LTTng » LTTng-tools

h1. LTTng-tools Wiki

h2. Quickstart

This is a quick start guide for the complete LTTng tool chain. This is divided

in three sections respectively kernel tracing, user-space tracing and reading a

trace.

See the README file for installation procedure or use the various Linux

distribution packages.

In order to trace the kernel, you''ll need the lttng-modules 2.0 compiled and

installed. See http://lttng.org/lttng2.0 for more instructions for that part.

For user-space tracing, you''ll need an instrumented application with lttng-ust

2.0.

lttng-tools provide a session daemon (lttng-sessiond) that acts as a tracing

registry. To trace any instrumented applications or the kernel, a registered

tracing session is needed beforehand. To interact with the session daemon and a

tracing session, you should use the lttng command line UI (lttng). It is also

possible to use the liblttngctl library for tracing control (lttng.h).

Here is a list of some powerful features the LTTng 2.0 kernel tracer offers:

	* Kprobes support

	* Function Tracer support

	* Context information support (add context data to an event)

	* Perf counter support

	* Tracepoint support

And for the LTTng UST 2.0 tracer:

	* Applications registration

	* Automatic tracepoints activation upon app. registration

	* Context information support

	* Safe buffers after application crash

	* Per-user tracing (root access *not* mandatory)

The next sections explains how to do tracing :)

h3. Kernel Tracing

You can start the session daemon by invoking the command "lttng-sessiond", or

let the lttng command line tool do it for you. The session daemon loads the

LTTng tracer modules for you if those modules can be found on your system. If

they are not found, the kernel tracing feature will be unavailable.

List available kernel events:

<pre>

# lttng list -k

</pre>

1) Create a tracing session. The .lttng directory will be created with .lttngrc

file in $HOME containing the session name (here ''mysession'') you are working

on.

<pre>

# lttng create mysession

</pre>

If you have multiple sessions, you can change the current session by using

<pre>

# lttng set-session myothersession

</pre>

2) Enable all tracepoints and all system call events.

<pre>

# lttng enable-event -a -k

</pre>

3) Enable tracepoint event(s). Here for example, we want only

''sched_switch'' and ''sched_wakeup'' events for the kernel (-k/--kernel).

<pre>

# lttng enable-event sched_switch,sched_wakeup -k

</pre>

or enable ALL tracepoint events:

<pre>

# lttng enable-event -a -k --tracepoint

</pre>

4) Enable all system call event(s).

<pre>

# lttng enable-event -a -k --syscall

</pre>

5) Enable kprobes and/or the function tracer with lttng

This is a new feature made possible by the new LTTng 2.0 kernel tracer. You can

enable a dynamic probe and data will be output in the trace along side with

your tracing data.

<pre>

# lttng enable-event aname -k --probe symbol+0x0

</pre>

or

<pre>

# lttng enable-event aname -k --probe 0xffff7260695

</pre>

Either an <address> or a <symbol+offset> can be used for probes.

You can also enable function tracer, which uses the Ftrace API (by Steven

Rostedt). Again, data will be output in the trace.

<pre>

# lttng enable-event aname -k --function <symbol_name>

</pre>

6) Enable context information for an event:

This is also a new feature which allows you to add context information to an

event. For example, you can add the PID along with the event information:

<pre>

# lttng add-context -k -e sched_switch -t pid

</pre>

At this point, you will have to look at ''lttng add-context --help'' for all

possible context type.

You can on the same line activate multiple context:

<pre>

# lttng add-context -k -e sched_switch -t pid -t nice -t tid

</pre>

7) Enable perf counter for an event:

Again, a new powerful feature is the possibility to add perf counter data

(using the perf API by Ingo Molnar and Thomas Gleixner) to the trace on a per

event basis. Let say we want to get the CPU cycles at each event:

<pre>

# lttng add-context -k -e sched_switch -t perf:cpu-cycles

</pre>

You''ll have to use the add-context help for all possible perf counter values.

8) Start tracing:

<pre>

# lttng start

</pre>

Tracing is in progress at this point and traces will be written in

$HOME/lttng-traces/mysession-<date>-<time>

NOTE: It will start tracing for *all* domain(s).

9) Stop tracing:

<pre>

# lttng stop

</pre>

NOTE: At this point, you can restart the trace (lttng start), enable/disable

events or just go take a break and come back 3 days later to start it again :).

You can also read the trace since the buffers are flushed on stop command.

10) Destroy your session after you are done with tracing

<pre>

# lttng destroy

</pre>

See Reading a trace section below to read you trace(s).

h3. User-space Tracing

Like kernel tracing, you can start the session daemon by invoking the command

"lttng-sessiond", or let the lttng command line tool do it for you.

NOTE: You do *not* need root credentials in order to tracer user-space

applications. However, if you run the session daemon under non-root user

rights, only applications of that user will be traced.

So, after instrumenting you applications with LTTng-ust 2.0

(http://lttng.org/lttng2.0), upon startup, it will automatically register to

the session daemon. If there is none running, it will simply wait on a seperate

thread for a session daemon to appear and then register.

Start your instrumented application at any time but at least before starting

tracing :).

List available registered applications:

<pre>

$ lttng list -u

</pre>

1) Create a tracing session. The .lttng directory will be created with a

.lttngrc file in $HOME containing the session name (here ''mysession'') you are

working on.

<pre>

$ lttng create mysession

</pre>

If you have multiple sessions, you can change the current session by using:

<pre>

$ lttng set-session myothersession

</pre>

2) Enable all tracepoints for the global UST domain ("-u" alone).

<pre>

$ lttng enable-event -a -u

</pre>

or enable a single tracepoint event.

<pre>

$ lttng enable-event ust_tests_hello:tptest -u

</pre>

3) This is also a new feature which allows you to add context information to an

event. For example, you can add the PID along with the event information:

<pre>

$ lttng add-context -t pid -e ust_tests_hello:tptest -u

</pre>

At this point, you will have to look at ''lttng add-context --help'' for all

possible context type.

You can on the same line activate multiple context:

<pre>

$ lttng add-context -u -e ust_tests_hello:tptest -t pid -t nice -t tid

</pre>

4) Start tracing:

<pre>

$ lttng start

</pre>

Tracing is in progress at this point and traces will be written in

$HOME/lttng-traces/mysession-<date>-<time>/ust/<procname>-<pid>-<date>-<time>

NOTE: It will start tracing for *all* domain(s).

5) Stop tracing:

<pre>

$ lttng stop

</pre>

NOTE: At this point, you can restart the trace (lttng start), enable/disable

events or just go take a break and come back 3 days later to start it again :).

You can also read the trace since the buffers are flushed on stop command.

6) Destroy your session after you are done with tracing

<pre>

$ lttng destroy

</pre>

h3. Reading a trace

The tool "Babeltrace" can be used to dump your binary trace into a

human-readable text format. Please see http://www.efficios.com/babeltrace and

git tree http://git.efficios.com/?p=babeltrace.git

<pre>

# babeltrace $HOME/lttng-traces/mysession-<date>-<time> | less

</pre>

Voilà