1143 |**************************************|
2 -> 3 : 420 |************* |
4 -> 7 : 159 |***** |
8 -> 15 : 295 |********* |
16 -> 31 : 25 | |
32 -> 63 : 5 | |
64 -> 127 : 1 | |
128 -> 255 : 0 | |
256 -> 511 : 0 | |
512 -> 1023 : 0 | |
1024 -> 2047 : 1 | |
2048 -> 4095 : 0 | |
4096 -> 8191 : 5 | |
8192 -> 16383 : 0 | |
16384 -> 32767 : 0 | |
32768 -> 65535 : 0 | |
65536 -> 131071 : 7 | |
131072 -> 262143 : 7 | |
262144 -> 524287 : 3 | |
524288 -> 1048575 : 7 | |
Detaching...
This shows a bimodal distribution. Many vfs_read() calls were faster than 15
microseconds, however, there was also a small handful between 65 milliseconds
and 1 second, seen at the bottom of the table. These are likely network reads
from SSH, waiting on interactive keystrokes.
Tracing do_nanosleep() in milliseconds:
# ./funclatency -m do_nanosleep
Tracing do_nanosleep... Hit Ctrl-C to end.
^C
msecs : count distribution
0 -> 1 : 0 | |
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 0 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 0 | |
128 -> 255 : 0 | |
256 -> 511 : 0 | |
512 -> 1023 : 328 |**************************************|
1024 -> 2047 : 0 | |
2048 -> 4095 : 0 | |
4096 -> 8191 : 32 |*** |
8192 -> 16383 : 0 | |
16384 -> 32767 : 0 | |
32768 -> 65535 : 2 | |
Detaching...
This looks like it has found threads that are sleeping every 1, 5, and 60
seconds.
An interval can be provided using -i, and timestamps added using -T. For
example, tracing vfs_read() latency in milliseconds and printing output
every 5 seconds:
# ./funclatency -mTi 5 vfs_read
Tracing vfs_read... Hit Ctrl-C to end.
20:10:08
msecs : count distribution
0 -> 1 : 1500 |*************************************+|
2 -> 3 : 3 | |
4 -> 7 : 1 | |
8 -> 15 : 2 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 4 | |
128 -> 255 : 3 | |
256 -> 511 : 1 | |
512 -> 1023 : 7 | |
20:10:13
msecs : count distribution
0 -> 1 : 1251 |*************************************+|
2 -> 3 : 3 | |
4 -> 7 : 2 | |
8 -> 15 : 0 | |
16 -> 31 : 2 | |
32 -> 63 : 3 | |
64 -> 127 : 5 | |
128 -> 255 : 5 | |
256 -> 511 : 3 | |
512 -> 1023 : 6 | |
1024 -> 2047 : 2 | |
20:10:18
msecs : count distribution
0 -> 1 : 1265 |*************************************+|
2 -> 3 : 0 | |
4 -> 7 : 5 | |
8 -> 15 : 9 | |
16 -> 31 : 7 | |
32 -> 63 : 1 | |
64 -> 127 : 2 | |
128 -> 255 : 3 | |
256 -> 511 : 5 | |
512 -> 1023 : 5 | |
1024 -> 2047 : 0 | |
2048 -> 4095 : 1 | |
^C
20:10:20
msecs : count distribution
0 -> 1 : 249 |*************************************+|
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 1 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 0 | |
128 -> 255 : 0 | |
256 -> 511 : 0 | |
512 -> 1023 : 1 | |
Detaching...
A single process can be traced, which filters in-kernel for efficiency. Here,
the vfs_read() function is timed as milliseconds for PID 17064, which is a
bash shell:
# ./funclatency -mp 17064 vfs_read
Tracing vfs_read... Hit Ctrl-C to end.
^C
msecs : count distribution
0 -> 1 : 1 |** |
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 1 |** |
16 -> 31 : 2 |***** |
32 -> 63 : 0 | |
64 -> 127 : 13 |**************************************|
128 -> 255 : 10 |***************************** |
256 -> 511 : 4 |*********** |
Detaching...
The distribution between 64 and 511 milliseconds shows keystroke latency.
USAGE message:
# ./funclatency -h
usage: funclatency [-h] [-p PID] [-i INTERVAL] [-T] [-u] [-m] [-r] pattern
Time kernel funcitons and print latency as a histogram
positional arguments:
pattern search expression for kernel functions
optional arguments:
-h, --help show this help message and exit
-p PID, --pid PID trace this PID only
-i INTERVAL, --interval INTERVAL
summary interval, seconds
-T, --timestamp include timestamp on output
-u, --microseconds microsecond histogram
-m, --milliseconds millisecond histogram
-r, --regexp use regular expressions. Default is "*" wildcards
only.
examples:
./funclatency do_sys_open # time the do_sys_open() kenel function
./funclatency -u vfs_read # time vfs_read(), in microseconds
./funclatency -m do_nanosleep # time do_nanosleep(), in milliseconds
./funclatency -mTi 5 vfs_read # output every 5 seconds, with timestamps
./funclatency -p 181 vfs_read # time process 181 only
./funclatency 'vfs_fstat*' # time both vfs_fstat() and vfs_fstatat()