Commit | Line | Data |
---|---|---|
5e0cbfb0 PP |
1 | --- |
2 | id: viewing-and-analyzing-your-traces | |
3 | --- | |
4 | ||
5 | This section describes how to visualize the data gathered after tracing | |
6 | the Linux kernel or a user space application. | |
7 | ||
1c3359df | 8 | Many ways exist to read LTTng traces: |
5e0cbfb0 PP |
9 | |
10 | * **`babeltrace`** is a command line utility which converts trace formats; | |
11 | it supports the format used by LTTng, | |
12 | <abbr title="Common Trace Format">CTF</abbr>, as well as a basic | |
13 | text output which may be `grep`ed. The `babeltrace` command is | |
14 | part of the | |
1c3359df PP |
15 | <a href="http://diamon.org/babeltrace" class="ext">Babeltrace</a> project. |
16 | * Babeltrace also includes **Python bindings** so that you may | |
5e0cbfb0 PP |
17 | easily open and read an LTTng trace with your own script, benefiting |
18 | from the power of Python. | |
1c3359df | 19 | * **<a href="http://tracecompass.org/" class="ext">Trace Compass</a>** |
b8734c94 | 20 | is an Eclipse plugin used to visualize and analyze various types of |
1c3359df | 21 | traces, including LTTng's. It also comes as a standalone application. |
5e0cbfb0 | 22 | |
1c3359df | 23 | LTTng trace files are recorded in the `~/lttng-traces` directory by default. |
5e0cbfb0 PP |
24 | Let's now view the trace and perform a basic analysis using |
25 | `babeltrace`. | |
26 | ||
27 | The simplest way to list all the recorded events of a trace is to pass its | |
28 | path to `babeltrace` with no options: | |
29 | ||
30 | <pre class="term"> | |
31 | babeltrace ~/lttng-traces/my-session | |
32 | </pre> | |
33 | ||
1c3359df | 34 | `babeltrace` finds all traces recursively within the given path and |
47bfcb75 | 35 | prints all their events, merging them in order of time. |
5e0cbfb0 PP |
36 | |
37 | Listing all the system calls of a Linux kernel trace with their arguments is | |
38 | easy with `babeltrace` and `grep`: | |
39 | ||
40 | <pre class="term"> | |
41 | babeltrace ~/lttng-traces/my-kernel-session | grep sys_ | |
42 | </pre> | |
43 | ||
44 | Counting events is also straightforward: | |
45 | ||
46 | <pre class="term"> | |
b80ba306 | 47 | babeltrace ~/lttng-traces/my-kernel-session | grep sys_read | wc --lines |
5e0cbfb0 PP |
48 | </pre> |
49 | ||
50 | The text output of `babeltrace` is useful for isolating events by simple | |
51 | matching using `grep` and similar utilities. However, more elaborate filters | |
52 | such as keeping only events with a field value falling within a specific range | |
53 | are not trivial to write using a shell. Moreover, reductions and even the | |
54 | most basic computations involving multiple events are virtually impossible | |
55 | to implement. | |
56 | ||
1c3359df | 57 | Fortunately, Babeltrace ships with Python 3 bindings which makes it |
5e0cbfb0 PP |
58 | really easy to read the events of an LTTng trace sequentially and compute |
59 | the desired information. | |
60 | ||
1c3359df | 61 | Here's a simple example using the Babeltrace Python bindings. The following |
5e0cbfb0 | 62 | script accepts an LTTng Linux kernel trace path as its first argument and |
1c3359df | 63 | prints the short names of the top 5 running processes on CPU 0 during the |
5e0cbfb0 PP |
64 | whole trace: |
65 | ||
66 | ~~~ python | |
67 | import sys | |
68 | from collections import Counter | |
69 | import babeltrace | |
70 | ||
71 | ||
72 | def top5proc(): | |
73 | if len(sys.argv) != 2: | |
74 | msg = 'Usage: python {} TRACEPATH'.format(sys.argv[0]) | |
75 | raise ValueError(msg) | |
76 | ||
77 | # a trace collection holds one to many traces | |
78 | col = babeltrace.TraceCollection() | |
79 | ||
80 | # add the trace provided by the user | |
81 | # (LTTng traces always have the 'ctf' format) | |
82 | if col.add_trace(sys.argv[1], 'ctf') is None: | |
83 | raise RuntimeError('Cannot add trace') | |
84 | ||
85 | # this counter dict will hold execution times: | |
86 | # | |
87 | # task command name -> total execution time (ns) | |
88 | exec_times = Counter() | |
89 | ||
90 | # this holds the last `sched_switch` timestamp | |
91 | last_ts = None | |
92 | ||
93 | # iterate events | |
94 | for event in col.events: | |
95 | # keep only `sched_switch` events | |
96 | if event.name != 'sched_switch': | |
97 | continue | |
98 | ||
99 | # keep only events which happened on CPU 0 | |
100 | if event['cpu_id'] != 0: | |
101 | continue | |
102 | ||
103 | # event timestamp | |
104 | cur_ts = event.timestamp | |
105 | ||
106 | if last_ts is None: | |
107 | # we start here | |
108 | last_ts = cur_ts | |
109 | ||
110 | # previous task command (short) name | |
111 | prev_comm = event['prev_comm'] | |
112 | ||
113 | # initialize entry in our dict if not yet done | |
114 | if prev_comm not in exec_times: | |
115 | exec_times[prev_comm] = 0 | |
116 | ||
117 | # compute previous command execution time | |
118 | diff = cur_ts - last_ts | |
119 | ||
120 | # update execution time of this command | |
121 | exec_times[prev_comm] += diff | |
122 | ||
123 | # update last timestamp | |
124 | last_ts = cur_ts | |
125 | ||
126 | # display top 10 | |
a345f8e0 | 127 | for name, ns in exec_times.most_common(5): |
5e0cbfb0 PP |
128 | s = ns / 1000000000 |
129 | print('{:20}{} s'.format(name, s)) | |
130 | ||
131 | ||
132 | if __name__ == '__main__': | |
133 | top5proc() | |
134 | ~~~ | |
135 | ||
136 | Save this script as `top5proc.py` and run it with Python 3, providing the | |
137 | path to an LTTng Linux kernel trace as the first argument: | |
138 | ||
139 | <pre class="term"> | |
140 | python3 top5proc.py ~/lttng-sessions/my-session-.../kernel | |
141 | </pre> | |
142 | ||
143 | Make sure the path you provide is the directory containing actual trace | |
ba3b1994 | 144 | files (`channel0_0`, `metadata`, and the rest): the `babeltrace` utility |
1c3359df | 145 | recurses directories, but the Python bindings do not. |
5e0cbfb0 PP |
146 | |
147 | Here's an example of output: | |
148 | ||
149 | ~~~ text | |
150 | swapper/0 48.607245889 s | |
151 | chromium 7.192738188 s | |
152 | pavucontrol 0.709894415 s | |
153 | Compositor 0.660867933 s | |
154 | Xorg.bin 0.616753786 s | |
155 | ~~~ | |
156 | ||
157 | Note that `swapper/0` is the "idle" process of CPU 0 on Linux; since we | |
158 | weren't using the CPU that much when tracing, its first position in the list | |
159 | makes sense. |