1 The LTTng Documentation
2 =======================
3 Philippe Proulx <pproulx@efficios.com>
4 v2.10, 25 February 2021
7 include::../common/copyright.txt[]
10 include::../common/warning-not-maintained.txt[]
13 include::../common/welcome.txt[]
16 include::../common/audience.txt[]
20 === What's in this documentation?
22 The LTTng Documentation is divided into the following sections:
24 * **<<nuts-and-bolts,Nuts and bolts>>** explains the
25 rudiments of software tracing and the rationale behind the
28 You can skip this section if you’re familiar with software tracing and
29 with the LTTng project.
31 * **<<installing-lttng,Installation>>** describes the steps to
32 install the LTTng packages on common Linux distributions and from
35 You can skip this section if you already properly installed LTTng on
38 * **<<getting-started,Quick start>>** is a concise guide to
39 getting started quickly with LTTng kernel and user space tracing.
41 We recommend this section if you're new to LTTng or to software tracing
44 You can skip this section if you're not new to LTTng.
46 * **<<core-concepts,Core concepts>>** explains the concepts at
49 It's a good idea to become familiar with the core concepts
50 before attempting to use the toolkit.
52 * **<<plumbing,Components of LTTng>>** describes the various components
53 of the LTTng machinery, like the daemons, the libraries, and the
54 command-line interface.
55 * **<<instrumenting,Instrumentation>>** shows different ways to
56 instrument user applications and the Linux kernel.
58 Instrumenting source code is essential to provide a meaningful
61 You can skip this section if you do not have a programming background.
63 * **<<controlling-tracing,Tracing control>>** is divided into topics
64 which demonstrate how to use the vast array of features that
65 LTTng{nbsp}{revision} offers.
66 * **<<reference,Reference>>** contains reference tables.
67 * **<<glossary,Glossary>>** is a specialized dictionary of terms related
68 to LTTng or to the field of software tracing.
71 include::../common/convention.txt[]
74 include::../common/acknowledgements.txt[]
78 == What's new in LTTng {revision}?
80 LTTng{nbsp}{revision} bears the name _KeKriek_. From
81 http://brasseriedunham.com/[Brasserie Dunham], the _**KeKriek**_ is a
82 sour mashed golden wheat ale fermented with local sour cherries from
83 Tougas orchards. Fresh sweet cherry notes with some tartness, lively
84 carbonation with a dry finish.
86 New features and changes in LTTng{nbsp}{revision}:
88 * **Tracing control**:
89 ** You can put more than one wildcard special character (`*`), and not
90 only at the end, when you <<enabling-disabling-events,create an event
91 rule>>, in both the instrumentation point name and the literal
93 link:/man/1/lttng-enable-event/v{revision}/#doc-filter-syntax[filter expressions]:
98 # lttng enable-event --kernel 'x86_*_local_timer_*' \
99 --filter='name == "*a*b*c*d*e" && count >= 23'
106 $ lttng enable-event --userspace '*_my_org:*msg*'
110 ** New trigger and notification API for
111 <<liblttng-ctl-lttng,`liblttng-ctl`>>. This new subsystem allows you
112 to register triggers which emit a notification when a given
113 condition is satisfied. As of LTTng{nbsp}{revision}, only
114 <<channel,channel>> buffer usage conditions are available.
115 Documentation is available in the
116 https://github.com/lttng/lttng-tools/tree/stable-{revision}/include/lttng[`liblttng-ctl`
118 <<notif-trigger-api,Get notified when a channel's buffer usage is too
121 ** You can now embed the whole textual LTTng-tools man pages into the
122 executables at build time with the `--enable-embedded-help`
123 configuration option. Thanks to this option, you don't need the
124 http://www.methods.co.nz/asciidoc/[AsciiDoc] and
125 https://directory.fsf.org/wiki/Xmlto[xmlto] tools at build time, and
126 a manual pager at run time, to get access to this documentation.
128 * **User space tracing**:
129 ** New blocking mode: an LTTng-UST tracepoint can now block until
130 <<channel,sub-buffer>> space is available instead of discarding event
131 records in <<channel-overwrite-mode-vs-discard-mode,discard mode>>.
132 With this feature, you can be sure that no event records are
133 discarded during your application's execution at the expense of
136 For example, the following command lines create a user space tracing
137 channel with an infinite blocking timeout and run an application
138 instrumented with LTTng-UST which is explicitly allowed to block:
144 $ lttng enable-channel --userspace --blocking-timeout=inf blocking-channel
145 $ lttng enable-event --userspace --channel=blocking-channel --all
147 $ LTTNG_UST_ALLOW_BLOCKING=1 my-app
151 See the complete <<blocking-timeout-example,blocking timeout example>>.
153 * **Linux kernel tracing**:
154 ** Linux 4.10, 4.11, and 4.12 support.
155 ** The thread state dump events recorded by LTTng-modules now contain
156 the task's CPU identifier. This improves the precision of the
157 scheduler model for analyses.
158 ** Extended man:socketpair(2) system call tracing data.
164 What is LTTng? As its name suggests, the _Linux Trace Toolkit: next
165 generation_ is a modern toolkit for tracing Linux systems and
166 applications. So your first question might be:
173 As the history of software engineering progressed and led to what
174 we now take for granted--complex, numerous and
175 interdependent software applications running in parallel on
176 sophisticated operating systems like Linux--the authors of such
177 components, software developers, began feeling a natural
178 urge to have tools that would ensure the robustness and good performance
179 of their masterpieces.
181 One major achievement in this field is, inarguably, the
182 https://www.gnu.org/software/gdb/[GNU debugger (GDB)],
183 an essential tool for developers to find and fix bugs. But even the best
184 debugger won't help make your software run faster, and nowadays, faster
185 software means either more work done by the same hardware, or cheaper
186 hardware for the same work.
188 A _profiler_ is often the tool of choice to identify performance
189 bottlenecks. Profiling is suitable to identify _where_ performance is
190 lost in a given software. The profiler outputs a profile, a statistical
191 summary of observed events, which you may use to discover which
192 functions took the most time to execute. However, a profiler won't
193 report _why_ some identified functions are the bottleneck. Bottlenecks
194 might only occur when specific conditions are met, conditions that are
195 sometimes impossible to capture by a statistical profiler, or impossible
196 to reproduce with an application altered by the overhead of an
197 event-based profiler. For a thorough investigation of software
198 performance issues, a history of execution is essential, with the
199 recorded values of variables and context fields you choose, and
200 with as little influence as possible on the instrumented software. This
201 is where tracing comes in handy.
203 _Tracing_ is a technique used to understand what goes on in a running
204 software system. The software used for tracing is called a _tracer_,
205 which is conceptually similar to a tape recorder. When recording,
206 specific instrumentation points placed in the software source code
207 generate events that are saved on a giant tape: a _trace_ file. You
208 can trace user applications and the operating system at the same time,
209 opening the possibility of resolving a wide range of problems that would
210 otherwise be extremely challenging.
212 Tracing is often compared to _logging_. However, tracers and loggers are
213 two different tools, serving two different purposes. Tracers are
214 designed to record much lower-level events that occur much more
215 frequently than log messages, often in the range of thousands per
216 second, with very little execution overhead. Logging is more appropriate
217 for a very high-level analysis of less frequent events: user accesses,
218 exceptional conditions (errors and warnings, for example), database
219 transactions, instant messaging communications, and such. Simply put,
220 logging is one of the many use cases that can be satisfied with tracing.
222 The list of recorded events inside a trace file can be read manually
223 like a log file for the maximum level of detail, but it is generally
224 much more interesting to perform application-specific analyses to
225 produce reduced statistics and graphs that are useful to resolve a
226 given problem. Trace viewers and analyzers are specialized tools
229 In the end, this is what LTTng is: a powerful, open source set of
230 tools to trace the Linux kernel and user applications at the same time.
231 LTTng is composed of several components actively maintained and
232 developed by its link:/community/#where[community].
235 [[lttng-alternatives]]
236 === Alternatives to noch:{LTTng}
238 Excluding proprietary solutions, a few competing software tracers
241 * https://github.com/dtrace4linux/linux[dtrace4linux] is a port of
242 Sun Microsystems's DTrace to Linux. The cmd:dtrace tool interprets
243 user scripts and is responsible for loading code into the
244 Linux kernel for further execution and collecting the outputted data.
245 * https://en.wikipedia.org/wiki/Berkeley_Packet_Filter[eBPF] is a
246 subsystem in the Linux kernel in which a virtual machine can execute
247 programs passed from the user space to the kernel. You can attach
248 such programs to tracepoints and KProbes thanks to a system call, and
249 they can output data to the user space when executed thanks to
250 different mechanisms (pipe, VM register values, and eBPF maps, to name
252 * https://www.kernel.org/doc/Documentation/trace/ftrace.txt[ftrace]
253 is the de facto function tracer of the Linux kernel. Its user
254 interface is a set of special files in sysfs.
255 * https://perf.wiki.kernel.org/[perf] is
256 a performance analyzing tool for Linux which supports hardware
257 performance counters, tracepoints, as well as other counters and
258 types of probes. perf's controlling utility is the cmd:perf command
260 * http://linux.die.net/man/1/strace[strace]
261 is a command-line utility which records system calls made by a
262 user process, as well as signal deliveries and changes of process
263 state. strace makes use of https://en.wikipedia.org/wiki/Ptrace[ptrace]
264 to fulfill its function.
265 * http://www.sysdig.org/[sysdig], like SystemTap, uses scripts to
266 analyze Linux kernel events. You write scripts, or _chisels_ in
267 sysdig's jargon, in Lua and sysdig executes them while the system is
268 being traced or afterwards. sysdig's interface is the cmd:sysdig
269 command-line tool as well as the curses-based cmd:csysdig tool.
270 * https://sourceware.org/systemtap/[SystemTap] is a Linux kernel and
271 user space tracer which uses custom user scripts to produce plain text
272 traces. SystemTap converts the scripts to the C language, and then
273 compiles them as Linux kernel modules which are loaded to produce
274 trace data. SystemTap's primary user interface is the cmd:stap
277 The main distinctive features of LTTng is that it produces correlated
278 kernel and user space traces, as well as doing so with the lowest
279 overhead amongst other solutions. It produces trace files in the
280 http://diamon.org/ctf[CTF] format, a file format optimized
281 for the production and analyses of multi-gigabyte data.
283 LTTng is the result of more than 10 years of active open source
284 development by a community of passionate developers.
285 LTTng{nbsp}{revision} is currently available on major desktop and server
288 The main interface for tracing control is a single command-line tool
289 named cmd:lttng. The latter can create several tracing sessions, enable
290 and disable events on the fly, filter events efficiently with custom
291 user expressions, start and stop tracing, and much more. LTTng can
292 record the traces on the file system or send them over the network, and
293 keep them totally or partially. You can view the traces once tracing
294 becomes inactive or in real-time.
296 <<installing-lttng,Install LTTng now>> and
297 <<getting-started,start tracing>>!
303 include::../common/warning-no-installation.txt[]
305 **LTTng** is a set of software <<plumbing,components>> which interact to
306 <<instrumenting,instrument>> the Linux kernel and user applications, and
307 to <<controlling-tracing,control tracing>> (start and stop
308 tracing, enable and disable event rules, and the rest). Those
309 components are bundled into the following packages:
311 * **LTTng-tools**: Libraries and command-line interface to
313 * **LTTng-modules**: Linux kernel modules to instrument and
315 * **LTTng-UST**: Libraries and Java/Python packages to instrument and
316 trace user applications.
318 Most distributions mark the LTTng-modules and LTTng-UST packages as
319 optional when installing LTTng-tools (which is always required). Note
322 * You only need to install LTTng-modules if you intend to trace the
324 * You only need to install LTTng-UST if you intend to trace user
328 [[building-from-source]]
329 === Build from source
331 To build and install LTTng{nbsp}{revision} from source:
333 . Using your distribution's package manager, or from source, install
334 the following dependencies of LTTng-tools and LTTng-UST:
337 * https://sourceforge.net/projects/libuuid/[libuuid]
338 * http://directory.fsf.org/wiki/Popt[popt]
339 * http://liburcu.org/[Userspace RCU]
340 * http://www.xmlsoft.org/[libxml2]
343 . Download, build, and install the latest LTTng-modules{nbsp}{revision}:
349 wget http://lttng.org/files/lttng-modules/lttng-modules-latest-2.10.tar.bz2 &&
350 tar -xf lttng-modules-latest-2.10.tar.bz2 &&
351 cd lttng-modules-2.10.* &&
353 sudo make modules_install &&
358 . Download, build, and install the latest LTTng-UST{nbsp}{revision}:
364 wget http://lttng.org/files/lttng-ust/lttng-ust-latest-2.10.tar.bz2 &&
365 tar -xf lttng-ust-latest-2.10.tar.bz2 &&
366 cd lttng-ust-2.10.* &&
376 .Java and Python application tracing
378 If you need to instrument and trace <<java-application,Java
379 applications>>, pass the `--enable-java-agent-jul`,
380 `--enable-java-agent-log4j`, or `--enable-java-agent-all` options to the
381 `configure` script, depending on which Java logging framework you use.
383 If you need to instrument and trace <<python-application,Python
384 applications>>, pass the `--enable-python-agent` option to the
385 `configure` script. You can set the `PYTHON` environment variable to the
386 path to the Python interpreter for which to install the LTTng-UST Python
394 By default, LTTng-UST libraries are installed to
395 dir:{/usr/local/lib}, which is the de facto directory in which to
396 keep self-compiled and third-party libraries.
398 When <<building-tracepoint-providers-and-user-application,linking an
399 instrumented user application with `liblttng-ust`>>:
401 * Append `/usr/local/lib` to the env:LD_LIBRARY_PATH environment
403 * Pass the `-L/usr/local/lib` and `-Wl,-rpath,/usr/local/lib` options to
404 man:gcc(1), man:g++(1), or man:clang(1).
408 . Download, build, and install the latest LTTng-tools{nbsp}{revision}:
414 wget http://lttng.org/files/lttng-tools/lttng-tools-latest-2.10.tar.bz2 &&
415 tar -xf lttng-tools-latest-2.10.tar.bz2 &&
416 cd lttng-tools-2.10.* &&
424 TIP: The https://github.com/eepp/vlttng[vlttng tool] can do all the
425 previous steps automatically for a given version of LTTng and confine
426 the installed files in a specific directory. This can be useful to test
427 LTTng without installing it on your system.
433 This is a short guide to get started quickly with LTTng kernel and user
436 Before you follow this guide, make sure to <<installing-lttng,install>>
439 This tutorial walks you through the steps to:
441 . <<tracing-the-linux-kernel,Trace the Linux kernel>>.
442 . <<tracing-your-own-user-application,Trace a user application>> written
444 . <<viewing-and-analyzing-your-traces,View and analyze the
448 [[tracing-the-linux-kernel]]
449 === Trace the Linux kernel
451 The following command lines start with the `#` prompt because you need
452 root privileges to trace the Linux kernel. You can also trace the kernel
453 as a regular user if your Unix user is a member of the
454 <<tracing-group,tracing group>>.
456 . Create a <<tracing-session,tracing session>> which writes its traces
457 to dir:{/tmp/my-kernel-trace}:
462 # lttng create my-kernel-session --output=/tmp/my-kernel-trace
466 . List the available kernel tracepoints and system calls:
471 # lttng list --kernel
472 # lttng list --kernel --syscall
476 . Create <<event,event rules>> which match the desired instrumentation
477 point names, for example the `sched_switch` and `sched_process_fork`
478 tracepoints, and the man:open(2) and man:close(2) system calls:
483 # lttng enable-event --kernel sched_switch,sched_process_fork
484 # lttng enable-event --kernel --syscall open,close
488 You can also create an event rule which matches _all_ the Linux kernel
489 tracepoints (this will generate a lot of data when tracing):
494 # lttng enable-event --kernel --all
498 . <<basic-tracing-session-control,Start tracing>>:
507 . Do some operation on your system for a few seconds. For example,
508 load a website, or list the files of a directory.
509 . <<creating-destroying-tracing-sessions,Destroy>> the current
519 The man:lttng-destroy(1) command does not destroy the trace data; it
520 only destroys the state of the tracing session.
522 The man:lttng-destroy(1) command also runs the man:lttng-stop(1) command
523 implicitly (see <<basic-tracing-session-control,Start and stop a tracing
524 session>>). You need to stop tracing to make LTTng flush the remaining
525 trace data and make the trace readable.
527 . For the sake of this example, make the recorded trace accessible to
533 # chown -R $(whoami) /tmp/my-kernel-trace
537 See <<viewing-and-analyzing-your-traces,View and analyze the
538 recorded events>> to view the recorded events.
541 [[tracing-your-own-user-application]]
542 === Trace a user application
544 This section steps you through a simple example to trace a
545 _Hello world_ program written in C.
547 To create the traceable user application:
549 . Create the tracepoint provider header file, which defines the
550 tracepoints and the events they can generate:
556 #undef TRACEPOINT_PROVIDER
557 #define TRACEPOINT_PROVIDER hello_world
559 #undef TRACEPOINT_INCLUDE
560 #define TRACEPOINT_INCLUDE "./hello-tp.h"
562 #if !defined(_HELLO_TP_H) || defined(TRACEPOINT_HEADER_MULTI_READ)
565 #include <lttng/tracepoint.h>
575 ctf_string(my_string_field, my_string_arg)
576 ctf_integer(int, my_integer_field, my_integer_arg)
580 #endif /* _HELLO_TP_H */
582 #include <lttng/tracepoint-event.h>
586 . Create the tracepoint provider package source file:
592 #define TRACEPOINT_CREATE_PROBES
593 #define TRACEPOINT_DEFINE
595 #include "hello-tp.h"
599 . Build the tracepoint provider package:
604 $ gcc -c -I. hello-tp.c
608 . Create the _Hello World_ application source file:
615 #include "hello-tp.h"
617 int main(int argc, char *argv[])
621 puts("Hello, World!\nPress Enter to continue...");
624 * The following getchar() call is only placed here for the purpose
625 * of this demonstration, to pause the application in order for
626 * you to have time to list its tracepoints. It is not
632 * A tracepoint() call.
634 * Arguments, as defined in hello-tp.h:
636 * 1. Tracepoint provider name (required)
637 * 2. Tracepoint name (required)
638 * 3. my_integer_arg (first user-defined argument)
639 * 4. my_string_arg (second user-defined argument)
641 * Notice the tracepoint provider and tracepoint names are
642 * NOT strings: they are in fact parts of variables that the
643 * macros in hello-tp.h create.
645 tracepoint(hello_world, my_first_tracepoint, 23, "hi there!");
647 for (x = 0; x < argc; ++x) {
648 tracepoint(hello_world, my_first_tracepoint, x, argv[x]);
651 puts("Quitting now!");
652 tracepoint(hello_world, my_first_tracepoint, x * x, "x^2");
659 . Build the application:
668 . Link the application with the tracepoint provider package,
669 `liblttng-ust`, and `libdl`:
674 $ gcc -o hello hello.o hello-tp.o -llttng-ust -ldl
678 Here's the whole build process:
681 .User space tracing tutorial's build steps.
682 image::ust-flow.png[]
684 To trace the user application:
686 . Run the application with a few arguments:
691 $ ./hello world and beyond
700 Press Enter to continue...
704 . Start an LTTng <<lttng-sessiond,session daemon>>:
709 $ lttng-sessiond --daemonize
713 Note that a session daemon might already be running, for example as
714 a service that the distribution's service manager started.
716 . List the available user space tracepoints:
721 $ lttng list --userspace
725 You see the `hello_world:my_first_tracepoint` tracepoint listed
726 under the `./hello` process.
728 . Create a <<tracing-session,tracing session>>:
733 $ lttng create my-user-space-session
737 . Create an <<event,event rule>> which matches the
738 `hello_world:my_first_tracepoint` event name:
743 $ lttng enable-event --userspace hello_world:my_first_tracepoint
747 . <<basic-tracing-session-control,Start tracing>>:
756 . Go back to the running `hello` application and press Enter. The
757 program executes all `tracepoint()` instrumentation points and exits.
758 . <<creating-destroying-tracing-sessions,Destroy>> the current
768 The man:lttng-destroy(1) command does not destroy the trace data; it
769 only destroys the state of the tracing session.
771 The man:lttng-destroy(1) command also runs the man:lttng-stop(1) command
772 implicitly (see <<basic-tracing-session-control,Start and stop a tracing
773 session>>). You need to stop tracing to make LTTng flush the remaining
774 trace data and make the trace readable.
776 By default, LTTng saves the traces in
777 +$LTTNG_HOME/lttng-traces/__name__-__date__-__time__+,
778 where +__name__+ is the tracing session name. The
779 env:LTTNG_HOME environment variable defaults to `$HOME` if not set.
781 See <<viewing-and-analyzing-your-traces,View and analyze the
782 recorded events>> to view the recorded events.
785 [[viewing-and-analyzing-your-traces]]
786 === View and analyze the recorded events
788 Once you have completed the <<tracing-the-linux-kernel,Trace the Linux
789 kernel>> and <<tracing-your-own-user-application,Trace a user
790 application>> tutorials, you can inspect the recorded events.
792 Many tools are available to read LTTng traces:
794 * **cmd:babeltrace** is a command-line utility which converts trace
795 formats; it supports the format that LTTng produces, CTF, as well as a
796 basic text output which can be ++grep++ed. The cmd:babeltrace command
797 is part of the http://diamon.org/babeltrace[Babeltrace] project.
798 * Babeltrace also includes
799 **https://www.python.org/[Python] bindings** so
800 that you can easily open and read an LTTng trace with your own script,
801 benefiting from the power of Python.
802 * http://tracecompass.org/[**Trace Compass**]
803 is a graphical user interface for viewing and analyzing any type of
804 logs or traces, including LTTng's.
805 * https://github.com/lttng/lttng-analyses[**LTTng analyses**] is a
806 project which includes many high-level analyses of LTTng kernel
807 traces, like scheduling statistics, interrupt frequency distribution,
808 top CPU usage, and more.
810 NOTE: This section assumes that the traces recorded during the previous
811 tutorials were saved to their default location, in the
812 dir:{$LTTNG_HOME/lttng-traces} directory. The env:LTTNG_HOME
813 environment variable defaults to `$HOME` if not set.
816 [[viewing-and-analyzing-your-traces-bt]]
817 ==== Use the cmd:babeltrace command-line tool
819 The simplest way to list all the recorded events of a trace is to pass
820 its path to cmd:babeltrace with no options:
824 $ babeltrace ~/lttng-traces/my-user-space-session*
827 cmd:babeltrace finds all traces recursively within the given path and
828 prints all their events, merging them in chronological order.
830 You can pipe the output of cmd:babeltrace into a tool like man:grep(1) for
835 $ babeltrace /tmp/my-kernel-trace | grep _switch
838 You can pipe the output of cmd:babeltrace into a tool like man:wc(1) to
839 count the recorded events:
843 $ babeltrace /tmp/my-kernel-trace | grep _open | wc --lines
847 [[viewing-and-analyzing-your-traces-bt-python]]
848 ==== Use the Babeltrace Python bindings
850 The <<viewing-and-analyzing-your-traces-bt,text output of cmd:babeltrace>>
851 is useful to isolate events by simple matching using man:grep(1) and
852 similar utilities. However, more elaborate filters, such as keeping only
853 event records with a field value falling within a specific range, are
854 not trivial to write using a shell. Moreover, reductions and even the
855 most basic computations involving multiple event records are virtually
856 impossible to implement.
858 Fortunately, Babeltrace ships with Python 3 bindings which makes it easy
859 to read the event records of an LTTng trace sequentially and compute the
862 The following script accepts an LTTng Linux kernel trace path as its
863 first argument and prints the short names of the top 5 running processes
864 on CPU 0 during the whole trace:
869 from collections import Counter
875 if len(sys.argv) != 2:
876 msg = 'Usage: python3 {} TRACEPATH'.format(sys.argv[0])
877 print(msg, file=sys.stderr)
880 # A trace collection contains one or more traces
881 col = babeltrace.TraceCollection()
883 # Add the trace provided by the user (LTTng traces always have
885 if col.add_trace(sys.argv[1], 'ctf') is None:
886 raise RuntimeError('Cannot add trace')
888 # This counter dict contains execution times:
890 # task command name -> total execution time (ns)
891 exec_times = Counter()
893 # This contains the last `sched_switch` timestamp
897 for event in col.events:
898 # Keep only `sched_switch` events
899 if event.name != 'sched_switch':
902 # Keep only events which happened on CPU 0
903 if event['cpu_id'] != 0:
907 cur_ts = event.timestamp
913 # Previous task command (short) name
914 prev_comm = event['prev_comm']
916 # Initialize entry in our dict if not yet done
917 if prev_comm not in exec_times:
918 exec_times[prev_comm] = 0
920 # Compute previous command execution time
921 diff = cur_ts - last_ts
923 # Update execution time of this command
924 exec_times[prev_comm] += diff
926 # Update last timestamp
930 for name, ns in exec_times.most_common(5):
932 print('{:20}{} s'.format(name, s))
937 if __name__ == '__main__':
938 sys.exit(0 if top5proc() else 1)
945 $ python3 top5proc.py /tmp/my-kernel-trace/kernel
951 swapper/0 48.607245889 s
952 chromium 7.192738188 s
953 pavucontrol 0.709894415 s
954 Compositor 0.660867933 s
955 Xorg.bin 0.616753786 s
958 Note that `swapper/0` is the "idle" process of CPU 0 on Linux; since we
959 weren't using the CPU that much when tracing, its first position in the
964 == [[understanding-lttng]]Core concepts
966 From a user's perspective, the LTTng system is built on a few concepts,
967 or objects, on which the <<lttng-cli,cmd:lttng command-line tool>>
968 operates by sending commands to the <<lttng-sessiond,session daemon>>.
969 Understanding how those objects relate to eachother is key in mastering
972 The core concepts are:
974 * <<tracing-session,Tracing session>>
975 * <<domain,Tracing domain>>
976 * <<channel,Channel and ring buffer>>
977 * <<"event","Instrumentation point, event rule, event, and event record">>
983 A _tracing session_ is a stateful dialogue between you and
984 a <<lttng-sessiond,session daemon>>. You can
985 <<creating-destroying-tracing-sessions,create a new tracing
986 session>> with the `lttng create` command.
988 Anything that you do when you control LTTng tracers happens within a
989 tracing session. In particular, a tracing session:
992 * Has its own set of trace files.
993 * Has its own state of activity (started or stopped).
994 * Has its own <<tracing-session-mode,mode>> (local, network streaming,
996 * Has its own <<channel,channels>> which have their own
997 <<event,event rules>>.
1000 .A _tracing session_ contains <<channel,channels>> that are members of <<domain,tracing domains>> and contain <<event,event rules>>.
1001 image::concepts.png[]
1003 Those attributes and objects are completely isolated between different
1006 A tracing session is analogous to a cash machine session:
1007 the operations you do on the banking system through the cash machine do
1008 not alter the data of other users of the same system. In the case of
1009 the cash machine, a session lasts as long as your bank card is inside.
1010 In the case of LTTng, a tracing session lasts from the `lttng create`
1011 command to the `lttng destroy` command.
1014 .Each Unix user has its own set of tracing sessions.
1015 image::many-sessions.png[]
1018 [[tracing-session-mode]]
1019 ==== Tracing session mode
1021 LTTng can send the generated trace data to different locations. The
1022 _tracing session mode_ dictates where to send it. The following modes
1023 are available in LTTng{nbsp}{revision}:
1025 [[local-mode]]Local mode::
1026 LTTng writes the traces to the file system of the machine being traced
1029 [[net-streaming-mode]]Network streaming mode::
1030 LTTng sends the traces over the network to a
1031 <<lttng-relayd,relay daemon>> running on a remote system.
1034 LTTng does not write the traces by default. Instead, you can request
1035 LTTng to <<taking-a-snapshot,take a snapshot>>, that is, a copy of the
1036 current tracing buffers, and to write it to the target's file system
1037 or to send it over the network to a <<lttng-relayd,relay daemon>>
1038 running on a remote system.
1040 [[live-mode]]Live mode::
1041 This mode is similar to the network streaming mode, but a live
1042 trace viewer can connect to the distant relay daemon to
1043 <<lttng-live,view event records as LTTng generates them>> by
1050 A _tracing domain_ is a namespace for event sources. A tracing domain
1051 has its own properties and features.
1053 There are currently five available tracing domains:
1057 * `java.util.logging` (JUL)
1061 You must specify a tracing domain when using some commands to avoid
1062 ambiguity. For example, since all the domains support named tracepoints
1063 as event sources (instrumentation points that you manually insert in the
1064 source code), you need to specify a tracing domain when
1065 <<enabling-disabling-events,creating an event rule>> because all the
1066 tracing domains could have tracepoints with the same names.
1068 Some features are reserved to specific tracing domains. Dynamic function
1069 entry and return instrumentation points, for example, are currently only
1070 supported in the Linux kernel tracing domain, but support for other
1071 tracing domains could be added in the future.
1073 You can create <<channel,channels>> in the Linux kernel and user space
1074 tracing domains. The other tracing domains have a single default
1079 === Channel and ring buffer
1081 A _channel_ is an object which is responsible for a set of ring buffers.
1082 Each ring buffer is divided into multiple sub-buffers. When an LTTng
1083 tracer emits an event, it can record it to one or more
1084 sub-buffers. The attributes of a channel determine what to do when
1085 there's no space left for a new event record because all sub-buffers
1086 are full, where to send a full sub-buffer, and other behaviours.
1088 A channel is always associated to a <<domain,tracing domain>>. The
1089 `java.util.logging` (JUL), log4j, and Python tracing domains each have
1090 a default channel which you cannot configure.
1092 A channel also owns <<event,event rules>>. When an LTTng tracer emits
1093 an event, it records it to the sub-buffers of all
1094 the enabled channels with a satisfied event rule, as long as those
1095 channels are part of active <<tracing-session,tracing sessions>>.
1098 [[channel-buffering-schemes]]
1099 ==== Per-user vs. per-process buffering schemes
1101 A channel has at least one ring buffer _per CPU_. LTTng always
1102 records an event to the ring buffer associated to the CPU on which it
1105 Two _buffering schemes_ are available when you
1106 <<enabling-disabling-channels,create a channel>> in the
1107 user space <<domain,tracing domain>>:
1109 Per-user buffering::
1110 Allocate one set of ring buffers--one per CPU--shared by all the
1111 instrumented processes of each Unix user.
1115 .Per-user buffering scheme.
1116 image::per-user-buffering.png[]
1119 Per-process buffering::
1120 Allocate one set of ring buffers--one per CPU--for each
1121 instrumented process.
1125 .Per-process buffering scheme.
1126 image::per-process-buffering.png[]
1129 The per-process buffering scheme tends to consume more memory than the
1130 per-user option because systems generally have more instrumented
1131 processes than Unix users running instrumented processes. However, the
1132 per-process buffering scheme ensures that one process having a high
1133 event throughput won't fill all the shared sub-buffers of the same
1136 The Linux kernel tracing domain has only one available buffering scheme
1137 which is to allocate a single set of ring buffers for the whole system.
1138 This scheme is similar to the per-user option, but with a single, global
1139 user "running" the kernel.
1142 [[channel-overwrite-mode-vs-discard-mode]]
1143 ==== Overwrite vs. discard event loss modes
1145 When an event occurs, LTTng records it to a specific sub-buffer (yellow
1146 arc in the following animation) of a specific channel's ring buffer.
1147 When there's no space left in a sub-buffer, the tracer marks it as
1148 consumable (red) and another, empty sub-buffer starts receiving the
1149 following event records. A <<lttng-consumerd,consumer daemon>>
1150 eventually consumes the marked sub-buffer (returns to white).
1153 [role="docsvg-channel-subbuf-anim"]
1158 In an ideal world, sub-buffers are consumed faster than they are filled,
1159 as is the case in the previous animation. In the real world,
1160 however, all sub-buffers can be full at some point, leaving no space to
1161 record the following events.
1163 By default, LTTng-modules and LTTng-UST are _non-blocking_ tracers: when
1164 no empty sub-buffer is available, it is acceptable to lose event records
1165 when the alternative would be to cause substantial delays in the
1166 instrumented application's execution. LTTng privileges performance over
1167 integrity; it aims at perturbing the traced system as little as possible
1168 in order to make tracing of subtle race conditions and rare interrupt
1171 Starting from LTTng{nbsp}2.10, the LTTng user space tracer, LTTng-UST,
1172 supports a _blocking mode_. See the <<blocking-timeout-example,blocking
1173 timeout example>> to learn how to use the blocking mode.
1175 When it comes to losing event records because no empty sub-buffer is
1176 available, or because the <<opt-blocking-timeout,blocking timeout>> is
1177 reached, the channel's _event loss mode_ determines what to do. The
1178 available event loss modes are:
1181 Drop the newest event records until a the tracer releases a
1184 This is the only available mode when you specify a
1185 <<opt-blocking-timeout,blocking timeout>>.
1188 Clear the sub-buffer containing the oldest event records and start
1189 writing the newest event records there.
1191 This mode is sometimes called _flight recorder mode_ because it's
1193 https://en.wikipedia.org/wiki/Flight_recorder[flight recorder]:
1194 always keep a fixed amount of the latest data.
1196 Which mechanism you should choose depends on your context: prioritize
1197 the newest or the oldest event records in the ring buffer?
1199 Beware that, in overwrite mode, the tracer abandons a _whole sub-buffer_
1200 as soon as a there's no space left for a new event record, whereas in
1201 discard mode, the tracer only discards the event record that doesn't
1204 In discard mode, LTTng increments a count of lost event records when an
1205 event record is lost and saves this count to the trace. Since
1206 LTTng{nbsp}2.8, in overwrite mode, LTTng writes to a given sub-buffer
1207 its sequence number within its data stream. With a <<local-mode,local>>,
1208 <<net-streaming-mode,network streaming>>, or <<live-mode,live>>
1209 <<tracing-session,tracing session>>, a trace reader can use such
1210 sequence numbers to report lost packets. In overwrite mode, LTTng
1211 doesn't write to the trace the exact number of lost event records in
1212 those lost sub-buffers.
1214 Trace analyses can use saved discarded event record and sub-buffer
1215 (packet) counts of the trace to decide whether or not to perform the
1216 analyses even if trace data is known to be missing.
1218 There are a few ways to decrease your probability of losing event
1220 <<channel-subbuf-size-vs-subbuf-count,Sub-buffer count and size>> shows
1221 how you can fine-tune the sub-buffer count and size of a channel to
1222 virtually stop losing event records, though at the cost of greater
1226 [[channel-subbuf-size-vs-subbuf-count]]
1227 ==== Sub-buffer count and size
1229 When you <<enabling-disabling-channels,create a channel>>, you can
1230 set its number of sub-buffers and their size.
1232 Note that there is noticeable CPU overhead introduced when
1233 switching sub-buffers (marking a full one as consumable and switching
1234 to an empty one for the following events to be recorded). Knowing this,
1235 the following list presents a few practical situations along with how
1236 to configure the sub-buffer count and size for them:
1238 * **High event throughput**: In general, prefer bigger sub-buffers to
1239 lower the risk of losing event records.
1241 Having bigger sub-buffers also ensures a lower
1242 <<channel-switch-timer,sub-buffer switching frequency>>.
1244 The number of sub-buffers is only meaningful if you create the channel
1245 in overwrite mode: in this case, if a sub-buffer overwrite happens, the
1246 other sub-buffers are left unaltered.
1248 * **Low event throughput**: In general, prefer smaller sub-buffers
1249 since the risk of losing event records is low.
1251 Because events occur less frequently, the sub-buffer switching frequency
1252 should remain low and thus the tracer's overhead should not be a
1255 * **Low memory system**: If your target system has a low memory
1256 limit, prefer fewer first, then smaller sub-buffers.
1258 Even if the system is limited in memory, you want to keep the
1259 sub-buffers as big as possible to avoid a high sub-buffer switching
1262 Note that LTTng uses http://diamon.org/ctf/[CTF] as its trace format,
1263 which means event data is very compact. For example, the average
1264 LTTng kernel event record weights about 32{nbsp}bytes. Thus, a
1265 sub-buffer size of 1{nbsp}MiB is considered big.
1267 The previous situations highlight the major trade-off between a few big
1268 sub-buffers and more, smaller sub-buffers: sub-buffer switching
1269 frequency vs. how much data is lost in overwrite mode. Assuming a
1270 constant event throughput and using the overwrite mode, the two
1271 following configurations have the same ring buffer total size:
1274 [role="docsvg-channel-subbuf-size-vs-count-anim"]
1279 * **2 sub-buffers of 4{nbsp}MiB each**: Expect a very low sub-buffer
1280 switching frequency, but if a sub-buffer overwrite happens, half of
1281 the event records so far (4{nbsp}MiB) are definitely lost.
1282 * **8 sub-buffers of 1{nbsp}MiB each**: Expect 4{nbsp}times the tracer's
1283 overhead as the previous configuration, but if a sub-buffer
1284 overwrite happens, only the eighth of event records so far are
1287 In discard mode, the sub-buffers count parameter is pointless: use two
1288 sub-buffers and set their size according to the requirements of your
1292 [[channel-switch-timer]]
1293 ==== Switch timer period
1295 The _switch timer period_ is an important configurable attribute of
1296 a channel to ensure periodic sub-buffer flushing.
1298 When the _switch timer_ expires, a sub-buffer switch happens. You can
1299 set the switch timer period attribute when you
1300 <<enabling-disabling-channels,create a channel>> to ensure that event
1301 data is consumed and committed to trace files or to a distant relay
1302 daemon periodically in case of a low event throughput.
1305 [role="docsvg-channel-switch-timer"]
1310 This attribute is also convenient when you use big sub-buffers to cope
1311 with a sporadic high event throughput, even if the throughput is
1315 [[channel-read-timer]]
1316 ==== Read timer period
1318 By default, the LTTng tracers use a notification mechanism to signal a
1319 full sub-buffer so that a consumer daemon can consume it. When such
1320 notifications must be avoided, for example in real-time applications,
1321 you can use the channel's _read timer_ instead. When the read timer
1322 fires, the <<lttng-consumerd,consumer daemon>> checks for full,
1323 consumable sub-buffers.
1326 [[tracefile-rotation]]
1327 ==== Trace file count and size
1329 By default, trace files can grow as large as needed. You can set the
1330 maximum size of each trace file that a channel writes when you
1331 <<enabling-disabling-channels,create a channel>>. When the size of
1332 a trace file reaches the channel's fixed maximum size, LTTng creates
1333 another file to contain the next event records. LTTng appends a file
1334 count to each trace file name in this case.
1336 If you set the trace file size attribute when you create a channel, the
1337 maximum number of trace files that LTTng creates is _unlimited_ by
1338 default. To limit them, you can also set a maximum number of trace
1339 files. When the number of trace files reaches the channel's fixed
1340 maximum count, the oldest trace file is overwritten. This mechanism is
1341 called _trace file rotation_.
1345 === Instrumentation point, event rule, event, and event record
1347 An _event rule_ is a set of conditions which must be **all** satisfied
1348 for LTTng to record an occuring event.
1350 You set the conditions when you <<enabling-disabling-events,create
1353 You always attach an event rule to <<channel,channel>> when you create
1356 When an event passes the conditions of an event rule, LTTng records it
1357 in one of the attached channel's sub-buffers.
1359 The available conditions, as of LTTng{nbsp}{revision}, are:
1361 * The event rule _is enabled_.
1362 * The instrumentation point's type _is{nbsp}T_.
1363 * The instrumentation point's name (sometimes called _event name_)
1364 _matches{nbsp}N_, but _is not{nbsp}E_.
1365 * The instrumentation point's log level _is as severe as{nbsp}L_, or
1366 _is exactly{nbsp}L_.
1367 * The fields of the event's payload _satisfy_ a filter
1368 expression{nbsp}__F__.
1370 As you can see, all the conditions but the dynamic filter are related to
1371 the event rule's status or to the instrumentation point, not to the
1372 occurring events. This is why, without a filter, checking if an event
1373 passes an event rule is not a dynamic task: when you create or modify an
1374 event rule, all the tracers of its tracing domain enable or disable the
1375 instrumentation points themselves once. This is possible because the
1376 attributes of an instrumentation point (type, name, and log level) are
1377 defined statically. In other words, without a dynamic filter, the tracer
1378 _does not evaluate_ the arguments of an instrumentation point unless it
1379 matches an enabled event rule.
1381 Note that, for LTTng to record an event, the <<channel,channel>> to
1382 which a matching event rule is attached must also be enabled, and the
1383 tracing session owning this channel must be active.
1386 .Logical path from an instrumentation point to an event record.
1387 image::event-rule.png[]
1389 .Event, event record, or event rule?
1391 With so many similar terms, it's easy to get confused.
1393 An **event** is the consequence of the execution of an _instrumentation
1394 point_, like a tracepoint that you manually place in some source code,
1395 or a Linux kernel KProbe. An event is said to _occur_ at a specific
1396 time. Different actions can be taken upon the occurrence of an event,
1397 like record the event's payload to a buffer.
1399 An **event record** is the representation of an event in a sub-buffer. A
1400 tracer is responsible for capturing the payload of an event, current
1401 context variables, the event's ID, and the event's timestamp. LTTng
1402 can append this sub-buffer to a trace file.
1404 An **event rule** is a set of conditions which must all be satisfied for
1405 LTTng to record an occuring event. Events still occur without
1406 satisfying event rules, but LTTng does not record them.
1411 == Components of noch:{LTTng}
1413 The second _T_ in _LTTng_ stands for _toolkit_: it would be wrong
1414 to call LTTng a simple _tool_ since it is composed of multiple
1415 interacting components. This section describes those components,
1416 explains their respective roles, and shows how they connect together to
1417 form the LTTng ecosystem.
1419 The following diagram shows how the most important components of LTTng
1420 interact with user applications, the Linux kernel, and you:
1423 .Control and trace data paths between LTTng components.
1424 image::plumbing.png[]
1426 The LTTng project incorporates:
1428 * **LTTng-tools**: Libraries and command-line interface to
1429 control tracing sessions.
1430 ** <<lttng-sessiond,Session daemon>> (man:lttng-sessiond(8)).
1431 ** <<lttng-consumerd,Consumer daemon>> (cmd:lttng-consumerd).
1432 ** <<lttng-relayd,Relay daemon>> (man:lttng-relayd(8)).
1433 ** <<liblttng-ctl-lttng,Tracing control library>> (`liblttng-ctl`).
1434 ** <<lttng-cli,Tracing control command-line tool>> (man:lttng(1)).
1435 * **LTTng-UST**: Libraries and Java/Python packages to trace user
1437 ** <<lttng-ust,User space tracing library>> (`liblttng-ust`) and its
1438 headers to instrument and trace any native user application.
1439 ** <<prebuilt-ust-helpers,Preloadable user space tracing helpers>>:
1440 *** `liblttng-ust-libc-wrapper`
1441 *** `liblttng-ust-pthread-wrapper`
1442 *** `liblttng-ust-cyg-profile`
1443 *** `liblttng-ust-cyg-profile-fast`
1444 *** `liblttng-ust-dl`
1445 ** User space tracepoint provider source files generator command-line
1446 tool (man:lttng-gen-tp(1)).
1447 ** <<lttng-ust-agents,LTTng-UST Java agent>> to instrument and trace
1448 Java applications using `java.util.logging` or
1449 Apache log4j 1.2 logging.
1450 ** <<lttng-ust-agents,LTTng-UST Python agent>> to instrument
1451 Python applications using the standard `logging` package.
1452 * **LTTng-modules**: <<lttng-modules,Linux kernel modules>> to trace
1454 ** LTTng kernel tracer module.
1455 ** Tracing ring buffer kernel modules.
1456 ** Probe kernel modules.
1457 ** LTTng logger kernel module.
1461 === Tracing control command-line interface
1464 .The tracing control command-line interface.
1465 image::plumbing-lttng-cli.png[]
1467 The _man:lttng(1) command-line tool_ is the standard user interface to
1468 control LTTng <<tracing-session,tracing sessions>>. The cmd:lttng tool
1469 is part of LTTng-tools.
1471 The cmd:lttng tool is linked with
1472 <<liblttng-ctl-lttng,`liblttng-ctl`>> to communicate with
1473 one or more <<lttng-sessiond,session daemons>> behind the scenes.
1475 The cmd:lttng tool has a Git-like interface:
1479 $ lttng <GENERAL OPTIONS> <COMMAND> <COMMAND OPTIONS>
1482 The <<controlling-tracing,Tracing control>> section explores the
1483 available features of LTTng using the cmd:lttng tool.
1486 [[liblttng-ctl-lttng]]
1487 === Tracing control library
1490 .The tracing control library.
1491 image::plumbing-liblttng-ctl.png[]
1493 The _LTTng control library_, `liblttng-ctl`, is used to communicate
1494 with a <<lttng-sessiond,session daemon>> using a C API that hides the
1495 underlying protocol's details. `liblttng-ctl` is part of LTTng-tools.
1497 The <<lttng-cli,cmd:lttng command-line tool>>
1498 is linked with `liblttng-ctl`.
1500 You can use `liblttng-ctl` in C or $$C++$$ source code by including its
1505 #include <lttng/lttng.h>
1508 Some objects are referenced by name (C string), such as tracing
1509 sessions, but most of them require to create a handle first using
1510 `lttng_create_handle()`.
1512 The best available developer documentation for `liblttng-ctl` is, as of
1513 LTTng{nbsp}{revision}, its installed header files. Every function and
1514 structure is thoroughly documented.
1518 === User space tracing library
1521 .The user space tracing library.
1522 image::plumbing-liblttng-ust.png[]
1524 The _user space tracing library_, `liblttng-ust` (see man:lttng-ust(3)),
1525 is the LTTng user space tracer. It receives commands from a
1526 <<lttng-sessiond,session daemon>>, for example to
1527 enable and disable specific instrumentation points, and writes event
1528 records to ring buffers shared with a
1529 <<lttng-consumerd,consumer daemon>>.
1530 `liblttng-ust` is part of LTTng-UST.
1532 Public C header files are installed beside `liblttng-ust` to
1533 instrument any <<c-application,C or $$C++$$ application>>.
1535 <<lttng-ust-agents,LTTng-UST agents>>, which are regular Java and Python
1536 packages, use their own library providing tracepoints which is
1537 linked with `liblttng-ust`.
1539 An application or library does not have to initialize `liblttng-ust`
1540 manually: its constructor does the necessary tasks to properly register
1541 to a session daemon. The initialization phase also enables the
1542 instrumentation points matching the <<event,event rules>> that you
1546 [[lttng-ust-agents]]
1547 === User space tracing agents
1550 .The user space tracing agents.
1551 image::plumbing-lttng-ust-agents.png[]
1553 The _LTTng-UST Java and Python agents_ are regular Java and Python
1554 packages which add LTTng tracing capabilities to the
1555 native logging frameworks. The LTTng-UST agents are part of LTTng-UST.
1557 In the case of Java, the
1558 https://docs.oracle.com/javase/7/docs/api/java/util/logging/package-summary.html[`java.util.logging`
1559 core logging facilities] and
1560 https://logging.apache.org/log4j/1.2/[Apache log4j 1.2] are supported.
1561 Note that Apache Log4{nbsp}2 is not supported.
1563 In the case of Python, the standard
1564 https://docs.python.org/3/library/logging.html[`logging`] package
1565 is supported. Both Python 2 and Python 3 modules can import the
1566 LTTng-UST Python agent package.
1568 The applications using the LTTng-UST agents are in the
1569 `java.util.logging` (JUL),
1570 log4j, and Python <<domain,tracing domains>>.
1572 Both agents use the same mechanism to trace the log statements. When an
1573 agent is initialized, it creates a log handler that attaches to the root
1574 logger. The agent also registers to a <<lttng-sessiond,session daemon>>.
1575 When the application executes a log statement, it is passed to the
1576 agent's log handler by the root logger. The agent's log handler calls a
1577 native function in a tracepoint provider package shared library linked
1578 with <<lttng-ust,`liblttng-ust`>>, passing the formatted log message and
1579 other fields, like its logger name and its log level. This native
1580 function contains a user space instrumentation point, hence tracing the
1583 The log level condition of an
1584 <<event,event rule>> is considered when tracing
1585 a Java or a Python application, and it's compatible with the standard
1586 JUL, log4j, and Python log levels.
1590 === LTTng kernel modules
1593 .The LTTng kernel modules.
1594 image::plumbing-lttng-modules.png[]
1596 The _LTTng kernel modules_ are a set of Linux kernel modules
1597 which implement the kernel tracer of the LTTng project. The LTTng
1598 kernel modules are part of LTTng-modules.
1600 The LTTng kernel modules include:
1602 * A set of _probe_ modules.
1604 Each module attaches to a specific subsystem
1605 of the Linux kernel using its tracepoint instrument points. There are
1606 also modules to attach to the entry and return points of the Linux
1607 system call functions.
1609 * _Ring buffer_ modules.
1611 A ring buffer implementation is provided as kernel modules. The LTTng
1612 kernel tracer writes to the ring buffer; a
1613 <<lttng-consumerd,consumer daemon>> reads from the ring buffer.
1615 * The _LTTng kernel tracer_ module.
1616 * The _LTTng logger_ module.
1618 The LTTng logger module implements the special path:{/proc/lttng-logger}
1619 file so that any executable can generate LTTng events by opening and
1620 writing to this file.
1622 See <<proc-lttng-logger-abi,LTTng logger>>.
1624 Generally, you do not have to load the LTTng kernel modules manually
1625 (using man:modprobe(8), for example): a root <<lttng-sessiond,session
1626 daemon>> loads the necessary modules when starting. If you have extra
1627 probe modules, you can specify to load them to the session daemon on
1630 The LTTng kernel modules are installed in
1631 +/usr/lib/modules/__release__/extra+ by default, where +__release__+ is
1632 the kernel release (see `uname --kernel-release`).
1639 .The session daemon.
1640 image::plumbing-sessiond.png[]
1642 The _session daemon_, man:lttng-sessiond(8), is a daemon responsible for
1643 managing tracing sessions and for controlling the various components of
1644 LTTng. The session daemon is part of LTTng-tools.
1646 The session daemon sends control requests to and receives control
1649 * The <<lttng-ust,user space tracing library>>.
1651 Any instance of the user space tracing library first registers to
1652 a session daemon. Then, the session daemon can send requests to
1653 this instance, such as:
1656 ** Get the list of tracepoints.
1657 ** Share an <<event,event rule>> so that the user space tracing library
1658 can enable or disable tracepoints. Amongst the possible conditions
1659 of an event rule is a filter expression which `liblttng-ust` evalutes
1660 when an event occurs.
1661 ** Share <<channel,channel>> attributes and ring buffer locations.
1664 The session daemon and the user space tracing library use a Unix
1665 domain socket for their communication.
1667 * The <<lttng-ust-agents,user space tracing agents>>.
1669 Any instance of a user space tracing agent first registers to
1670 a session daemon. Then, the session daemon can send requests to
1671 this instance, such as:
1674 ** Get the list of loggers.
1675 ** Enable or disable a specific logger.
1678 The session daemon and the user space tracing agent use a TCP connection
1679 for their communication.
1681 * The <<lttng-modules,LTTng kernel tracer>>.
1682 * The <<lttng-consumerd,consumer daemon>>.
1684 The session daemon sends requests to the consumer daemon to instruct
1685 it where to send the trace data streams, amongst other information.
1687 * The <<lttng-relayd,relay daemon>>.
1689 The session daemon receives commands from the
1690 <<liblttng-ctl-lttng,tracing control library>>.
1692 The root session daemon loads the appropriate
1693 <<lttng-modules,LTTng kernel modules>> on startup. It also spawns
1694 a <<lttng-consumerd,consumer daemon>> as soon as you create
1695 an <<event,event rule>>.
1697 The session daemon does not send and receive trace data: this is the
1698 role of the <<lttng-consumerd,consumer daemon>> and
1699 <<lttng-relayd,relay daemon>>. It does, however, generate the
1700 http://diamon.org/ctf/[CTF] metadata stream.
1702 Each Unix user can have its own session daemon instance. The
1703 tracing sessions managed by different session daemons are completely
1706 The root user's session daemon is the only one which is
1707 allowed to control the LTTng kernel tracer, and its spawned consumer
1708 daemon is the only one which is allowed to consume trace data from the
1709 LTTng kernel tracer. Note, however, that any Unix user which is a member
1710 of the <<tracing-group,tracing group>> is allowed
1711 to create <<channel,channels>> in the
1712 Linux kernel <<domain,tracing domain>>, and thus to trace the Linux
1715 The <<lttng-cli,cmd:lttng command-line tool>> automatically starts a
1716 session daemon when using its `create` command if none is currently
1717 running. You can also start the session daemon manually.
1724 .The consumer daemon.
1725 image::plumbing-consumerd.png[]
1727 The _consumer daemon_, cmd:lttng-consumerd, is a daemon which shares
1728 ring buffers with user applications or with the LTTng kernel modules to
1729 collect trace data and send it to some location (on disk or to a
1730 <<lttng-relayd,relay daemon>> over the network). The consumer daemon
1731 is part of LTTng-tools.
1733 You do not start a consumer daemon manually: a consumer daemon is always
1734 spawned by a <<lttng-sessiond,session daemon>> as soon as you create an
1735 <<event,event rule>>, that is, before you start tracing. When you kill
1736 its owner session daemon, the consumer daemon also exits because it is
1737 the session daemon's child process. Command-line options of
1738 man:lttng-sessiond(8) target the consumer daemon process.
1740 There are up to two running consumer daemons per Unix user, whereas only
1741 one session daemon can run per user. This is because each process can be
1742 either 32-bit or 64-bit: if the target system runs a mixture of 32-bit
1743 and 64-bit processes, it is more efficient to have separate
1744 corresponding 32-bit and 64-bit consumer daemons. The root user is an
1745 exception: it can have up to _three_ running consumer daemons: 32-bit
1746 and 64-bit instances for its user applications, and one more
1747 reserved for collecting kernel trace data.
1755 image::plumbing-relayd.png[]
1757 The _relay daemon_, man:lttng-relayd(8), is a daemon acting as a bridge
1758 between remote session and consumer daemons, local trace files, and a
1759 remote live trace viewer. The relay daemon is part of LTTng-tools.
1761 The main purpose of the relay daemon is to implement a receiver of
1762 <<sending-trace-data-over-the-network,trace data over the network>>.
1763 This is useful when the target system does not have much file system
1764 space to record trace files locally.
1766 The relay daemon is also a server to which a
1767 <<lttng-live,live trace viewer>> can
1768 connect. The live trace viewer sends requests to the relay daemon to
1769 receive trace data as the target system emits events. The
1770 communication protocol is named _LTTng live_; it is used over TCP
1773 Note that you can start the relay daemon on the target system directly.
1774 This is the setup of choice when the use case is to view events as
1775 the target system emits them without the need of a remote system.
1779 == [[using-lttng]]Instrumentation
1781 There are many examples of tracing and monitoring in our everyday life:
1783 * You have access to real-time and historical weather reports and
1784 forecasts thanks to weather stations installed around the country.
1785 * You know your heart is safe thanks to an electrocardiogram.
1786 * You make sure not to drive your car too fast and to have enough fuel
1787 to reach your destination thanks to gauges visible on your dashboard.
1789 All the previous examples have something in common: they rely on
1790 **instruments**. Without the electrodes attached to the surface of your
1791 body's skin, cardiac monitoring is futile.
1793 LTTng, as a tracer, is no different from those real life examples. If
1794 you're about to trace a software system or, in other words, record its
1795 history of execution, you better have **instrumentation points** in the
1796 subject you're tracing, that is, the actual software.
1798 Various ways were developed to instrument a piece of software for LTTng
1799 tracing. The most straightforward one is to manually place
1800 instrumentation points, called _tracepoints_, in the software's source
1801 code. It is also possible to add instrumentation points dynamically in
1802 the Linux kernel <<domain,tracing domain>>.
1804 If you're only interested in tracing the Linux kernel, your
1805 instrumentation needs are probably already covered by LTTng's built-in
1806 <<lttng-modules,Linux kernel tracepoints>>. You may also wish to trace a
1807 user application which is already instrumented for LTTng tracing.
1808 In such cases, you can skip this whole section and read the topics of
1809 the <<controlling-tracing,Tracing control>> section.
1811 Many methods are available to instrument a piece of software for LTTng
1814 * <<c-application,User space instrumentation for C and $$C++$$
1816 * <<prebuilt-ust-helpers,Prebuilt user space tracing helpers>>.
1817 * <<java-application,User space Java agent>>.
1818 * <<python-application,User space Python agent>>.
1819 * <<proc-lttng-logger-abi,LTTng logger>>.
1820 * <<instrumenting-linux-kernel,LTTng kernel tracepoints>>.
1824 === [[cxx-application]]User space instrumentation for C and $$C++$$ applications
1826 The procedure to instrument a C or $$C++$$ user application with
1827 the <<lttng-ust,LTTng user space tracing library>>, `liblttng-ust`, is:
1829 . <<tracepoint-provider,Create the source files of a tracepoint provider
1831 . <<probing-the-application-source-code,Add tracepoints to
1832 the application's source code>>.
1833 . <<building-tracepoint-providers-and-user-application,Build and link
1834 a tracepoint provider package and the user application>>.
1836 If you need quick, man:printf(3)-like instrumentation, you can skip
1837 those steps and use <<tracef,`tracef()`>> or <<tracelog,`tracelog()`>>
1840 IMPORTANT: You need to <<installing-lttng,install>> LTTng-UST to
1841 instrument a user application with `liblttng-ust`.
1844 [[tracepoint-provider]]
1845 ==== Create the source files of a tracepoint provider package
1847 A _tracepoint provider_ is a set of compiled functions which provide
1848 **tracepoints** to an application, the type of instrumentation point
1849 supported by LTTng-UST. Those functions can emit events with
1850 user-defined fields and serialize those events as event records to one
1851 or more LTTng-UST <<channel,channel>> sub-buffers. The `tracepoint()`
1852 macro, which you <<probing-the-application-source-code,insert in a user
1853 application's source code>>, calls those functions.
1855 A _tracepoint provider package_ is an object file (`.o`) or a shared
1856 library (`.so`) which contains one or more tracepoint providers.
1857 Its source files are:
1859 * One or more <<tpp-header,tracepoint provider header>> (`.h`).
1860 * A <<tpp-source,tracepoint provider package source>> (`.c`).
1862 A tracepoint provider package is dynamically linked with `liblttng-ust`,
1863 the LTTng user space tracer, at run time.
1866 .User application linked with `liblttng-ust` and containing a tracepoint provider.
1867 image::ust-app.png[]
1869 NOTE: If you need quick, man:printf(3)-like instrumentation, you can
1870 skip creating and using a tracepoint provider and use
1871 <<tracef,`tracef()`>> or <<tracelog,`tracelog()`>> instead.
1875 ===== Create a tracepoint provider header file template
1877 A _tracepoint provider header file_ contains the tracepoint
1878 definitions of a tracepoint provider.
1880 To create a tracepoint provider header file:
1882 . Start from this template:
1886 .Tracepoint provider header file template (`.h` file extension).
1888 #undef TRACEPOINT_PROVIDER
1889 #define TRACEPOINT_PROVIDER provider_name
1891 #undef TRACEPOINT_INCLUDE
1892 #define TRACEPOINT_INCLUDE "./tp.h"
1894 #if !defined(_TP_H) || defined(TRACEPOINT_HEADER_MULTI_READ)
1897 #include <lttng/tracepoint.h>
1900 * Use TRACEPOINT_EVENT(), TRACEPOINT_EVENT_CLASS(),
1901 * TRACEPOINT_EVENT_INSTANCE(), and TRACEPOINT_LOGLEVEL() here.
1906 #include <lttng/tracepoint-event.h>
1912 * `provider_name` with the name of your tracepoint provider.
1913 * `"tp.h"` with the name of your tracepoint provider header file.
1915 . Below the `#include <lttng/tracepoint.h>` line, put your
1916 <<defining-tracepoints,tracepoint definitions>>.
1918 Your tracepoint provider name must be unique amongst all the possible
1919 tracepoint provider names used on the same target system. We
1920 suggest to include the name of your project or company in the name,
1921 for example, `org_lttng_my_project_tpp`.
1923 TIP: [[lttng-gen-tp]]You can use the man:lttng-gen-tp(1) tool to create
1924 this boilerplate for you. When using cmd:lttng-gen-tp, all you need to
1925 write are the <<defining-tracepoints,tracepoint definitions>>.
1928 [[defining-tracepoints]]
1929 ===== Create a tracepoint definition
1931 A _tracepoint definition_ defines, for a given tracepoint:
1933 * Its **input arguments**. They are the macro parameters that the
1934 `tracepoint()` macro accepts for this particular tracepoint
1935 in the user application's source code.
1936 * Its **output event fields**. They are the sources of event fields
1937 that form the payload of any event that the execution of the
1938 `tracepoint()` macro emits for this particular tracepoint.
1940 You can create a tracepoint definition by using the
1941 `TRACEPOINT_EVENT()` macro below the `#include <lttng/tracepoint.h>`
1943 <<tpp-header,tracepoint provider header file template>>.
1945 The syntax of the `TRACEPOINT_EVENT()` macro is:
1948 .`TRACEPOINT_EVENT()` macro syntax.
1951 /* Tracepoint provider name */
1954 /* Tracepoint name */
1957 /* Input arguments */
1962 /* Output event fields */
1971 * `provider_name` with your tracepoint provider name.
1972 * `tracepoint_name` with your tracepoint name.
1973 * `arguments` with the <<tpp-def-input-args,input arguments>>.
1974 * `fields` with the <<tpp-def-output-fields,output event field>>
1977 This tracepoint emits events named `provider_name:tracepoint_name`.
1980 .Event name's length limitation
1982 The concatenation of the tracepoint provider name and the
1983 tracepoint name must not exceed **254 characters**. If it does, the
1984 instrumented application compiles and runs, but LTTng throws multiple
1985 warnings and you could experience serious issues.
1988 [[tpp-def-input-args]]The syntax of the `TP_ARGS()` macro is:
1991 .`TP_ARGS()` macro syntax.
2000 * `type` with the C type of the argument.
2001 * `arg_name` with the argument name.
2003 You can repeat `type` and `arg_name` up to 10 times to have
2004 more than one argument.
2006 .`TP_ARGS()` usage with three arguments.
2018 The `TP_ARGS()` and `TP_ARGS(void)` forms are valid to create a
2019 tracepoint definition with no input arguments.
2021 [[tpp-def-output-fields]]The `TP_FIELDS()` macro contains a list of
2022 `ctf_*()` macros. Each `ctf_*()` macro defines one event field. See
2023 man:lttng-ust(3) for a complete description of the available `ctf_*()`
2024 macros. A `ctf_*()` macro specifies the type, size, and byte order of
2027 Each `ctf_*()` macro takes an _argument expression_ parameter. This is a
2028 C expression that the tracer evalutes at the `tracepoint()` macro site
2029 in the application's source code. This expression provides a field's
2030 source of data. The argument expression can include input argument names
2031 listed in the `TP_ARGS()` macro.
2033 Each `ctf_*()` macro also takes a _field name_ parameter. Field names
2034 must be unique within a given tracepoint definition.
2036 Here's a complete tracepoint definition example:
2038 .Tracepoint definition.
2040 The following tracepoint definition defines a tracepoint which takes
2041 three input arguments and has four output event fields.
2045 #include "my-custom-structure.h"
2051 const struct my_custom_structure*, my_custom_structure,
2056 ctf_string(query_field, query)
2057 ctf_float(double, ratio_field, ratio)
2058 ctf_integer(int, recv_size, my_custom_structure->recv_size)
2059 ctf_integer(int, send_size, my_custom_structure->send_size)
2064 You can refer to this tracepoint definition with the `tracepoint()`
2065 macro in your application's source code like this:
2069 tracepoint(my_provider, my_tracepoint,
2070 my_structure, some_ratio, the_query);
2074 NOTE: The LTTng tracer only evaluates tracepoint arguments at run time
2075 if they satisfy an enabled <<event,event rule>>.
2078 [[using-tracepoint-classes]]
2079 ===== Use a tracepoint class
2081 A _tracepoint class_ is a class of tracepoints which share the same
2082 output event field definitions. A _tracepoint instance_ is one
2083 instance of such a defined tracepoint class, with its own tracepoint
2086 The <<defining-tracepoints,`TRACEPOINT_EVENT()` macro>> is actually a
2087 shorthand which defines both a tracepoint class and a tracepoint
2088 instance at the same time.
2090 When you build a tracepoint provider package, the C or $$C++$$ compiler
2091 creates one serialization function for each **tracepoint class**. A
2092 serialization function is responsible for serializing the event fields
2093 of a tracepoint to a sub-buffer when tracing.
2095 For various performance reasons, when your situation requires multiple
2096 tracepoint definitions with different names, but with the same event
2097 fields, we recommend that you manually create a tracepoint class
2098 and instantiate as many tracepoint instances as needed. One positive
2099 effect of such a design, amongst other advantages, is that all
2100 tracepoint instances of the same tracepoint class reuse the same
2101 serialization function, thus reducing
2102 https://en.wikipedia.org/wiki/Cache_pollution[cache pollution].
2104 .Use a tracepoint class and tracepoint instances.
2106 Consider the following three tracepoint definitions:
2118 ctf_integer(int, userid, userid)
2119 ctf_integer(size_t, len, len)
2131 ctf_integer(int, userid, userid)
2132 ctf_integer(size_t, len, len)
2144 ctf_integer(int, userid, userid)
2145 ctf_integer(size_t, len, len)
2150 In this case, we create three tracepoint classes, with one implicit
2151 tracepoint instance for each of them: `get_account`, `get_settings`, and
2152 `get_transaction`. However, they all share the same event field names
2153 and types. Hence three identical, yet independent serialization
2154 functions are created when you build the tracepoint provider package.
2156 A better design choice is to define a single tracepoint class and three
2157 tracepoint instances:
2161 /* The tracepoint class */
2162 TRACEPOINT_EVENT_CLASS(
2163 /* Tracepoint provider name */
2166 /* Tracepoint class name */
2169 /* Input arguments */
2175 /* Output event fields */
2177 ctf_integer(int, userid, userid)
2178 ctf_integer(size_t, len, len)
2182 /* The tracepoint instances */
2183 TRACEPOINT_EVENT_INSTANCE(
2184 /* Tracepoint provider name */
2187 /* Tracepoint class name */
2190 /* Tracepoint name */
2193 /* Input arguments */
2199 TRACEPOINT_EVENT_INSTANCE(
2208 TRACEPOINT_EVENT_INSTANCE(
2221 [[assigning-log-levels]]
2222 ===== Assign a log level to a tracepoint definition
2224 You can assign an optional _log level_ to a
2225 <<defining-tracepoints,tracepoint definition>>.
2227 Assigning different levels of severity to tracepoint definitions can
2228 be useful: when you <<enabling-disabling-events,create an event rule>>,
2229 you can target tracepoints having a log level as severe as a specific
2232 The concept of LTTng-UST log levels is similar to the levels found
2233 in typical logging frameworks:
2235 * In a logging framework, the log level is given by the function
2236 or method name you use at the log statement site: `debug()`,
2237 `info()`, `warn()`, `error()`, and so on.
2238 * In LTTng-UST, you statically assign the log level to a tracepoint
2239 definition; any `tracepoint()` macro invocation which refers to
2240 this definition has this log level.
2242 You can assign a log level to a tracepoint definition with the
2243 `TRACEPOINT_LOGLEVEL()` macro. You must use this macro _after_ the
2244 <<defining-tracepoints,`TRACEPOINT_EVENT()`>> or
2245 <<using-tracepoint-classes,`TRACEPOINT_INSTANCE()`>> macro for a given
2248 The syntax of the `TRACEPOINT_LOGLEVEL()` macro is:
2251 .`TRACEPOINT_LOGLEVEL()` macro syntax.
2253 TRACEPOINT_LOGLEVEL(provider_name, tracepoint_name, log_level)
2258 * `provider_name` with the tracepoint provider name.
2259 * `tracepoint_name` with the tracepoint name.
2260 * `log_level` with the log level to assign to the tracepoint
2261 definition named `tracepoint_name` in the `provider_name`
2262 tracepoint provider.
2264 See man:lttng-ust(3) for a list of available log level names.
2266 .Assign the `TRACE_DEBUG_UNIT` log level to a tracepoint definition.
2270 /* Tracepoint definition */
2279 ctf_integer(int, userid, userid)
2280 ctf_integer(size_t, len, len)
2284 /* Log level assignment */
2285 TRACEPOINT_LOGLEVEL(my_app, get_transaction, TRACE_DEBUG_UNIT)
2291 ===== Create a tracepoint provider package source file
2293 A _tracepoint provider package source file_ is a C source file which
2294 includes a <<tpp-header,tracepoint provider header file>> to expand its
2295 macros into event serialization and other functions.
2297 You can always use the following tracepoint provider package source
2301 .Tracepoint provider package source file template.
2303 #define TRACEPOINT_CREATE_PROBES
2308 Replace `tp.h` with the name of your <<tpp-header,tracepoint provider
2309 header file>> name. You may also include more than one tracepoint
2310 provider header file here to create a tracepoint provider package
2311 holding more than one tracepoint providers.
2314 [[probing-the-application-source-code]]
2315 ==== Add tracepoints to an application's source code
2317 Once you <<tpp-header,create a tracepoint provider header file>>, you
2318 can use the `tracepoint()` macro in your application's
2319 source code to insert the tracepoints that this header
2320 <<defining-tracepoints,defines>>.
2322 The `tracepoint()` macro takes at least two parameters: the tracepoint
2323 provider name and the tracepoint name. The corresponding tracepoint
2324 definition defines the other parameters.
2326 .`tracepoint()` usage.
2328 The following <<defining-tracepoints,tracepoint definition>> defines a
2329 tracepoint which takes two input arguments and has two output event
2333 .Tracepoint provider header file.
2335 #include "my-custom-structure.h"
2342 const char*, cmd_name
2345 ctf_string(cmd_name, cmd_name)
2346 ctf_integer(int, number_of_args, argc)
2351 You can refer to this tracepoint definition with the `tracepoint()`
2352 macro in your application's source code like this:
2355 .Application's source file.
2359 int main(int argc, char* argv[])
2361 tracepoint(my_provider, my_tracepoint, argc, argv[0]);
2367 Note how the application's source code includes
2368 the tracepoint provider header file containing the tracepoint
2369 definitions to use, path:{tp.h}.
2372 .`tracepoint()` usage with a complex tracepoint definition.
2374 Consider this complex tracepoint definition, where multiple event
2375 fields refer to the same input arguments in their argument expression
2379 .Tracepoint provider header file.
2381 /* For `struct stat` */
2382 #include <sys/types.h>
2383 #include <sys/stat.h>
2395 ctf_integer(int, my_constant_field, 23 + 17)
2396 ctf_integer(int, my_int_arg_field, my_int_arg)
2397 ctf_integer(int, my_int_arg_field2, my_int_arg * my_int_arg)
2398 ctf_integer(int, sum4_field, my_str_arg[0] + my_str_arg[1] +
2399 my_str_arg[2] + my_str_arg[3])
2400 ctf_string(my_str_arg_field, my_str_arg)
2401 ctf_integer_hex(off_t, size_field, st->st_size)
2402 ctf_float(double, size_dbl_field, (double) st->st_size)
2403 ctf_sequence_text(char, half_my_str_arg_field, my_str_arg,
2404 size_t, strlen(my_str_arg) / 2)
2409 You can refer to this tracepoint definition with the `tracepoint()`
2410 macro in your application's source code like this:
2413 .Application's source file.
2415 #define TRACEPOINT_DEFINE
2422 stat("/etc/fstab", &s);
2423 tracepoint(my_provider, my_tracepoint, 23, "Hello, World!", &s);
2429 If you look at the event record that LTTng writes when tracing this
2430 program, assuming the file size of path:{/etc/fstab} is 301{nbsp}bytes,
2431 it should look like this:
2433 .Event record fields
2435 |Field's name |Field's value
2436 |`my_constant_field` |40
2437 |`my_int_arg_field` |23
2438 |`my_int_arg_field2` |529
2440 |`my_str_arg_field` |`Hello, World!`
2441 |`size_field` |0x12d
2442 |`size_dbl_field` |301.0
2443 |`half_my_str_arg_field` |`Hello,`
2447 Sometimes, the arguments you pass to `tracepoint()` are expensive to
2448 compute--they use the call stack, for example. To avoid this
2449 computation when the tracepoint is disabled, you can use the
2450 `tracepoint_enabled()` and `do_tracepoint()` macros.
2452 The syntax of the `tracepoint_enabled()` and `do_tracepoint()` macros
2456 .`tracepoint_enabled()` and `do_tracepoint()` macros syntax.
2458 tracepoint_enabled(provider_name, tracepoint_name)
2459 do_tracepoint(provider_name, tracepoint_name, ...)
2464 * `provider_name` with the tracepoint provider name.
2465 * `tracepoint_name` with the tracepoint name.
2467 `tracepoint_enabled()` returns a non-zero value if the tracepoint named
2468 `tracepoint_name` from the provider named `provider_name` is enabled
2471 `do_tracepoint()` is like `tracepoint()`, except that it doesn't check
2472 if the tracepoint is enabled. Using `tracepoint()` with
2473 `tracepoint_enabled()` is dangerous since `tracepoint()` also contains
2474 the `tracepoint_enabled()` check, thus a race condition is
2475 possible in this situation:
2478 .Possible race condition when using `tracepoint_enabled()` with `tracepoint()`.
2480 if (tracepoint_enabled(my_provider, my_tracepoint)) {
2481 stuff = prepare_stuff();
2484 tracepoint(my_provider, my_tracepoint, stuff);
2487 If the tracepoint is enabled after the condition, then `stuff` is not
2488 prepared: the emitted event will either contain wrong data, or the whole
2489 application could crash (segmentation fault, for example).
2491 NOTE: Neither `tracepoint_enabled()` nor `do_tracepoint()` have an
2492 `STAP_PROBEV()` call. If you need it, you must emit
2496 [[building-tracepoint-providers-and-user-application]]
2497 ==== Build and link a tracepoint provider package and an application
2499 Once you have one or more <<tpp-header,tracepoint provider header
2500 files>> and a <<tpp-source,tracepoint provider package source file>>,
2501 you can create the tracepoint provider package by compiling its source
2502 file. From here, multiple build and run scenarios are possible. The
2503 following table shows common application and library configurations
2504 along with the required command lines to achieve them.
2506 In the following diagrams, we use the following file names:
2509 Executable application.
2512 Application's object file.
2515 Tracepoint provider package object file.
2518 Tracepoint provider package archive file.
2521 Tracepoint provider package shared object file.
2524 User library object file.
2527 User library shared object file.
2529 We use the following symbols in the diagrams of table below:
2532 .Symbols used in the build scenario diagrams.
2533 image::ust-sit-symbols.png[]
2535 We assume that path:{.} is part of the env:LD_LIBRARY_PATH environment
2536 variable in the following instructions.
2538 [role="growable ust-scenarios",cols="asciidoc,asciidoc"]
2539 .Common tracepoint provider package scenarios.
2541 |Scenario |Instructions
2544 The instrumented application is statically linked with
2545 the tracepoint provider package object.
2547 image::ust-sit+app-linked-with-tp-o+app-instrumented.png[]
2550 include::../common/ust-sit-step-tp-o.txt[]
2552 To build the instrumented application:
2554 . In path:{app.c}, before including path:{tpp.h}, add the following line:
2559 #define TRACEPOINT_DEFINE
2563 . Compile the application source file:
2572 . Build the application:
2577 $ gcc -o app app.o tpp.o -llttng-ust -ldl
2581 To run the instrumented application:
2583 * Start the application:
2593 The instrumented application is statically linked with the
2594 tracepoint provider package archive file.
2596 image::ust-sit+app-linked-with-tp-a+app-instrumented.png[]
2599 To create the tracepoint provider package archive file:
2601 . Compile the <<tpp-source,tracepoint provider package source file>>:
2610 . Create the tracepoint provider package archive file:
2615 $ ar rcs tpp.a tpp.o
2619 To build the instrumented application:
2621 . In path:{app.c}, before including path:{tpp.h}, add the following line:
2626 #define TRACEPOINT_DEFINE
2630 . Compile the application source file:
2639 . Build the application:
2644 $ gcc -o app app.o tpp.a -llttng-ust -ldl
2648 To run the instrumented application:
2650 * Start the application:
2660 The instrumented application is linked with the tracepoint provider
2661 package shared object.
2663 image::ust-sit+app-linked-with-tp-so+app-instrumented.png[]
2666 include::../common/ust-sit-step-tp-so.txt[]
2668 To build the instrumented application:
2670 . In path:{app.c}, before including path:{tpp.h}, add the following line:
2675 #define TRACEPOINT_DEFINE
2679 . Compile the application source file:
2688 . Build the application:
2693 $ gcc -o app app.o -ldl -L. -ltpp
2697 To run the instrumented application:
2699 * Start the application:
2709 The tracepoint provider package shared object is preloaded before the
2710 instrumented application starts.
2712 image::ust-sit+tp-so-preloaded+app-instrumented.png[]
2715 include::../common/ust-sit-step-tp-so.txt[]
2717 To build the instrumented application:
2719 . In path:{app.c}, before including path:{tpp.h}, add the
2725 #define TRACEPOINT_DEFINE
2726 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
2730 . Compile the application source file:
2739 . Build the application:
2744 $ gcc -o app app.o -ldl
2748 To run the instrumented application with tracing support:
2750 * Preload the tracepoint provider package shared object and
2751 start the application:
2756 $ LD_PRELOAD=./libtpp.so ./app
2760 To run the instrumented application without tracing support:
2762 * Start the application:
2772 The instrumented application dynamically loads the tracepoint provider
2773 package shared object.
2775 See the <<dlclose-warning,warning about `dlclose()`>>.
2777 image::ust-sit+app-dlopens-tp-so+app-instrumented.png[]
2780 include::../common/ust-sit-step-tp-so.txt[]
2782 To build the instrumented application:
2784 . In path:{app.c}, before including path:{tpp.h}, add the
2790 #define TRACEPOINT_DEFINE
2791 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
2795 . Compile the application source file:
2804 . Build the application:
2809 $ gcc -o app app.o -ldl
2813 To run the instrumented application:
2815 * Start the application:
2825 The application is linked with the instrumented user library.
2827 The instrumented user library is statically linked with the tracepoint
2828 provider package object file.
2830 image::ust-sit+app-linked-with-lib+lib-linked-with-tp-o+lib-instrumented.png[]
2833 include::../common/ust-sit-step-tp-o-fpic.txt[]
2835 To build the instrumented user library:
2837 . In path:{emon.c}, before including path:{tpp.h}, add the
2843 #define TRACEPOINT_DEFINE
2847 . Compile the user library source file:
2852 $ gcc -I. -fpic -c emon.c
2856 . Build the user library shared object:
2861 $ gcc -shared -o libemon.so emon.o tpp.o -llttng-ust -ldl
2865 To build the application:
2867 . Compile the application source file:
2876 . Build the application:
2881 $ gcc -o app app.o -L. -lemon
2885 To run the application:
2887 * Start the application:
2897 The application is linked with the instrumented user library.
2899 The instrumented user library is linked with the tracepoint provider
2900 package shared object.
2902 image::ust-sit+app-linked-with-lib+lib-linked-with-tp-so+lib-instrumented.png[]
2905 include::../common/ust-sit-step-tp-so.txt[]
2907 To build the instrumented user library:
2909 . In path:{emon.c}, before including path:{tpp.h}, add the
2915 #define TRACEPOINT_DEFINE
2919 . Compile the user library source file:
2924 $ gcc -I. -fpic -c emon.c
2928 . Build the user library shared object:
2933 $ gcc -shared -o libemon.so emon.o -ldl -L. -ltpp
2937 To build the application:
2939 . Compile the application source file:
2948 . Build the application:
2953 $ gcc -o app app.o -L. -lemon
2957 To run the application:
2959 * Start the application:
2969 The tracepoint provider package shared object is preloaded before the
2972 The application is linked with the instrumented user library.
2974 image::ust-sit+tp-so-preloaded+app-linked-with-lib+lib-instrumented.png[]
2977 include::../common/ust-sit-step-tp-so.txt[]
2979 To build the instrumented user library:
2981 . In path:{emon.c}, before including path:{tpp.h}, add the
2987 #define TRACEPOINT_DEFINE
2988 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
2992 . Compile the user library source file:
2997 $ gcc -I. -fpic -c emon.c
3001 . Build the user library shared object:
3006 $ gcc -shared -o libemon.so emon.o -ldl
3010 To build the application:
3012 . Compile the application source file:
3021 . Build the application:
3026 $ gcc -o app app.o -L. -lemon
3030 To run the application with tracing support:
3032 * Preload the tracepoint provider package shared object and
3033 start the application:
3038 $ LD_PRELOAD=./libtpp.so ./app
3042 To run the application without tracing support:
3044 * Start the application:
3054 The application is linked with the instrumented user library.
3056 The instrumented user library dynamically loads the tracepoint provider
3057 package shared object.
3059 See the <<dlclose-warning,warning about `dlclose()`>>.
3061 image::ust-sit+app-linked-with-lib+lib-dlopens-tp-so+lib-instrumented.png[]
3064 include::../common/ust-sit-step-tp-so.txt[]
3066 To build the instrumented user library:
3068 . In path:{emon.c}, before including path:{tpp.h}, add the
3074 #define TRACEPOINT_DEFINE
3075 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
3079 . Compile the user library source file:
3084 $ gcc -I. -fpic -c emon.c
3088 . Build the user library shared object:
3093 $ gcc -shared -o libemon.so emon.o -ldl
3097 To build the application:
3099 . Compile the application source file:
3108 . Build the application:
3113 $ gcc -o app app.o -L. -lemon
3117 To run the application:
3119 * Start the application:
3129 The application dynamically loads the instrumented user library.
3131 The instrumented user library is linked with the tracepoint provider
3132 package shared object.
3134 See the <<dlclose-warning,warning about `dlclose()`>>.
3136 image::ust-sit+app-dlopens-lib+lib-linked-with-tp-so+lib-instrumented.png[]
3139 include::../common/ust-sit-step-tp-so.txt[]
3141 To build the instrumented user library:
3143 . In path:{emon.c}, before including path:{tpp.h}, add the
3149 #define TRACEPOINT_DEFINE
3153 . Compile the user library source file:
3158 $ gcc -I. -fpic -c emon.c
3162 . Build the user library shared object:
3167 $ gcc -shared -o libemon.so emon.o -ldl -L. -ltpp
3171 To build the application:
3173 . Compile the application source file:
3182 . Build the application:
3187 $ gcc -o app app.o -ldl -L. -lemon
3191 To run the application:
3193 * Start the application:
3203 The application dynamically loads the instrumented user library.
3205 The instrumented user library dynamically loads the tracepoint provider
3206 package shared object.
3208 See the <<dlclose-warning,warning about `dlclose()`>>.
3210 image::ust-sit+app-dlopens-lib+lib-dlopens-tp-so+lib-instrumented.png[]
3213 include::../common/ust-sit-step-tp-so.txt[]
3215 To build the instrumented user library:
3217 . In path:{emon.c}, before including path:{tpp.h}, add the
3223 #define TRACEPOINT_DEFINE
3224 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
3228 . Compile the user library source file:
3233 $ gcc -I. -fpic -c emon.c
3237 . Build the user library shared object:
3242 $ gcc -shared -o libemon.so emon.o -ldl
3246 To build the application:
3248 . Compile the application source file:
3257 . Build the application:
3262 $ gcc -o app app.o -ldl -L. -lemon
3266 To run the application:
3268 * Start the application:
3278 The tracepoint provider package shared object is preloaded before the
3281 The application dynamically loads the instrumented user library.
3283 image::ust-sit+tp-so-preloaded+app-dlopens-lib+lib-instrumented.png[]
3286 include::../common/ust-sit-step-tp-so.txt[]
3288 To build the instrumented user library:
3290 . In path:{emon.c}, before including path:{tpp.h}, add the
3296 #define TRACEPOINT_DEFINE
3297 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
3301 . Compile the user library source file:
3306 $ gcc -I. -fpic -c emon.c
3310 . Build the user library shared object:
3315 $ gcc -shared -o libemon.so emon.o -ldl
3319 To build the application:
3321 . Compile the application source file:
3330 . Build the application:
3335 $ gcc -o app app.o -L. -lemon
3339 To run the application with tracing support:
3341 * Preload the tracepoint provider package shared object and
3342 start the application:
3347 $ LD_PRELOAD=./libtpp.so ./app
3351 To run the application without tracing support:
3353 * Start the application:
3363 The application is statically linked with the tracepoint provider
3364 package object file.
3366 The application is linked with the instrumented user library.
3368 image::ust-sit+app-linked-with-tp-o+app-linked-with-lib+lib-instrumented.png[]
3371 include::../common/ust-sit-step-tp-o.txt[]
3373 To build the instrumented user library:
3375 . In path:{emon.c}, before including path:{tpp.h}, add the
3381 #define TRACEPOINT_DEFINE
3385 . Compile the user library source file:
3390 $ gcc -I. -fpic -c emon.c
3394 . Build the user library shared object:
3399 $ gcc -shared -o libemon.so emon.o
3403 To build the application:
3405 . Compile the application source file:
3414 . Build the application:
3419 $ gcc -o app app.o tpp.o -llttng-ust -ldl -L. -lemon
3423 To run the instrumented application:
3425 * Start the application:
3435 The application is statically linked with the tracepoint provider
3436 package object file.
3438 The application dynamically loads the instrumented user library.
3440 image::ust-sit+app-linked-with-tp-o+app-dlopens-lib+lib-instrumented.png[]
3443 include::../common/ust-sit-step-tp-o.txt[]
3445 To build the application:
3447 . In path:{app.c}, before including path:{tpp.h}, add the following line:
3452 #define TRACEPOINT_DEFINE
3456 . Compile the application source file:
3465 . Build the application:
3470 $ gcc -Wl,--export-dynamic -o app app.o tpp.o \
3475 The `--export-dynamic` option passed to the linker is necessary for the
3476 dynamically loaded library to ``see'' the tracepoint symbols defined in
3479 To build the instrumented user library:
3481 . Compile the user library source file:
3486 $ gcc -I. -fpic -c emon.c
3490 . Build the user library shared object:
3495 $ gcc -shared -o libemon.so emon.o
3499 To run the application:
3501 * Start the application:
3513 .Do not use man:dlclose(3) on a tracepoint provider package
3515 Never use man:dlclose(3) on any shared object which:
3517 * Is linked with, statically or dynamically, a tracepoint provider
3519 * Calls man:dlopen(3) itself to dynamically open a tracepoint provider
3520 package shared object.
3522 This is currently considered **unsafe** due to a lack of reference
3523 counting from LTTng-UST to the shared object.
3525 A known workaround (available since glibc 2.2) is to use the
3526 `RTLD_NODELETE` flag when calling man:dlopen(3) initially. This has the
3527 effect of not unloading the loaded shared object, even if man:dlclose(3)
3530 You can also preload the tracepoint provider package shared object with
3531 the env:LD_PRELOAD environment variable to overcome this limitation.
3535 [[using-lttng-ust-with-daemons]]
3536 ===== Use noch:{LTTng-UST} with daemons
3538 If your instrumented application calls man:fork(2), man:clone(2),
3539 or BSD's man:rfork(2), without a following man:exec(3)-family
3540 system call, you must preload the path:{liblttng-ust-fork.so} shared
3541 object when you start the application.
3545 $ LD_PRELOAD=liblttng-ust-fork.so ./my-app
3548 If your tracepoint provider package is
3549 a shared library which you also preload, you must put both
3550 shared objects in env:LD_PRELOAD:
3554 $ LD_PRELOAD=liblttng-ust-fork.so:/path/to/tp.so ./my-app
3560 ===== Use noch:{LTTng-UST} with applications which close file descriptors that don't belong to them
3562 If your instrumented application closes one or more file descriptors
3563 which it did not open itself, you must preload the
3564 path:{liblttng-ust-fd.so} shared object when you start the application:
3568 $ LD_PRELOAD=liblttng-ust-fd.so ./my-app
3571 Typical use cases include closing all the file descriptors after
3572 man:fork(2) or man:rfork(2) and buggy applications doing
3576 [[lttng-ust-pkg-config]]
3577 ===== Use noch:{pkg-config}
3579 On some distributions, LTTng-UST ships with a
3580 https://www.freedesktop.org/wiki/Software/pkg-config/[pkg-config]
3581 metadata file. If this is your case, then you can use cmd:pkg-config to
3582 build an application on the command line:
3586 $ gcc -o my-app my-app.o tp.o $(pkg-config --cflags --libs lttng-ust)
3590 [[instrumenting-32-bit-app-on-64-bit-system]]
3591 ===== [[advanced-instrumenting-techniques]]Build a 32-bit instrumented application for a 64-bit target system
3593 In order to trace a 32-bit application running on a 64-bit system,
3594 LTTng must use a dedicated 32-bit
3595 <<lttng-consumerd,consumer daemon>>.
3597 The following steps show how to build and install a 32-bit consumer
3598 daemon, which is _not_ part of the default 64-bit LTTng build, how to
3599 build and install the 32-bit LTTng-UST libraries, and how to build and
3600 link an instrumented 32-bit application in that context.
3602 To build a 32-bit instrumented application for a 64-bit target system,
3603 assuming you have a fresh target system with no installed Userspace RCU
3606 . Download, build, and install a 32-bit version of Userspace RCU:
3611 $ cd $(mktemp -d) &&
3612 wget http://lttng.org/files/urcu/userspace-rcu-latest-0.9.tar.bz2 &&
3613 tar -xf userspace-rcu-latest-0.9.tar.bz2 &&
3614 cd userspace-rcu-0.9.* &&
3615 ./configure --libdir=/usr/local/lib32 CFLAGS=-m32 &&
3617 sudo make install &&
3622 . Using your distribution's package manager, or from source, install
3623 the following 32-bit versions of the following dependencies of
3624 LTTng-tools and LTTng-UST:
3627 * https://sourceforge.net/projects/libuuid/[libuuid]
3628 * http://directory.fsf.org/wiki/Popt[popt]
3629 * http://www.xmlsoft.org/[libxml2]
3632 . Download, build, and install a 32-bit version of the latest
3633 LTTng-UST{nbsp}{revision}:
3638 $ cd $(mktemp -d) &&
3639 wget http://lttng.org/files/lttng-ust/lttng-ust-latest-2.10.tar.bz2 &&
3640 tar -xf lttng-ust-latest-2.10.tar.bz2 &&
3641 cd lttng-ust-2.10.* &&
3642 ./configure --libdir=/usr/local/lib32 \
3643 CFLAGS=-m32 CXXFLAGS=-m32 \
3644 LDFLAGS='-L/usr/local/lib32 -L/usr/lib32' &&
3646 sudo make install &&
3653 Depending on your distribution,
3654 32-bit libraries could be installed at a different location than
3655 `/usr/lib32`. For example, Debian is known to install
3656 some 32-bit libraries in `/usr/lib/i386-linux-gnu`.
3658 In this case, make sure to set `LDFLAGS` to all the
3659 relevant 32-bit library paths, for example:
3663 $ LDFLAGS='-L/usr/lib/i386-linux-gnu -L/usr/lib32'
3667 . Download the latest LTTng-tools{nbsp}{revision}, build, and install
3668 the 32-bit consumer daemon:
3673 $ cd $(mktemp -d) &&
3674 wget http://lttng.org/files/lttng-tools/lttng-tools-latest-2.10.tar.bz2 &&
3675 tar -xf lttng-tools-latest-2.10.tar.bz2 &&
3676 cd lttng-tools-2.10.* &&
3677 ./configure --libdir=/usr/local/lib32 CFLAGS=-m32 CXXFLAGS=-m32 \
3678 LDFLAGS='-L/usr/local/lib32 -L/usr/lib32' \
3679 --disable-bin-lttng --disable-bin-lttng-crash \
3680 --disable-bin-lttng-relayd --disable-bin-lttng-sessiond &&
3682 cd src/bin/lttng-consumerd &&
3683 sudo make install &&
3688 . From your distribution or from source,
3689 <<installing-lttng,install>> the 64-bit versions of
3690 LTTng-UST and Userspace RCU.
3691 . Download, build, and install the 64-bit version of the
3692 latest LTTng-tools{nbsp}{revision}:
3697 $ cd $(mktemp -d) &&
3698 wget http://lttng.org/files/lttng-tools/lttng-tools-latest-2.10.tar.bz2 &&
3699 tar -xf lttng-tools-latest-2.10.tar.bz2 &&
3700 cd lttng-tools-2.10.* &&
3701 ./configure --with-consumerd32-libdir=/usr/local/lib32 \
3702 --with-consumerd32-bin=/usr/local/lib32/lttng/libexec/lttng-consumerd &&
3704 sudo make install &&
3709 . Pass the following options to man:gcc(1), man:g++(1), or man:clang(1)
3710 when linking your 32-bit application:
3713 -m32 -L/usr/lib32 -L/usr/local/lib32 \
3714 -Wl,-rpath,/usr/lib32,-rpath,/usr/local/lib32
3717 For example, let's rebuild the quick start example in
3718 <<tracing-your-own-user-application,Trace a user application>> as an
3719 instrumented 32-bit application:
3724 $ gcc -m32 -c -I. hello-tp.c
3725 $ gcc -m32 -c hello.c
3726 $ gcc -m32 -o hello hello.o hello-tp.o \
3727 -L/usr/lib32 -L/usr/local/lib32 \
3728 -Wl,-rpath,/usr/lib32,-rpath,/usr/local/lib32 \
3733 No special action is required to execute the 32-bit application and
3734 to trace it: use the command-line man:lttng(1) tool as usual.
3741 man:tracef(3) is a small LTTng-UST API designed for quick,
3742 man:printf(3)-like instrumentation without the burden of
3743 <<tracepoint-provider,creating>> and
3744 <<building-tracepoint-providers-and-user-application,building>>
3745 a tracepoint provider package.
3747 To use `tracef()` in your application:
3749 . In the C or C++ source files where you need to use `tracef()`,
3750 include `<lttng/tracef.h>`:
3755 #include <lttng/tracef.h>
3759 . In the application's source code, use `tracef()` like you would use
3767 tracef("my message: %d (%s)", my_integer, my_string);
3773 . Link your application with `liblttng-ust`:
3778 $ gcc -o app app.c -llttng-ust
3782 To trace the events that `tracef()` calls emit:
3784 * <<enabling-disabling-events,Create an event rule>> which matches the
3785 `lttng_ust_tracef:*` event name:
3790 $ lttng enable-event --userspace 'lttng_ust_tracef:*'
3795 .Limitations of `tracef()`
3797 The `tracef()` utility function was developed to make user space tracing
3798 super simple, albeit with notable disadvantages compared to
3799 <<defining-tracepoints,user-defined tracepoints>>:
3801 * All the emitted events have the same tracepoint provider and
3802 tracepoint names, respectively `lttng_ust_tracef` and `event`.
3803 * There is no static type checking.
3804 * The only event record field you actually get, named `msg`, is a string
3805 potentially containing the values you passed to `tracef()`
3806 using your own format string. This also means that you cannot filter
3807 events with a custom expression at run time because there are no
3809 * Since `tracef()` uses the C standard library's man:vasprintf(3)
3810 function behind the scenes to format the strings at run time, its
3811 expected performance is lower than with user-defined tracepoints,
3812 which do not require a conversion to a string.
3814 Taking this into consideration, `tracef()` is useful for some quick
3815 prototyping and debugging, but you should not consider it for any
3816 permanent and serious applicative instrumentation.
3822 ==== Use `tracelog()`
3824 The man:tracelog(3) API is very similar to <<tracef,`tracef()`>>, with
3825 the difference that it accepts an additional log level parameter.
3827 The goal of `tracelog()` is to ease the migration from logging to
3830 To use `tracelog()` in your application:
3832 . In the C or C++ source files where you need to use `tracelog()`,
3833 include `<lttng/tracelog.h>`:
3838 #include <lttng/tracelog.h>
3842 . In the application's source code, use `tracelog()` like you would use
3843 man:printf(3), except for the first parameter which is the log
3851 tracelog(TRACE_WARNING, "my message: %d (%s)",
3852 my_integer, my_string);
3858 See man:lttng-ust(3) for a list of available log level names.
3860 . Link your application with `liblttng-ust`:
3865 $ gcc -o app app.c -llttng-ust
3869 To trace the events that `tracelog()` calls emit with a log level
3870 _as severe as_ a specific log level:
3872 * <<enabling-disabling-events,Create an event rule>> which matches the
3873 `lttng_ust_tracelog:*` event name and a minimum level
3879 $ lttng enable-event --userspace 'lttng_ust_tracelog:*'
3880 --loglevel=TRACE_WARNING
3884 To trace the events that `tracelog()` calls emit with a
3885 _specific log level_:
3887 * Create an event rule which matches the `lttng_ust_tracelog:*`
3888 event name and a specific log level:
3893 $ lttng enable-event --userspace 'lttng_ust_tracelog:*'
3894 --loglevel-only=TRACE_INFO
3899 [[prebuilt-ust-helpers]]
3900 === Prebuilt user space tracing helpers
3902 The LTTng-UST package provides a few helpers in the form of preloadable
3903 shared objects which automatically instrument system functions and
3906 The helper shared objects are normally found in dir:{/usr/lib}. If you
3907 built LTTng-UST <<building-from-source,from source>>, they are probably
3908 located in dir:{/usr/local/lib}.
3910 The installed user space tracing helpers in LTTng-UST{nbsp}{revision}
3913 path:{liblttng-ust-libc-wrapper.so}::
3914 path:{liblttng-ust-pthread-wrapper.so}::
3915 <<liblttng-ust-libc-pthread-wrapper,C{nbsp}standard library
3916 memory and POSIX threads function tracing>>.
3918 path:{liblttng-ust-cyg-profile.so}::
3919 path:{liblttng-ust-cyg-profile-fast.so}::
3920 <<liblttng-ust-cyg-profile,Function entry and exit tracing>>.
3922 path:{liblttng-ust-dl.so}::
3923 <<liblttng-ust-dl,Dynamic linker tracing>>.
3925 To use a user space tracing helper with any user application:
3927 * Preload the helper shared object when you start the application:
3932 $ LD_PRELOAD=liblttng-ust-libc-wrapper.so my-app
3936 You can preload more than one helper:
3941 $ LD_PRELOAD=liblttng-ust-libc-wrapper.so:liblttng-ust-dl.so my-app
3947 [[liblttng-ust-libc-pthread-wrapper]]
3948 ==== Instrument C standard library memory and POSIX threads functions
3950 The path:{liblttng-ust-libc-wrapper.so} and
3951 path:{liblttng-ust-pthread-wrapper.so} helpers
3952 add instrumentation to some C standard library and POSIX
3956 .Functions instrumented by preloading path:{liblttng-ust-libc-wrapper.so}.
3958 |TP provider name |TP name |Instrumented function
3960 .6+|`lttng_ust_libc` |`malloc` |man:malloc(3)
3961 |`calloc` |man:calloc(3)
3962 |`realloc` |man:realloc(3)
3963 |`free` |man:free(3)
3964 |`memalign` |man:memalign(3)
3965 |`posix_memalign` |man:posix_memalign(3)
3969 .Functions instrumented by preloading path:{liblttng-ust-pthread-wrapper.so}.
3971 |TP provider name |TP name |Instrumented function
3973 .4+|`lttng_ust_pthread` |`pthread_mutex_lock_req` |man:pthread_mutex_lock(3p) (request time)
3974 |`pthread_mutex_lock_acq` |man:pthread_mutex_lock(3p) (acquire time)
3975 |`pthread_mutex_trylock` |man:pthread_mutex_trylock(3p)
3976 |`pthread_mutex_unlock` |man:pthread_mutex_unlock(3p)
3979 When you preload the shared object, it replaces the functions listed
3980 in the previous tables by wrappers which contain tracepoints and call
3981 the replaced functions.
3984 [[liblttng-ust-cyg-profile]]
3985 ==== Instrument function entry and exit
3987 The path:{liblttng-ust-cyg-profile*.so} helpers can add instrumentation
3988 to the entry and exit points of functions.
3990 man:gcc(1) and man:clang(1) have an option named
3991 https://gcc.gnu.org/onlinedocs/gcc/Instrumentation-Options.html[`-finstrument-functions`]
3992 which generates instrumentation calls for entry and exit to functions.
3993 The LTTng-UST function tracing helpers,
3994 path:{liblttng-ust-cyg-profile.so} and
3995 path:{liblttng-ust-cyg-profile-fast.so}, take advantage of this feature
3996 to add tracepoints to the two generated functions (which contain
3997 `cyg_profile` in their names, hence the helper's name).
3999 To use the LTTng-UST function tracing helper, the source files to
4000 instrument must be built using the `-finstrument-functions` compiler
4003 There are two versions of the LTTng-UST function tracing helper:
4005 * **path:{liblttng-ust-cyg-profile-fast.so}** is a lightweight variant
4006 that you should only use when it can be _guaranteed_ that the
4007 complete event stream is recorded without any lost event record.
4008 Any kind of duplicate information is left out.
4010 Assuming no event record is lost, having only the function addresses on
4011 entry is enough to create a call graph, since an event record always
4012 contains the ID of the CPU that generated it.
4014 You can use a tool like man:addr2line(1) to convert function addresses
4015 back to source file names and line numbers.
4017 * **path:{liblttng-ust-cyg-profile.so}** is a more robust variant
4018 which also works in use cases where event records might get discarded or
4019 not recorded from application startup.
4020 In these cases, the trace analyzer needs more information to be
4021 able to reconstruct the program flow.
4023 See man:lttng-ust-cyg-profile(3) to learn more about the instrumentation
4024 points of this helper.
4026 All the tracepoints that this helper provides have the
4027 log level `TRACE_DEBUG_FUNCTION` (see man:lttng-ust(3)).
4029 TIP: It's sometimes a good idea to limit the number of source files that
4030 you compile with the `-finstrument-functions` option to prevent LTTng
4031 from writing an excessive amount of trace data at run time. When using
4032 man:gcc(1), you can use the
4033 `-finstrument-functions-exclude-function-list` option to avoid
4034 instrument entries and exits of specific function names.
4039 ==== Instrument the dynamic linker
4041 The path:{liblttng-ust-dl.so} helper adds instrumentation to the
4042 man:dlopen(3) and man:dlclose(3) function calls.
4044 See man:lttng-ust-dl(3) to learn more about the instrumentation points
4049 [[java-application]]
4050 === User space Java agent
4052 You can instrument any Java application which uses one of the following
4055 * The https://docs.oracle.com/javase/7/docs/api/java/util/logging/package-summary.html[**`java.util.logging`**]
4056 (JUL) core logging facilities.
4057 * http://logging.apache.org/log4j/1.2/[**Apache log4j 1.2**], since
4058 LTTng 2.6. Note that Apache Log4j{nbsp}2 is not supported.
4061 .LTTng-UST Java agent imported by a Java application.
4062 image::java-app.png[]
4064 Note that the methods described below are new in LTTng{nbsp}2.8.
4065 Previous LTTng versions use another technique.
4067 NOTE: We use http://openjdk.java.net/[OpenJDK]{nbsp}8 for development
4068 and https://ci.lttng.org/[continuous integration], thus this version is
4069 directly supported. However, the LTTng-UST Java agent is also tested
4070 with OpenJDK{nbsp}7.
4075 ==== Use the LTTng-UST Java agent for `java.util.logging`
4077 To use the LTTng-UST Java agent in a Java application which uses
4078 `java.util.logging` (JUL):
4080 . In the Java application's source code, import the LTTng-UST
4081 log handler package for `java.util.logging`:
4086 import org.lttng.ust.agent.jul.LttngLogHandler;
4090 . Create an LTTng-UST JUL log handler:
4095 Handler lttngUstLogHandler = new LttngLogHandler();
4099 . Add this handler to the JUL loggers which should emit LTTng events:
4104 Logger myLogger = Logger.getLogger("some-logger");
4106 myLogger.addHandler(lttngUstLogHandler);
4110 . Use `java.util.logging` log statements and configuration as usual.
4111 The loggers with an attached LTTng-UST log handler can emit
4114 . Before exiting the application, remove the LTTng-UST log handler from
4115 the loggers attached to it and call its `close()` method:
4120 myLogger.removeHandler(lttngUstLogHandler);
4121 lttngUstLogHandler.close();
4125 This is not strictly necessary, but it is recommended for a clean
4126 disposal of the handler's resources.
4128 . Include the LTTng-UST Java agent's common and JUL-specific JAR files,
4129 path:{lttng-ust-agent-common.jar} and path:{lttng-ust-agent-jul.jar},
4131 https://docs.oracle.com/javase/tutorial/essential/environment/paths.html[class
4132 path] when you build the Java application.
4134 The JAR files are typically located in dir:{/usr/share/java}.
4136 IMPORTANT: The LTTng-UST Java agent must be
4137 <<installing-lttng,installed>> for the logging framework your
4140 .Use the LTTng-UST Java agent for `java.util.logging`.
4145 import java.io.IOException;
4146 import java.util.logging.Handler;
4147 import java.util.logging.Logger;
4148 import org.lttng.ust.agent.jul.LttngLogHandler;
4152 private static final int answer = 42;
4154 public static void main(String[] argv) throws Exception
4157 Logger logger = Logger.getLogger("jello");
4159 // Create an LTTng-UST log handler
4160 Handler lttngUstLogHandler = new LttngLogHandler();
4162 // Add the LTTng-UST log handler to our logger
4163 logger.addHandler(lttngUstLogHandler);
4166 logger.info("some info");
4167 logger.warning("some warning");
4169 logger.finer("finer information; the answer is " + answer);
4171 logger.severe("error!");
4173 // Not mandatory, but cleaner
4174 logger.removeHandler(lttngUstLogHandler);
4175 lttngUstLogHandler.close();
4184 $ javac -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-jul.jar Test.java
4187 <<creating-destroying-tracing-sessions,Create a tracing session>>,
4188 <<enabling-disabling-events,create an event rule>> matching the
4189 `jello` JUL logger, and <<basic-tracing-session-control,start tracing>>:
4194 $ lttng enable-event --jul jello
4198 Run the compiled class:
4202 $ java -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-jul.jar:. Test
4205 <<basic-tracing-session-control,Stop tracing>> and inspect the
4215 In the resulting trace, an <<event,event record>> generated by a Java
4216 application using `java.util.logging` is named `lttng_jul:event` and
4217 has the following fields:
4220 Log record's message.
4226 Name of the class in which the log statement was executed.
4229 Name of the method in which the log statement was executed.
4232 Logging time (timestamp in milliseconds).
4235 Log level integer value.
4238 ID of the thread in which the log statement was executed.
4240 You can use the opt:lttng-enable-event(1):--loglevel or
4241 opt:lttng-enable-event(1):--loglevel-only option of the
4242 man:lttng-enable-event(1) command to target a range of JUL log levels
4243 or a specific JUL log level.
4248 ==== Use the LTTng-UST Java agent for Apache log4j
4250 To use the LTTng-UST Java agent in a Java application which uses
4253 . In the Java application's source code, import the LTTng-UST
4254 log appender package for Apache log4j:
4259 import org.lttng.ust.agent.log4j.LttngLogAppender;
4263 . Create an LTTng-UST log4j log appender:
4268 Appender lttngUstLogAppender = new LttngLogAppender();
4272 . Add this appender to the log4j loggers which should emit LTTng events:
4277 Logger myLogger = Logger.getLogger("some-logger");
4279 myLogger.addAppender(lttngUstLogAppender);
4283 . Use Apache log4j log statements and configuration as usual. The
4284 loggers with an attached LTTng-UST log appender can emit LTTng events.
4286 . Before exiting the application, remove the LTTng-UST log appender from
4287 the loggers attached to it and call its `close()` method:
4292 myLogger.removeAppender(lttngUstLogAppender);
4293 lttngUstLogAppender.close();
4297 This is not strictly necessary, but it is recommended for a clean
4298 disposal of the appender's resources.
4300 . Include the LTTng-UST Java agent's common and log4j-specific JAR
4301 files, path:{lttng-ust-agent-common.jar} and
4302 path:{lttng-ust-agent-log4j.jar}, in the
4303 https://docs.oracle.com/javase/tutorial/essential/environment/paths.html[class
4304 path] when you build the Java application.
4306 The JAR files are typically located in dir:{/usr/share/java}.
4308 IMPORTANT: The LTTng-UST Java agent must be
4309 <<installing-lttng,installed>> for the logging framework your
4312 .Use the LTTng-UST Java agent for Apache log4j.
4317 import org.apache.log4j.Appender;
4318 import org.apache.log4j.Logger;
4319 import org.lttng.ust.agent.log4j.LttngLogAppender;
4323 private static final int answer = 42;
4325 public static void main(String[] argv) throws Exception
4328 Logger logger = Logger.getLogger("jello");
4330 // Create an LTTng-UST log appender
4331 Appender lttngUstLogAppender = new LttngLogAppender();
4333 // Add the LTTng-UST log appender to our logger
4334 logger.addAppender(lttngUstLogAppender);
4337 logger.info("some info");
4338 logger.warn("some warning");
4340 logger.debug("debug information; the answer is " + answer);
4342 logger.fatal("error!");
4344 // Not mandatory, but cleaner
4345 logger.removeAppender(lttngUstLogAppender);
4346 lttngUstLogAppender.close();
4352 Build this example (`$LOG4JPATH` is the path to the Apache log4j JAR
4357 $ javac -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-log4j.jar:$LOG4JPATH Test.java
4360 <<creating-destroying-tracing-sessions,Create a tracing session>>,
4361 <<enabling-disabling-events,create an event rule>> matching the
4362 `jello` log4j logger, and <<basic-tracing-session-control,start tracing>>:
4367 $ lttng enable-event --log4j jello
4371 Run the compiled class:
4375 $ java -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-log4j.jar:$LOG4JPATH:. Test
4378 <<basic-tracing-session-control,Stop tracing>> and inspect the
4388 In the resulting trace, an <<event,event record>> generated by a Java
4389 application using log4j is named `lttng_log4j:event` and
4390 has the following fields:
4393 Log record's message.
4399 Name of the class in which the log statement was executed.
4402 Name of the method in which the log statement was executed.
4405 Name of the file in which the executed log statement is located.
4408 Line number at which the log statement was executed.
4414 Log level integer value.
4417 Name of the Java thread in which the log statement was executed.
4419 You can use the opt:lttng-enable-event(1):--loglevel or
4420 opt:lttng-enable-event(1):--loglevel-only option of the
4421 man:lttng-enable-event(1) command to target a range of Apache log4j log levels
4422 or a specific log4j log level.
4426 [[java-application-context]]
4427 ==== Provide application-specific context fields in a Java application
4429 A Java application-specific context field is a piece of state provided
4430 by the application which <<adding-context,you can add>>, using the
4431 man:lttng-add-context(1) command, to each <<event,event record>>
4432 produced by the log statements of this application.
4434 For example, a given object might have a current request ID variable.
4435 You can create a context information retriever for this object and
4436 assign a name to this current request ID. You can then, using the
4437 man:lttng-add-context(1) command, add this context field by name to
4438 the JUL or log4j <<channel,channel>>.
4440 To provide application-specific context fields in a Java application:
4442 . In the Java application's source code, import the LTTng-UST
4443 Java agent context classes and interfaces:
4448 import org.lttng.ust.agent.context.ContextInfoManager;
4449 import org.lttng.ust.agent.context.IContextInfoRetriever;
4453 . Create a context information retriever class, that is, a class which
4454 implements the `IContextInfoRetriever` interface:
4459 class MyContextInfoRetriever implements IContextInfoRetriever
4462 public Object retrieveContextInfo(String key)
4464 if (key.equals("intCtx")) {
4466 } else if (key.equals("strContext")) {
4467 return "context value!";
4476 This `retrieveContextInfo()` method is the only member of the
4477 `IContextInfoRetriever` interface. Its role is to return the current
4478 value of a state by name to create a context field. The names of the
4479 context fields and which state variables they return depends on your
4482 All primitive types and objects are supported as context fields.
4483 When `retrieveContextInfo()` returns an object, the context field
4484 serializer calls its `toString()` method to add a string field to
4485 event records. The method can also return `null`, which means that
4486 no context field is available for the required name.
4488 . Register an instance of your context information retriever class to
4489 the context information manager singleton:
4494 IContextInfoRetriever cir = new MyContextInfoRetriever();
4495 ContextInfoManager cim = ContextInfoManager.getInstance();
4496 cim.registerContextInfoRetriever("retrieverName", cir);
4500 . Before exiting the application, remove your context information
4501 retriever from the context information manager singleton:
4506 ContextInfoManager cim = ContextInfoManager.getInstance();
4507 cim.unregisterContextInfoRetriever("retrieverName");
4511 This is not strictly necessary, but it is recommended for a clean
4512 disposal of some manager's resources.
4514 . Build your Java application with LTTng-UST Java agent support as
4515 usual, following the procedure for either the <<jul,JUL>> or
4516 <<log4j,Apache log4j>> framework.
4519 .Provide application-specific context fields in a Java application.
4524 import java.util.logging.Handler;
4525 import java.util.logging.Logger;
4526 import org.lttng.ust.agent.jul.LttngLogHandler;
4527 import org.lttng.ust.agent.context.ContextInfoManager;
4528 import org.lttng.ust.agent.context.IContextInfoRetriever;
4532 // Our context information retriever class
4533 private static class MyContextInfoRetriever
4534 implements IContextInfoRetriever
4537 public Object retrieveContextInfo(String key) {
4538 if (key.equals("intCtx")) {
4540 } else if (key.equals("strContext")) {
4541 return "context value!";
4548 private static final int answer = 42;
4550 public static void main(String args[]) throws Exception
4552 // Get the context information manager instance
4553 ContextInfoManager cim = ContextInfoManager.getInstance();
4555 // Create and register our context information retriever
4556 IContextInfoRetriever cir = new MyContextInfoRetriever();
4557 cim.registerContextInfoRetriever("myRetriever", cir);
4560 Logger logger = Logger.getLogger("jello");
4562 // Create an LTTng-UST log handler
4563 Handler lttngUstLogHandler = new LttngLogHandler();
4565 // Add the LTTng-UST log handler to our logger
4566 logger.addHandler(lttngUstLogHandler);
4569 logger.info("some info");
4570 logger.warning("some warning");
4572 logger.finer("finer information; the answer is " + answer);
4574 logger.severe("error!");
4576 // Not mandatory, but cleaner
4577 logger.removeHandler(lttngUstLogHandler);
4578 lttngUstLogHandler.close();
4579 cim.unregisterContextInfoRetriever("myRetriever");
4588 $ javac -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-jul.jar Test.java
4591 <<creating-destroying-tracing-sessions,Create a tracing session>>
4592 and <<enabling-disabling-events,create an event rule>> matching the
4598 $ lttng enable-event --jul jello
4601 <<adding-context,Add the application-specific context fields>> to the
4606 $ lttng add-context --jul --type='$app.myRetriever:intCtx'
4607 $ lttng add-context --jul --type='$app.myRetriever:strContext'
4610 <<basic-tracing-session-control,Start tracing>>:
4617 Run the compiled class:
4621 $ java -cp /usr/share/java/jarpath/lttng-ust-agent-common.jar:/usr/share/java/jarpath/lttng-ust-agent-jul.jar:. Test
4624 <<basic-tracing-session-control,Stop tracing>> and inspect the
4636 [[python-application]]
4637 === User space Python agent
4639 You can instrument a Python 2 or Python 3 application which uses the
4640 standard https://docs.python.org/3/library/logging.html[`logging`]
4643 Each log statement emits an LTTng event once the
4644 application module imports the
4645 <<lttng-ust-agents,LTTng-UST Python agent>> package.
4648 .A Python application importing the LTTng-UST Python agent.
4649 image::python-app.png[]
4651 To use the LTTng-UST Python agent:
4653 . In the Python application's source code, import the LTTng-UST Python
4663 The LTTng-UST Python agent automatically adds its logging handler to the
4664 root logger at import time.
4666 Any log statement that the application executes before this import does
4667 not emit an LTTng event.
4669 IMPORTANT: The LTTng-UST Python agent must be
4670 <<installing-lttng,installed>>.
4672 . Use log statements and logging configuration as usual.
4673 Since the LTTng-UST Python agent adds a handler to the _root_
4674 logger, you can trace any log statement from any logger.
4676 .Use the LTTng-UST Python agent.
4687 logging.basicConfig()
4688 logger = logging.getLogger('my-logger')
4691 logger.debug('debug message')
4692 logger.info('info message')
4693 logger.warn('warn message')
4694 logger.error('error message')
4695 logger.critical('critical message')
4699 if __name__ == '__main__':
4703 NOTE: `logging.basicConfig()`, which adds to the root logger a basic
4704 logging handler which prints to the standard error stream, is not
4705 strictly required for LTTng-UST tracing to work, but in versions of
4706 Python preceding 3.2, you could see a warning message which indicates
4707 that no handler exists for the logger `my-logger`.
4709 <<creating-destroying-tracing-sessions,Create a tracing session>>,
4710 <<enabling-disabling-events,create an event rule>> matching the
4711 `my-logger` Python logger, and <<basic-tracing-session-control,start
4717 $ lttng enable-event --python my-logger
4721 Run the Python script:
4728 <<basic-tracing-session-control,Stop tracing>> and inspect the recorded
4738 In the resulting trace, an <<event,event record>> generated by a Python
4739 application is named `lttng_python:event` and has the following fields:
4742 Logging time (string).
4745 Log record's message.
4751 Name of the function in which the log statement was executed.
4754 Line number at which the log statement was executed.
4757 Log level integer value.
4760 ID of the Python thread in which the log statement was executed.
4763 Name of the Python thread in which the log statement was executed.
4765 You can use the opt:lttng-enable-event(1):--loglevel or
4766 opt:lttng-enable-event(1):--loglevel-only option of the
4767 man:lttng-enable-event(1) command to target a range of Python log levels
4768 or a specific Python log level.
4770 When an application imports the LTTng-UST Python agent, the agent tries
4771 to register to a <<lttng-sessiond,session daemon>>. Note that you must
4772 <<start-sessiond,start the session daemon>> _before_ you run the Python
4773 application. If a session daemon is found, the agent tries to register
4774 to it during 5{nbsp}seconds, after which the application continues
4775 without LTTng tracing support. You can override this timeout value with
4776 the env:LTTNG_UST_PYTHON_REGISTER_TIMEOUT environment variable
4779 If the session daemon stops while a Python application with an imported
4780 LTTng-UST Python agent runs, the agent retries to connect and to
4781 register to a session daemon every 3{nbsp}seconds. You can override this
4782 delay with the env:LTTNG_UST_PYTHON_REGISTER_RETRY_DELAY environment
4787 [[proc-lttng-logger-abi]]
4790 The `lttng-tracer` Linux kernel module, part of
4791 <<lttng-modules,LTTng-modules>>, creates the special LTTng logger file
4792 path:{/proc/lttng-logger} when it's loaded. Any application can write
4793 text data to this file to emit an LTTng event.
4796 .An application writes to the LTTng logger file to emit an LTTng event.
4797 image::lttng-logger.png[]
4799 The LTTng logger is the quickest method--not the most efficient,
4800 however--to add instrumentation to an application. It is designed
4801 mostly to instrument shell scripts:
4805 $ echo "Some message, some $variable" > /proc/lttng-logger
4808 Any event that the LTTng logger emits is named `lttng_logger` and
4809 belongs to the Linux kernel <<domain,tracing domain>>. However, unlike
4810 other instrumentation points in the kernel tracing domain, **any Unix
4811 user** can <<enabling-disabling-events,create an event rule>> which
4812 matches its event name, not only the root user or users in the
4813 <<tracing-group,tracing group>>.
4815 To use the LTTng logger:
4817 * From any application, write text data to the path:{/proc/lttng-logger}
4820 The `msg` field of `lttng_logger` event records contains the
4823 NOTE: The maximum message length of an LTTng logger event is
4824 1024{nbsp}bytes. Writing more than this makes the LTTng logger emit more
4825 than one event to contain the remaining data.
4827 You should not use the LTTng logger to trace a user application which
4828 can be instrumented in a more efficient way, namely:
4830 * <<c-application,C and $$C++$$ applications>>.
4831 * <<java-application,Java applications>>.
4832 * <<python-application,Python applications>>.
4834 .Use the LTTng logger.
4839 echo 'Hello, World!' > /proc/lttng-logger
4841 df --human-readable --print-type / > /proc/lttng-logger
4844 <<creating-destroying-tracing-sessions,Create a tracing session>>,
4845 <<enabling-disabling-events,create an event rule>> matching the
4846 `lttng_logger` Linux kernel tracepoint, and
4847 <<basic-tracing-session-control,start tracing>>:
4852 $ lttng enable-event --kernel lttng_logger
4856 Run the Bash script:
4863 <<basic-tracing-session-control,Stop tracing>> and inspect the recorded
4874 [[instrumenting-linux-kernel]]
4875 === LTTng kernel tracepoints
4877 NOTE: This section shows how to _add_ instrumentation points to the
4878 Linux kernel. The kernel's subsystems are already thoroughly
4879 instrumented at strategic places for LTTng when you
4880 <<installing-lttng,install>> the <<lttng-modules,LTTng-modules>>
4884 There are two methods to instrument the Linux kernel:
4886 . <<linux-add-lttng-layer,Add an LTTng layer>> over an existing ftrace
4887 tracepoint which uses the `TRACE_EVENT()` API.
4889 Choose this if you want to instrumentation a Linux kernel tree with an
4890 instrumentation point compatible with ftrace, perf, and SystemTap.
4892 . Use an <<linux-lttng-tracepoint-event,LTTng-only approach>> to
4893 instrument an out-of-tree kernel module.
4895 Choose this if you don't need ftrace, perf, or SystemTap support.
4899 [[linux-add-lttng-layer]]
4900 ==== [[instrumenting-linux-kernel-itself]][[mainline-trace-event]][[lttng-adaptation-layer]]Add an LTTng layer to an existing ftrace tracepoint
4902 This section shows how to add an LTTng layer to existing ftrace
4903 instrumentation using the `TRACE_EVENT()` API.
4905 This section does not document the `TRACE_EVENT()` macro. You can
4906 read the following articles to learn more about this API:
4908 * http://lwn.net/Articles/379903/[Using the TRACE_EVENT() macro (Part 1)]
4909 * http://lwn.net/Articles/381064/[Using the TRACE_EVENT() macro (Part 2)]
4910 * http://lwn.net/Articles/383362/[Using the TRACE_EVENT() macro (Part 3)]
4912 The following procedure assumes that your ftrace tracepoints are
4913 correctly defined in their own header and that they are created in
4914 one source file using the `CREATE_TRACE_POINTS` definition.
4916 To add an LTTng layer over an existing ftrace tracepoint:
4918 . Make sure the following kernel configuration options are
4924 * `CONFIG_HIGH_RES_TIMERS`
4925 * `CONFIG_TRACEPOINTS`
4928 . Build the Linux source tree with your custom ftrace tracepoints.
4929 . Boot the resulting Linux image on your target system.
4931 Confirm that the tracepoints exist by looking for their names in the
4932 dir:{/sys/kernel/debug/tracing/events/subsys} directory, where `subsys`
4933 is your subsystem's name.
4935 . Get a copy of the latest LTTng-modules{nbsp}{revision}:
4940 $ cd $(mktemp -d) &&
4941 wget http://lttng.org/files/lttng-modules/lttng-modules-latest-2.10.tar.bz2 &&
4942 tar -xf lttng-modules-latest-2.10.tar.bz2 &&
4943 cd lttng-modules-2.10.*
4947 . In dir:{instrumentation/events/lttng-module}, relative to the root
4948 of the LTTng-modules source tree, create a header file named
4949 +__subsys__.h+ for your custom subsystem +__subsys__+ and write your
4950 LTTng-modules tracepoint definitions using the LTTng-modules
4953 Start with this template:
4957 .path:{instrumentation/events/lttng-module/my_subsys.h}
4960 #define TRACE_SYSTEM my_subsys
4962 #if !defined(_LTTNG_MY_SUBSYS_H) || defined(TRACE_HEADER_MULTI_READ)
4963 #define _LTTNG_MY_SUBSYS_H
4965 #include "../../../probes/lttng-tracepoint-event.h"
4966 #include <linux/tracepoint.h>
4968 LTTNG_TRACEPOINT_EVENT(
4970 * Format is identical to TRACE_EVENT()'s version for the three
4971 * following macro parameters:
4974 TP_PROTO(int my_int, const char *my_string),
4975 TP_ARGS(my_int, my_string),
4977 /* LTTng-modules specific macros */
4979 ctf_integer(int, my_int_field, my_int)
4980 ctf_string(my_bar_field, my_bar)
4984 #endif /* !defined(_LTTNG_MY_SUBSYS_H) || defined(TRACE_HEADER_MULTI_READ) */
4986 #include "../../../probes/define_trace.h"
4990 The entries in the `TP_FIELDS()` section are the list of fields for the
4991 LTTng tracepoint. This is similar to the `TP_STRUCT__entry()` part of
4992 ftrace's `TRACE_EVENT()` macro.
4994 See <<lttng-modules-tp-fields,Tracepoint fields macros>> for a
4995 complete description of the available `ctf_*()` macros.
4997 . Create the LTTng-modules probe's kernel module C source file,
4998 +probes/lttng-probe-__subsys__.c+, where +__subsys__+ is your
5003 .path:{probes/lttng-probe-my-subsys.c}
5005 #include <linux/module.h>
5006 #include "../lttng-tracer.h"
5009 * Build-time verification of mismatch between mainline
5010 * TRACE_EVENT() arguments and the LTTng-modules adaptation
5011 * layer LTTNG_TRACEPOINT_EVENT() arguments.
5013 #include <trace/events/my_subsys.h>
5015 /* Create LTTng tracepoint probes */
5016 #define LTTNG_PACKAGE_BUILD
5017 #define CREATE_TRACE_POINTS
5018 #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
5020 #include "../instrumentation/events/lttng-module/my_subsys.h"
5022 MODULE_LICENSE("GPL and additional rights");
5023 MODULE_AUTHOR("Your name <your-email>");
5024 MODULE_DESCRIPTION("LTTng my_subsys probes");
5025 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
5026 __stringify(LTTNG_MODULES_MINOR_VERSION) "."
5027 __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
5028 LTTNG_MODULES_EXTRAVERSION);
5032 . Edit path:{probes/KBuild} and add your new kernel module object
5033 next to the existing ones:
5037 .path:{probes/KBuild}
5041 obj-m += lttng-probe-module.o
5042 obj-m += lttng-probe-power.o
5044 obj-m += lttng-probe-my-subsys.o
5050 . Build and install the LTTng kernel modules:
5055 $ make KERNELDIR=/path/to/linux
5056 # make modules_install && depmod -a
5060 Replace `/path/to/linux` with the path to the Linux source tree where
5061 you defined and used tracepoints with ftrace's `TRACE_EVENT()` macro.
5063 Note that you can also use the
5064 <<lttng-tracepoint-event-code,`LTTNG_TRACEPOINT_EVENT_CODE()` macro>>
5065 instead of `LTTNG_TRACEPOINT_EVENT()` to use custom local variables and
5066 C code that need to be executed before the event fields are recorded.
5068 The best way to learn how to use the previous LTTng-modules macros is to
5069 inspect the existing LTTng-modules tracepoint definitions in the
5070 dir:{instrumentation/events/lttng-module} header files. Compare them
5071 with the Linux kernel mainline versions in the
5072 dir:{include/trace/events} directory of the Linux source tree.
5076 [[lttng-tracepoint-event-code]]
5077 ===== Use custom C code to access the data for tracepoint fields
5079 Although we recommended to always use the
5080 <<lttng-adaptation-layer,`LTTNG_TRACEPOINT_EVENT()`>> macro to describe
5081 the arguments and fields of an LTTng-modules tracepoint when possible,
5082 sometimes you need a more complex process to access the data that the
5083 tracer records as event record fields. In other words, you need local
5084 variables and multiple C{nbsp}statements instead of simple
5085 argument-based expressions that you pass to the
5086 <<lttng-modules-tp-fields,`ctf_*()` macros of `TP_FIELDS()`>>.
5088 You can use the `LTTNG_TRACEPOINT_EVENT_CODE()` macro instead of
5089 `LTTNG_TRACEPOINT_EVENT()` to declare custom local variables and define
5090 a block of C{nbsp}code to be executed before LTTng records the fields.
5091 The structure of this macro is:
5094 .`LTTNG_TRACEPOINT_EVENT_CODE()` macro syntax.
5096 LTTNG_TRACEPOINT_EVENT_CODE(
5098 * Format identical to the LTTNG_TRACEPOINT_EVENT()
5099 * version for the following three macro parameters:
5102 TP_PROTO(int my_int, const char *my_string),
5103 TP_ARGS(my_int, my_string),
5105 /* Declarations of custom local variables */
5108 unsigned long b = 0;
5109 const char *name = "(undefined)";
5110 struct my_struct *my_struct;
5114 * Custom code which uses both tracepoint arguments
5115 * (in TP_ARGS()) and local variables (in TP_locvar()).
5117 * Local variables are actually members of a structure pointed
5118 * to by the special variable tp_locvar.
5122 tp_locvar->a = my_int + 17;
5123 tp_locvar->my_struct = get_my_struct_at(tp_locvar->a);
5124 tp_locvar->b = my_struct_compute_b(tp_locvar->my_struct);
5125 tp_locvar->name = my_struct_get_name(tp_locvar->my_struct);
5126 put_my_struct(tp_locvar->my_struct);
5135 * Format identical to the LTTNG_TRACEPOINT_EVENT()
5136 * version for this, except that tp_locvar members can be
5137 * used in the argument expression parameters of
5138 * the ctf_*() macros.
5141 ctf_integer(unsigned long, my_struct_b, tp_locvar->b)
5142 ctf_integer(int, my_struct_a, tp_locvar->a)
5143 ctf_string(my_string_field, my_string)
5144 ctf_string(my_struct_name, tp_locvar->name)
5149 IMPORTANT: The C code defined in `TP_code()` must not have any side
5150 effects when executed. In particular, the code must not allocate
5151 memory or get resources without deallocating this memory or putting
5152 those resources afterwards.
5155 [[instrumenting-linux-kernel-tracing]]
5156 ==== Load and unload a custom probe kernel module
5158 You must load a <<lttng-adaptation-layer,created LTTng-modules probe
5159 kernel module>> in the kernel before it can emit LTTng events.
5161 To load the default probe kernel modules and a custom probe kernel
5164 * Use the opt:lttng-sessiond(8):--extra-kmod-probes option to give extra
5165 probe modules to load when starting a root <<lttng-sessiond,session
5169 .Load the `my_subsys`, `usb`, and the default probe modules.
5173 # lttng-sessiond --extra-kmod-probes=my_subsys,usb
5178 You only need to pass the subsystem name, not the whole kernel module
5181 To load _only_ a given custom probe kernel module:
5183 * Use the opt:lttng-sessiond(8):--kmod-probes option to give the probe
5184 modules to load when starting a root session daemon:
5187 .Load only the `my_subsys` and `usb` probe modules.
5191 # lttng-sessiond --kmod-probes=my_subsys,usb
5196 To confirm that a probe module is loaded:
5203 $ lsmod | grep lttng_probe_usb
5207 To unload the loaded probe modules:
5209 * Kill the session daemon with `SIGTERM`:
5214 # pkill lttng-sessiond
5218 You can also use man:modprobe(8)'s `--remove` option if the session
5219 daemon terminates abnormally.
5222 [[controlling-tracing]]
5225 Once an application or a Linux kernel is
5226 <<instrumenting,instrumented>> for LTTng tracing,
5229 This section is divided in topics on how to use the various
5230 <<plumbing,components of LTTng>>, in particular the <<lttng-cli,cmd:lttng
5231 command-line tool>>, to _control_ the LTTng daemons and tracers.
5233 NOTE: In the following subsections, we refer to an man:lttng(1) command
5234 using its man page name. For example, instead of _Run the `create`
5235 command to..._, we use _Run the man:lttng-create(1) command to..._.
5239 === Start a session daemon
5241 In some situations, you need to run a <<lttng-sessiond,session daemon>>
5242 (man:lttng-sessiond(8)) _before_ you can use the man:lttng(1)
5245 You will see the following error when you run a command while no session
5249 Error: No session daemon is available
5252 The only command that automatically runs a session daemon is
5253 man:lttng-create(1), which you use to
5254 <<creating-destroying-tracing-sessions,create a tracing session>>. While
5255 this is most of the time the first operation that you do, sometimes it's
5256 not. Some examples are:
5258 * <<list-instrumentation-points,List the available instrumentation points>>.
5259 * <<saving-loading-tracing-session,Load a tracing session configuration>>.
5261 [[tracing-group]] Each Unix user must have its own running session
5262 daemon to trace user applications. The session daemon that the root user
5263 starts is the only one allowed to control the LTTng kernel tracer. Users
5264 that are part of the _tracing group_ can control the root session
5265 daemon. The default tracing group name is `tracing`; you can set it to
5266 something else with the opt:lttng-sessiond(8):--group option when you
5267 start the root session daemon.
5269 To start a user session daemon:
5271 * Run man:lttng-sessiond(8):
5276 $ lttng-sessiond --daemonize
5280 To start the root session daemon:
5282 * Run man:lttng-sessiond(8) as the root user:
5287 # lttng-sessiond --daemonize
5291 In both cases, remove the opt:lttng-sessiond(8):--daemonize option to
5292 start the session daemon in foreground.
5294 To stop a session daemon, use man:kill(1) on its process ID (standard
5297 Note that some Linux distributions could manage the LTTng session daemon
5298 as a service. In this case, you should use the service manager to
5299 start, restart, and stop session daemons.
5302 [[creating-destroying-tracing-sessions]]
5303 === Create and destroy a tracing session
5305 Almost all the LTTng control operations happen in the scope of
5306 a <<tracing-session,tracing session>>, which is the dialogue between the
5307 <<lttng-sessiond,session daemon>> and you.
5309 To create a tracing session with a generated name:
5311 * Use the man:lttng-create(1) command:
5320 The created tracing session's name is `auto` followed by the
5323 To create a tracing session with a specific name:
5325 * Use the optional argument of the man:lttng-create(1) command:
5330 $ lttng create my-session
5334 Replace `my-session` with the specific tracing session name.
5336 LTTng appends the creation date to the created tracing session's name.
5338 LTTng writes the traces of a tracing session in
5339 +$LTTNG_HOME/lttng-trace/__name__+ by default, where +__name__+ is the
5340 name of the tracing session. Note that the env:LTTNG_HOME environment
5341 variable defaults to `$HOME` if not set.
5343 To output LTTng traces to a non-default location:
5345 * Use the opt:lttng-create(1):--output option of the man:lttng-create(1) command:
5350 $ lttng create my-session --output=/tmp/some-directory
5354 You may create as many tracing sessions as you wish.
5356 To list all the existing tracing sessions for your Unix user:
5358 * Use the man:lttng-list(1) command:
5367 When you create a tracing session, it is set as the _current tracing
5368 session_. The following man:lttng(1) commands operate on the current
5369 tracing session when you don't specify one:
5371 [role="list-3-cols"]
5388 To change the current tracing session:
5390 * Use the man:lttng-set-session(1) command:
5395 $ lttng set-session new-session
5399 Replace `new-session` by the name of the new current tracing session.
5401 When you are done tracing in a given tracing session, you can destroy
5402 it. This operation frees the resources taken by the tracing session
5403 to destroy; it does not destroy the trace data that LTTng wrote for
5404 this tracing session.
5406 To destroy the current tracing session:
5408 * Use the man:lttng-destroy(1) command:
5417 The man:lttng-destroy(1) command also runs the man:lttng-stop(1)
5418 command implicitly (see <<basic-tracing-session-control,Start and stop a
5419 tracing session>>). You need to stop tracing to make LTTng flush the
5420 remaining trace data and make the trace readable.
5423 [[list-instrumentation-points]]
5424 === List the available instrumentation points
5426 The <<lttng-sessiond,session daemon>> can query the running instrumented
5427 user applications and the Linux kernel to get a list of available
5428 instrumentation points. For the Linux kernel <<domain,tracing domain>>,
5429 they are tracepoints and system calls. For the user space tracing
5430 domain, they are tracepoints. For the other tracing domains, they are
5433 To list the available instrumentation points:
5435 * Use the man:lttng-list(1) command with the requested tracing domain's
5439 * opt:lttng-list(1):--kernel: Linux kernel tracepoints (your Unix user
5440 must be a root user, or it must be a member of the
5441 <<tracing-group,tracing group>>).
5442 * opt:lttng-list(1):--kernel with opt:lttng-list(1):--syscall: Linux
5443 kernel system calls (your Unix user must be a root user, or it must be
5444 a member of the tracing group).
5445 * opt:lttng-list(1):--userspace: user space tracepoints.
5446 * opt:lttng-list(1):--jul: `java.util.logging` loggers.
5447 * opt:lttng-list(1):--log4j: Apache log4j loggers.
5448 * opt:lttng-list(1):--python: Python loggers.
5451 .List the available user space tracepoints.
5455 $ lttng list --userspace
5459 .List the available Linux kernel system call tracepoints.
5463 $ lttng list --kernel --syscall
5468 [[enabling-disabling-events]]
5469 === Create and enable an event rule
5471 Once you <<creating-destroying-tracing-sessions,create a tracing
5472 session>>, you can create <<event,event rules>> with the
5473 man:lttng-enable-event(1) command.
5475 You specify each condition with a command-line option. The available
5476 condition options are shown in the following table.
5478 [role="growable",cols="asciidoc,asciidoc,default"]
5479 .Condition command-line options for the man:lttng-enable-event(1) command.
5481 |Option |Description |Applicable tracing domains
5487 . +--probe=__ADDR__+
5488 . +--function=__ADDR__+
5491 Instead of using the default _tracepoint_ instrumentation type, use:
5493 . A Linux system call.
5494 . A Linux https://lwn.net/Articles/132196/[KProbe] (symbol or address).
5495 . The entry and return points of a Linux function (symbol or address).
5499 |First positional argument.
5502 Tracepoint or system call name. In the case of a Linux KProbe or
5503 function, this is a custom name given to the event rule. With the
5504 JUL, log4j, and Python domains, this is a logger name.
5506 With a tracepoint, logger, or system call name, you can use the special
5507 `*` globbing character to match anything (for example, `sched_*`,
5515 . +--loglevel=__LEVEL__+
5516 . +--loglevel-only=__LEVEL__+
5519 . Match only tracepoints or log statements with a logging level at
5520 least as severe as +__LEVEL__+.
5521 . Match only tracepoints or log statements with a logging level
5522 equal to +__LEVEL__+.
5524 See man:lttng-enable-event(1) for the list of available logging level
5527 |User space, JUL, log4j, and Python.
5529 |+--exclude=__EXCLUSIONS__+
5532 When you use a `*` character at the end of the tracepoint or logger
5533 name (first positional argument), exclude the specific names in the
5534 comma-delimited list +__EXCLUSIONS__+.
5537 User space, JUL, log4j, and Python.
5539 |+--filter=__EXPR__+
5542 Match only events which satisfy the expression +__EXPR__+.
5544 See man:lttng-enable-event(1) to learn more about the syntax of a
5551 You attach an event rule to a <<channel,channel>> on creation. If you do
5552 not specify the channel with the opt:lttng-enable-event(1):--channel
5553 option, and if the event rule to create is the first in its
5554 <<domain,tracing domain>> for a given tracing session, then LTTng
5555 creates a _default channel_ for you. This default channel is reused in
5556 subsequent invocations of the man:lttng-enable-event(1) command for the
5557 same tracing domain.
5559 An event rule is always enabled at creation time.
5561 The following examples show how you can combine the previous
5562 command-line options to create simple to more complex event rules.
5564 .Create an event rule targetting a Linux kernel tracepoint (default channel).
5568 $ lttng enable-event --kernel sched_switch
5572 .Create an event rule matching four Linux kernel system calls (default channel).
5576 $ lttng enable-event --kernel --syscall open,write,read,close
5580 .Create event rules matching tracepoints with filter expressions (default channel).
5584 $ lttng enable-event --kernel sched_switch --filter='prev_comm == "bash"'
5589 $ lttng enable-event --kernel --all \
5590 --filter='$ctx.tid == 1988 || $ctx.tid == 1534'
5595 $ lttng enable-event --jul my_logger \
5596 --filter='$app.retriever:cur_msg_id > 3'
5599 IMPORTANT: Make sure to always quote the filter string when you
5600 use man:lttng(1) from a shell.
5603 .Create an event rule matching any user space tracepoint of a given tracepoint provider with a log level range (default channel).
5607 $ lttng enable-event --userspace my_app:'*' --loglevel=TRACE_INFO
5610 IMPORTANT: Make sure to always quote the wildcard character when you
5611 use man:lttng(1) from a shell.
5614 .Create an event rule matching multiple Python loggers with a wildcard and with exclusions (default channel).
5618 $ lttng enable-event --python my-app.'*' \
5619 --exclude='my-app.module,my-app.hello'
5623 .Create an event rule matching any Apache log4j logger with a specific log level (default channel).
5627 $ lttng enable-event --log4j --all --loglevel-only=LOG4J_WARN
5631 .Create an event rule attached to a specific channel matching a specific user space tracepoint provider and tracepoint.
5635 $ lttng enable-event --userspace my_app:my_tracepoint --channel=my-channel
5639 The event rules of a given channel form a whitelist: as soon as an
5640 emitted event passes one of them, LTTng can record the event. For
5641 example, an event named `my_app:my_tracepoint` emitted from a user space
5642 tracepoint with a `TRACE_ERROR` log level passes both of the following
5647 $ lttng enable-event --userspace my_app:my_tracepoint
5648 $ lttng enable-event --userspace my_app:my_tracepoint \
5649 --loglevel=TRACE_INFO
5652 The second event rule is redundant: the first one includes
5656 [[disable-event-rule]]
5657 === Disable an event rule
5659 To disable an event rule that you <<enabling-disabling-events,created>>
5660 previously, use the man:lttng-disable-event(1) command. This command
5661 disables _all_ the event rules (of a given tracing domain and channel)
5662 which match an instrumentation point. The other conditions are not
5663 supported as of LTTng{nbsp}{revision}.
5665 The LTTng tracer does not record an emitted event which passes
5666 a _disabled_ event rule.
5668 .Disable an event rule matching a Python logger (default channel).
5672 $ lttng disable-event --python my-logger
5676 .Disable an event rule matching all `java.util.logging` loggers (default channel).
5680 $ lttng disable-event --jul '*'
5684 .Disable _all_ the event rules of the default channel.
5686 The opt:lttng-disable-event(1):--all-events option is not, like the
5687 opt:lttng-enable-event(1):--all option of man:lttng-enable-event(1), the
5688 equivalent of the event name `*` (wildcard): it disables _all_ the event
5689 rules of a given channel.
5693 $ lttng disable-event --jul --all-events
5697 NOTE: You cannot delete an event rule once you create it.
5701 === Get the status of a tracing session
5703 To get the status of the current tracing session, that is, its
5704 parameters, its channels, event rules, and their attributes:
5706 * Use the man:lttng-status(1) command:
5716 To get the status of any tracing session:
5718 * Use the man:lttng-list(1) command with the tracing session's name:
5723 $ lttng list my-session
5727 Replace `my-session` with the desired tracing session's name.
5730 [[basic-tracing-session-control]]
5731 === Start and stop a tracing session
5733 Once you <<creating-destroying-tracing-sessions,create a tracing
5735 <<enabling-disabling-events,create one or more event rules>>,
5736 you can start and stop the tracers for this tracing session.
5738 To start tracing in the current tracing session:
5740 * Use the man:lttng-start(1) command:
5749 LTTng is very flexible: you can launch user applications before
5750 or after the you start the tracers. The tracers only record the events
5751 if they pass enabled event rules and if they occur while the tracers are
5754 To stop tracing in the current tracing session:
5756 * Use the man:lttng-stop(1) command:
5765 If there were <<channel-overwrite-mode-vs-discard-mode,lost event
5766 records>> or lost sub-buffers since the last time you ran
5767 man:lttng-start(1), warnings are printed when you run the
5768 man:lttng-stop(1) command.
5770 IMPORTANT: You need to stop tracing to make LTTng flush the remaining
5771 trace data and make the trace readable. Note that the
5772 man:lttng-destroy(1) command (see
5773 <<creating-destroying-tracing-sessions,Create and destroy a tracing
5774 session>>) also runs the man:lttng-stop(1) command implicitly.
5777 [[enabling-disabling-channels]]
5778 === Create a channel
5780 Once you create a tracing session, you can create a <<channel,channel>>
5781 with the man:lttng-enable-channel(1) command.
5783 Note that LTTng automatically creates a default channel when, for a
5784 given <<domain,tracing domain>>, no channels exist and you
5785 <<enabling-disabling-events,create>> the first event rule. This default
5786 channel is named `channel0` and its attributes are set to reasonable
5787 values. Therefore, you only need to create a channel when you need
5788 non-default attributes.
5790 You specify each non-default channel attribute with a command-line
5791 option when you use the man:lttng-enable-channel(1) command. The
5792 available command-line options are:
5794 [role="growable",cols="asciidoc,asciidoc"]
5795 .Command-line options for the man:lttng-enable-channel(1) command.
5797 |Option |Description
5803 <<channel-overwrite-mode-vs-discard-mode,event loss mode>> instead of
5804 the default _discard_ mode.
5806 |`--buffers-pid` (user space tracing domain only)
5809 Use the per-process <<channel-buffering-schemes,buffering scheme>>
5810 instead of the default per-user buffering scheme.
5812 |+--subbuf-size=__SIZE__+
5815 Allocate sub-buffers of +__SIZE__+ bytes (power of two), for each CPU,
5816 either for each Unix user (default), or for each instrumented process.
5818 See <<channel-subbuf-size-vs-subbuf-count,Sub-buffer count and size>>.
5820 |+--num-subbuf=__COUNT__+
5823 Allocate +__COUNT__+ sub-buffers (power of two), for each CPU, either
5824 for each Unix user (default), or for each instrumented process.
5826 See <<channel-subbuf-size-vs-subbuf-count,Sub-buffer count and size>>.
5828 |+--tracefile-size=__SIZE__+
5831 Set the maximum size of each trace file that this channel writes within
5832 a stream to +__SIZE__+ bytes instead of no maximum.
5834 See <<tracefile-rotation,Trace file count and size>>.
5836 |+--tracefile-count=__COUNT__+
5839 Limit the number of trace files that this channel creates to
5840 +__COUNT__+ channels instead of no limit.
5842 See <<tracefile-rotation,Trace file count and size>>.
5844 |+--switch-timer=__PERIODUS__+
5847 Set the <<channel-switch-timer,switch timer period>>
5848 to +__PERIODUS__+{nbsp}µs.
5850 |+--read-timer=__PERIODUS__+
5853 Set the <<channel-read-timer,read timer period>>
5854 to +__PERIODUS__+{nbsp}µs.
5856 |[[opt-blocking-timeout]]+--blocking-timeout=__TIMEOUTUS__+
5859 Set the timeout of user space applications which load LTTng-UST
5860 in blocking mode to +__TIMEOUTUS__+:
5863 Never block (non-blocking mode).
5866 Block forever until space is available in a sub-buffer to record
5869 __n__, a positive value::
5870 Wait for at most __n__ µs when trying to write into a sub-buffer.
5872 Note that, for this option to have any effect on an instrumented
5873 user space application, you need to run the application with a set
5874 env:LTTNG_UST_ALLOW_BLOCKING environment variable.
5876 |+--output=__TYPE__+ (Linux kernel tracing domain only)
5879 Set the channel's output type to +__TYPE__+, either `mmap` or `splice`.
5883 You can only create a channel in the Linux kernel and user space
5884 <<domain,tracing domains>>: other tracing domains have their own channel
5885 created on the fly when <<enabling-disabling-events,creating event
5890 Because of a current LTTng limitation, you must create all channels
5891 _before_ you <<basic-tracing-session-control,start tracing>> in a given
5892 tracing session, that is, before the first time you run
5895 Since LTTng automatically creates a default channel when you use the
5896 man:lttng-enable-event(1) command with a specific tracing domain, you
5897 cannot, for example, create a Linux kernel event rule, start tracing,
5898 and then create a user space event rule, because no user space channel
5899 exists yet and it's too late to create one.
5901 For this reason, make sure to configure your channels properly
5902 before starting the tracers for the first time!
5905 The following examples show how you can combine the previous
5906 command-line options to create simple to more complex channels.
5908 .Create a Linux kernel channel with default attributes.
5912 $ lttng enable-channel --kernel my-channel
5916 .Create a user space channel with 4 sub-buffers or 1{nbsp}MiB each, per CPU, per instrumented process.
5920 $ lttng enable-channel --userspace --num-subbuf=4 --subbuf-size=1M \
5921 --buffers-pid my-channel
5925 .[[blocking-timeout-example]]Create a default user space channel with an infinite blocking timeout.
5927 <<creating-destroying-tracing-sessions,Create a tracing-session>>,
5928 create the channel, <<enabling-disabling-events,create an event rule>>,
5929 and <<basic-tracing-session-control,start tracing>>:
5934 $ lttng enable-channel --userspace --blocking-timeout=inf blocking-channel
5935 $ lttng enable-event --userspace --channel=blocking-channel --all
5939 Run an application instrumented with LTTng-UST and allow it to block:
5943 $ LTTNG_UST_ALLOW_BLOCKING=1 my-app
5947 .Create a Linux kernel channel which rotates 8 trace files of 4{nbsp}MiB each for each stream
5951 $ lttng enable-channel --kernel --tracefile-count=8 \
5952 --tracefile-size=4194304 my-channel
5956 .Create a user space channel in overwrite (or _flight recorder_) mode.
5960 $ lttng enable-channel --userspace --overwrite my-channel
5964 You can <<enabling-disabling-events,create>> the same event rule in
5965 two different channels:
5969 $ lttng enable-event --userspace --channel=my-channel app:tp
5970 $ lttng enable-event --userspace --channel=other-channel app:tp
5973 If both channels are enabled, when a tracepoint named `app:tp` is
5974 reached, LTTng records two events, one for each channel.
5978 === Disable a channel
5980 To disable a specific channel that you <<enabling-disabling-channels,created>>
5981 previously, use the man:lttng-disable-channel(1) command.
5983 .Disable a specific Linux kernel channel.
5987 $ lttng disable-channel --kernel my-channel
5991 The state of a channel precedes the individual states of event rules
5992 attached to it: event rules which belong to a disabled channel, even if
5993 they are enabled, are also considered disabled.
5997 === Add context fields to a channel
5999 Event record fields in trace files provide important information about
6000 events that occured previously, but sometimes some external context may
6001 help you solve a problem faster. Examples of context fields are:
6003 * The **process ID**, **thread ID**, **process name**, and
6004 **process priority** of the thread in which the event occurs.
6005 * The **hostname** of the system on which the event occurs.
6006 * The current values of many possible **performance counters** using
6008 ** CPU cycles, stalled cycles, idle cycles, and the other cycle types.
6010 ** Branch instructions, misses, and loads.
6012 * Any context defined at the application level (supported for the
6013 JUL and log4j <<domain,tracing domains>>).
6015 To get the full list of available context fields, see
6016 `lttng add-context --list`. Some context fields are reserved for a
6017 specific <<domain,tracing domain>> (Linux kernel or user space).
6019 You add context fields to <<channel,channels>>. All the events
6020 that a channel with added context fields records contain those fields.
6022 To add context fields to one or all the channels of a given tracing
6025 * Use the man:lttng-add-context(1) command.
6027 .Add context fields to all the channels of the current tracing session.
6029 The following command line adds the virtual process identifier and
6030 the per-thread CPU cycles count fields to all the user space channels
6031 of the current tracing session.
6035 $ lttng add-context --userspace --type=vpid --type=perf:thread:cpu-cycles
6039 .Add performance counter context fields by raw ID
6041 See man:lttng-add-context(1) for the exact format of the context field
6042 type, which is partly compatible with the format used in
6047 $ lttng add-context --userspace --type=perf:thread:raw:r0110:test
6048 $ lttng add-context --kernel --type=perf:cpu:raw:r0013c:x86unhalted
6052 .Add a context field to a specific channel.
6054 The following command line adds the thread identifier context field
6055 to the Linux kernel channel named `my-channel` in the current
6060 $ lttng add-context --kernel --channel=my-channel --type=tid
6064 .Add an application-specific context field to a specific channel.
6066 The following command line adds the `cur_msg_id` context field of the
6067 `retriever` context retriever for all the instrumented
6068 <<java-application,Java applications>> recording <<event,event records>>
6069 in the channel named `my-channel`:
6073 $ lttng add-context --kernel --channel=my-channel \
6074 --type='$app:retriever:cur_msg_id'
6077 IMPORTANT: Make sure to always quote the `$` character when you
6078 use man:lttng-add-context(1) from a shell.
6081 NOTE: You cannot remove context fields from a channel once you add it.
6086 === Track process IDs
6088 It's often useful to allow only specific process IDs (PIDs) to emit
6089 events. For example, you may wish to record all the system calls made by
6090 a given process (Ă la http://linux.die.net/man/1/strace[strace]).
6092 The man:lttng-track(1) and man:lttng-untrack(1) commands serve this
6093 purpose. Both commands operate on a whitelist of process IDs. You _add_
6094 entries to this whitelist with the man:lttng-track(1) command and remove
6095 entries with the man:lttng-untrack(1) command. Any process which has one
6096 of the PIDs in the whitelist is allowed to emit LTTng events which pass
6097 an enabled <<event,event rule>>.
6099 NOTE: The PID tracker tracks the _numeric process IDs_. Should a
6100 process with a given tracked ID exit and another process be given this
6101 ID, then the latter would also be allowed to emit events.
6103 .Track and untrack process IDs.
6105 For the sake of the following example, assume the target system has 16
6109 <<creating-destroying-tracing-sessions,create a tracing session>>,
6110 the whitelist contains all the possible PIDs:
6113 .All PIDs are tracked.
6114 image::track-all.png[]
6116 When the whitelist is full and you use the man:lttng-track(1) command to
6117 specify some PIDs to track, LTTng first clears the whitelist, then it
6118 tracks the specific PIDs. After:
6122 $ lttng track --pid=3,4,7,10,13
6128 .PIDs 3, 4, 7, 10, and 13 are tracked.
6129 image::track-3-4-7-10-13.png[]
6131 You can add more PIDs to the whitelist afterwards:
6135 $ lttng track --pid=1,15,16
6141 .PIDs 1, 15, and 16 are added to the whitelist.
6142 image::track-1-3-4-7-10-13-15-16.png[]
6144 The man:lttng-untrack(1) command removes entries from the PID tracker's
6145 whitelist. Given the previous example, the following command:
6149 $ lttng untrack --pid=3,7,10,13
6152 leads to this whitelist:
6155 .PIDs 3, 7, 10, and 13 are removed from the whitelist.
6156 image::track-1-4-15-16.png[]
6158 LTTng can track all possible PIDs again using the
6159 opt:lttng-track(1):--all option:
6163 $ lttng track --pid --all
6166 The result is, again:
6169 .All PIDs are tracked.
6170 image::track-all.png[]
6173 .Track only specific PIDs
6175 A very typical use case with PID tracking is to start with an empty
6176 whitelist, then <<basic-tracing-session-control,start the tracers>>, and
6177 then add PIDs manually while tracers are active. You can accomplish this
6178 by using the opt:lttng-untrack(1):--all option of the
6179 man:lttng-untrack(1) command to clear the whitelist after you
6180 <<creating-destroying-tracing-sessions,create a tracing session>>:
6184 $ lttng untrack --pid --all
6190 .No PIDs are tracked.
6191 image::untrack-all.png[]
6193 If you trace with this whitelist configuration, the tracer records no
6194 events for this <<domain,tracing domain>> because no processes are
6195 tracked. You can use the man:lttng-track(1) command as usual to track
6196 specific PIDs, for example:
6200 $ lttng track --pid=6,11
6206 .PIDs 6 and 11 are tracked.
6207 image::track-6-11.png[]
6212 [[saving-loading-tracing-session]]
6213 === Save and load tracing session configurations
6215 Configuring a <<tracing-session,tracing session>> can be long. Some of
6216 the tasks involved are:
6218 * <<enabling-disabling-channels,Create channels>> with
6219 specific attributes.
6220 * <<adding-context,Add context fields>> to specific channels.
6221 * <<enabling-disabling-events,Create event rules>> with specific log
6222 level and filter conditions.
6224 If you use LTTng to solve real world problems, chances are you have to
6225 record events using the same tracing session setup over and over,
6226 modifying a few variables each time in your instrumented program
6227 or environment. To avoid constant tracing session reconfiguration,
6228 the man:lttng(1) command-line tool can save and load tracing session
6229 configurations to/from XML files.
6231 To save a given tracing session configuration:
6233 * Use the man:lttng-save(1) command:
6238 $ lttng save my-session
6242 Replace `my-session` with the name of the tracing session to save.
6244 LTTng saves tracing session configurations to
6245 dir:{$LTTNG_HOME/.lttng/sessions} by default. Note that the
6246 env:LTTNG_HOME environment variable defaults to `$HOME` if not set. Use
6247 the opt:lttng-save(1):--output-path option to change this destination
6250 LTTng saves all configuration parameters, for example:
6252 * The tracing session name.
6253 * The trace data output path.
6254 * The channels with their state and all their attributes.
6255 * The context fields you added to channels.
6256 * The event rules with their state, log level and filter conditions.
6258 To load a tracing session:
6260 * Use the man:lttng-load(1) command:
6265 $ lttng load my-session
6269 Replace `my-session` with the name of the tracing session to load.
6271 When LTTng loads a configuration, it restores your saved tracing session
6272 as if you just configured it manually.
6274 See man:lttng(1) for the complete list of command-line options. You
6275 can also save and load all many sessions at a time, and decide in which
6276 directory to output the XML files.
6279 [[sending-trace-data-over-the-network]]
6280 === Send trace data over the network
6282 LTTng can send the recorded trace data to a remote system over the
6283 network instead of writing it to the local file system.
6285 To send the trace data over the network:
6287 . On the _remote_ system (which can also be the target system),
6288 start an LTTng <<lttng-relayd,relay daemon>> (man:lttng-relayd(8)):
6297 . On the _target_ system, create a tracing session configured to
6298 send trace data over the network:
6303 $ lttng create my-session --set-url=net://remote-system
6307 Replace `remote-system` by the host name or IP address of the
6308 remote system. See man:lttng-create(1) for the exact URL format.
6310 . On the target system, use the man:lttng(1) command-line tool as usual.
6311 When tracing is active, the target's consumer daemon sends sub-buffers
6312 to the relay daemon running on the remote system instead of flushing
6313 them to the local file system. The relay daemon writes the received
6314 packets to the local file system.
6316 The relay daemon writes trace files to
6317 +$LTTNG_HOME/lttng-traces/__hostname__/__session__+ by default, where
6318 +__hostname__+ is the host name of the target system and +__session__+
6319 is the tracing session name. Note that the env:LTTNG_HOME environment
6320 variable defaults to `$HOME` if not set. Use the
6321 opt:lttng-relayd(8):--output option of man:lttng-relayd(8) to write
6322 trace files to another base directory.
6327 === View events as LTTng emits them (noch:{LTTng} live)
6329 LTTng live is a network protocol implemented by the <<lttng-relayd,relay
6330 daemon>> (man:lttng-relayd(8)) to allow compatible trace viewers to
6331 display events as LTTng emits them on the target system while tracing is
6334 The relay daemon creates a _tee_: it forwards the trace data to both
6335 the local file system and to connected live viewers:
6338 .The relay daemon creates a _tee_, forwarding the trace data to both trace files and a connected live viewer.
6343 . On the _target system_, create a <<tracing-session,tracing session>>
6349 $ lttng create my-session --live
6353 This spawns a local relay daemon.
6355 . Start the live viewer and configure it to connect to the relay
6356 daemon. For example, with http://diamon.org/babeltrace[Babeltrace]:
6361 $ babeltrace --input-format=lttng-live \
6362 net://localhost/host/hostname/my-session
6369 * `hostname` with the host name of the target system.
6370 * `my-session` with the name of the tracing session to view.
6373 . Configure the tracing session as usual with the man:lttng(1)
6374 command-line tool, and <<basic-tracing-session-control,start tracing>>.
6376 You can list the available live tracing sessions with Babeltrace:
6380 $ babeltrace --input-format=lttng-live net://localhost
6383 You can start the relay daemon on another system. In this case, you need
6384 to specify the relay daemon's URL when you create the tracing session
6385 with the opt:lttng-create(1):--set-url option. You also need to replace
6386 `localhost` in the procedure above with the host name of the system on
6387 which the relay daemon is running.
6389 See man:lttng-create(1) and man:lttng-relayd(8) for the complete list of
6390 command-line options.
6394 [[taking-a-snapshot]]
6395 === Take a snapshot of the current sub-buffers of a tracing session
6397 The normal behavior of LTTng is to append full sub-buffers to growing
6398 trace data files. This is ideal to keep a full history of the events
6399 that occurred on the target system, but it can
6400 represent too much data in some situations. For example, you may wish
6401 to trace your application continuously until some critical situation
6402 happens, in which case you only need the latest few recorded
6403 events to perform the desired analysis, not multi-gigabyte trace files.
6405 With the man:lttng-snapshot(1) command, you can take a snapshot of the
6406 current sub-buffers of a given <<tracing-session,tracing session>>.
6407 LTTng can write the snapshot to the local file system or send it over
6412 . Create a tracing session in _snapshot mode_:
6417 $ lttng create my-session --snapshot
6421 The <<channel-overwrite-mode-vs-discard-mode,event loss mode>> of
6422 <<channel,channels>> created in this mode is automatically set to
6423 _overwrite_ (flight recorder mode).
6425 . Configure the tracing session as usual with the man:lttng(1)
6426 command-line tool, and <<basic-tracing-session-control,start tracing>>.
6428 . **Optional**: When you need to take a snapshot,
6429 <<basic-tracing-session-control,stop tracing>>.
6431 You can take a snapshot when the tracers are active, but if you stop
6432 them first, you are sure that the data in the sub-buffers does not
6433 change before you actually take the snapshot.
6440 $ lttng snapshot record --name=my-first-snapshot
6444 LTTng writes the current sub-buffers of all the current tracing
6445 session's channels to trace files on the local file system. Those trace
6446 files have `my-first-snapshot` in their name.
6448 There is no difference between the format of a normal trace file and the
6449 format of a snapshot: viewers of LTTng traces also support LTTng
6452 By default, LTTng writes snapshot files to the path shown by
6453 `lttng snapshot list-output`. You can change this path or decide to send
6454 snapshots over the network using either:
6456 . An output path or URL that you specify when you create the
6458 . An snapshot output path or URL that you add using
6459 `lttng snapshot add-output`
6460 . An output path or URL that you provide directly to the
6461 `lttng snapshot record` command.
6463 Method 3 overrides method 2, which overrides method 1. When you
6464 specify a URL, a relay daemon must listen on a remote system (see
6465 <<sending-trace-data-over-the-network,Send trace data over the network>>).
6470 === Use the machine interface
6472 With any command of the man:lttng(1) command-line tool, you can set the
6473 opt:lttng(1):--mi option to `xml` (before the command name) to get an
6474 XML machine interface output, for example:
6478 $ lttng --mi=xml enable-event --kernel --syscall open
6481 A schema definition (XSD) is
6482 https://github.com/lttng/lttng-tools/blob/stable-2.10/src/common/mi-lttng-3.0.xsd[available]
6483 to ease the integration with external tools as much as possible.
6487 [[metadata-regenerate]]
6488 === Regenerate the metadata of an LTTng trace
6490 An LTTng trace, which is a http://diamon.org/ctf[CTF] trace, has both
6491 data stream files and a metadata file. This metadata file contains,
6492 amongst other things, information about the offset of the clock sources
6493 used to timestamp <<event,event records>> when tracing.
6495 If, once a <<tracing-session,tracing session>> is
6496 <<basic-tracing-session-control,started>>, a major
6497 https://en.wikipedia.org/wiki/Network_Time_Protocol[NTP] correction
6498 happens, the trace's clock offset also needs to be updated. You
6499 can use the `metadata` item of the man:lttng-regenerate(1) command
6502 The main use case of this command is to allow a system to boot with
6503 an incorrect wall time and trace it with LTTng before its wall time
6504 is corrected. Once the system is known to be in a state where its
6505 wall time is correct, it can run `lttng regenerate metadata`.
6507 To regenerate the metadata of an LTTng trace:
6509 * Use the `metadata` item of the man:lttng-regenerate(1) command:
6514 $ lttng regenerate metadata
6520 `lttng regenerate metadata` has the following limitations:
6522 * Tracing session <<creating-destroying-tracing-sessions,created>>
6524 * User space <<channel,channels>>, if any, are using
6525 <<channel-buffering-schemes,per-user buffering>>.
6530 [[regenerate-statedump]]
6531 === Regenerate the state dump of a tracing session
6533 The LTTng kernel and user space tracers generate state dump
6534 <<event,event records>> when the application starts or when you
6535 <<basic-tracing-session-control,start a tracing session>>. An analysis
6536 can use the state dump event records to set an initial state before it
6537 builds the rest of the state from the following event records.
6538 http://tracecompass.org/[Trace Compass] is a notable example of an
6539 application which uses the state dump of an LTTng trace.
6541 When you <<taking-a-snapshot,take a snapshot>>, it's possible that the
6542 state dump event records are not included in the snapshot because they
6543 were recorded to a sub-buffer that has been consumed or overwritten
6546 You can use the `lttng regenerate statedump` command to emit the state
6547 dump event records again.
6549 To regenerate the state dump of the current tracing session, provided
6550 create it in snapshot mode, before you take a snapshot:
6552 . Use the `statedump` item of the man:lttng-regenerate(1) command:
6557 $ lttng regenerate statedump
6561 . <<basic-tracing-session-control,Stop the tracing session>>:
6570 . <<taking-a-snapshot,Take a snapshot>>:
6575 $ lttng snapshot record --name=my-snapshot
6579 Depending on the event throughput, you should run steps 1 and 2
6580 as closely as possible.
6582 NOTE: To record the state dump events, you need to
6583 <<enabling-disabling-events,create event rules>> which enable them.
6584 LTTng-UST state dump tracepoints start with `lttng_ust_statedump:`.
6585 LTTng-modules state dump tracepoints start with `lttng_statedump_`.
6589 [[persistent-memory-file-systems]]
6590 === Record trace data on persistent memory file systems
6592 https://en.wikipedia.org/wiki/Non-volatile_random-access_memory[Non-volatile random-access memory]
6593 (NVRAM) is random-access memory that retains its information when power
6594 is turned off (non-volatile). Systems with such memory can store data
6595 structures in RAM and retrieve them after a reboot, without flushing
6596 to typical _storage_.
6598 Linux supports NVRAM file systems thanks to either
6599 http://pramfs.sourceforge.net/[PRAMFS] or
6600 https://www.kernel.org/doc/Documentation/filesystems/dax.txt[DAX]{nbsp}+{nbsp}http://lkml.iu.edu/hypermail/linux/kernel/1504.1/03463.html[pmem]
6601 (requires Linux 4.1+).
6603 This section does not describe how to operate such file systems;
6604 we assume that you have a working persistent memory file system.
6606 When you create a <<tracing-session,tracing session>>, you can specify
6607 the path of the shared memory holding the sub-buffers. If you specify a
6608 location on an NVRAM file system, then you can retrieve the latest
6609 recorded trace data when the system reboots after a crash.
6611 To record trace data on a persistent memory file system and retrieve the
6612 trace data after a system crash:
6614 . Create a tracing session with a sub-buffer shared memory path located
6615 on an NVRAM file system:
6620 $ lttng create my-session --shm-path=/path/to/shm
6624 . Configure the tracing session as usual with the man:lttng(1)
6625 command-line tool, and <<basic-tracing-session-control,start tracing>>.
6627 . After a system crash, use the man:lttng-crash(1) command-line tool to
6628 view the trace data recorded on the NVRAM file system:
6633 $ lttng-crash /path/to/shm
6637 The binary layout of the ring buffer files is not exactly the same as
6638 the trace files layout. This is why you need to use man:lttng-crash(1)
6639 instead of your preferred trace viewer directly.
6641 To convert the ring buffer files to LTTng trace files:
6643 * Use the opt:lttng-crash(1):--extract option of man:lttng-crash(1):
6648 $ lttng-crash --extract=/path/to/trace /path/to/shm
6654 [[notif-trigger-api]]
6655 === Get notified when a channel's buffer usage is too high or too low
6657 With LTTng's $$C/C++$$ notification and trigger API, your user
6658 application can get notified when the buffer usage of one or more
6659 <<channel,channels>> becomes too low or too high. You can use this API
6660 and enable or disable <<event,event rules>> during tracing to avoid
6661 <<channel-overwrite-mode-vs-discard-mode,discarded event records>>.
6663 .Have a user application get notified when an LTTng channel's buffer usage is too high.
6665 In this example, we create and build an application which gets notified
6666 when the buffer usage of a specific LTTng channel is higher than
6667 75{nbsp}%. We only print that it is the case in the example, but we
6668 could as well use the API of <<liblttng-ctl-lttng,`liblttng-ctl`>> to
6669 disable event rules when this happens.
6671 . Create the application's C source file:
6679 #include <lttng/domain.h>
6680 #include <lttng/action/action.h>
6681 #include <lttng/action/notify.h>
6682 #include <lttng/condition/condition.h>
6683 #include <lttng/condition/buffer-usage.h>
6684 #include <lttng/condition/evaluation.h>
6685 #include <lttng/notification/channel.h>
6686 #include <lttng/notification/notification.h>
6687 #include <lttng/trigger/trigger.h>
6688 #include <lttng/endpoint.h>
6690 int main(int argc, char *argv[])
6692 int exit_status = 0;
6693 struct lttng_notification_channel *notification_channel;
6694 struct lttng_condition *condition;
6695 struct lttng_action *action;
6696 struct lttng_trigger *trigger;
6697 const char *tracing_session_name;
6698 const char *channel_name;
6701 tracing_session_name = argv[1];
6702 channel_name = argv[2];
6705 * Create a notification channel. A notification channel
6706 * connects the user application to the LTTng session daemon.
6707 * This notification channel can be used to listen to various
6708 * types of notifications.
6710 notification_channel = lttng_notification_channel_create(
6711 lttng_session_daemon_notification_endpoint);
6714 * Create a "high buffer usage" condition. In this case, the
6715 * condition is reached when the buffer usage is greater than or
6716 * equal to 75 %. We create the condition for a specific tracing
6717 * session name, channel name, and for the user space tracing
6720 * The "low buffer usage" condition type also exists.
6722 condition = lttng_condition_buffer_usage_high_create();
6723 lttng_condition_buffer_usage_set_threshold_ratio(condition, .75);
6724 lttng_condition_buffer_usage_set_session_name(
6725 condition, tracing_session_name);
6726 lttng_condition_buffer_usage_set_channel_name(condition,
6728 lttng_condition_buffer_usage_set_domain_type(condition,
6732 * Create an action (get a notification) to take when the
6733 * condition created above is reached.
6735 action = lttng_action_notify_create();
6738 * Create a trigger. A trigger associates a condition to an
6739 * action: the action is executed when the condition is reached.
6741 trigger = lttng_trigger_create(condition, action);
6743 /* Register the trigger to LTTng. */
6744 lttng_register_trigger(trigger);
6747 * Now that we have registered a trigger, a notification will be
6748 * emitted everytime its condition is met. To receive this
6749 * notification, we must subscribe to notifications that match
6750 * the same condition.
6752 lttng_notification_channel_subscribe(notification_channel,
6756 * Notification loop. You can put this in a dedicated thread to
6757 * avoid blocking the main thread.
6760 struct lttng_notification *notification;
6761 enum lttng_notification_channel_status status;
6762 const struct lttng_evaluation *notification_evaluation;
6763 const struct lttng_condition *notification_condition;
6764 double buffer_usage;
6766 /* Receive the next notification. */
6767 status = lttng_notification_channel_get_next_notification(
6768 notification_channel, ¬ification);
6771 case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
6773 case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
6775 * The session daemon can drop notifications if
6776 * a monitoring application is not consuming the
6777 * notifications fast enough.
6780 case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
6782 * The notification channel has been closed by the
6783 * session daemon. This is typically caused by a session
6784 * daemon shutting down.
6788 /* Unhandled conditions or errors. */
6794 * A notification provides, amongst other things:
6796 * * The condition that caused this notification to be
6798 * * The condition evaluation, which provides more
6799 * specific information on the evaluation of the
6802 * The condition evaluation provides the buffer usage
6803 * value at the moment the condition was reached.
6805 notification_condition = lttng_notification_get_condition(
6807 notification_evaluation = lttng_notification_get_evaluation(
6810 /* We're subscribed to only one condition. */
6811 assert(lttng_condition_get_type(notification_condition) ==
6812 LTTNG_CONDITION_TYPE_BUFFER_USAGE_HIGH);
6815 * Get the exact sampled buffer usage from the
6816 * condition evaluation.
6818 lttng_evaluation_buffer_usage_get_usage_ratio(
6819 notification_evaluation, &buffer_usage);
6822 * At this point, instead of printing a message, we
6823 * could do something to reduce the channel's buffer
6824 * usage, like disable specific events.
6826 printf("Buffer usage is %f %% in tracing session \"%s\", "
6827 "user space channel \"%s\".\n", buffer_usage * 100,
6828 tracing_session_name, channel_name);
6829 lttng_notification_destroy(notification);
6833 lttng_action_destroy(action);
6834 lttng_condition_destroy(condition);
6835 lttng_trigger_destroy(trigger);
6836 lttng_notification_channel_destroy(notification_channel);
6842 . Build the `notif-app` application, linking it to `liblttng-ctl`:
6847 $ gcc -o notif-app notif-app.c -llttng-ctl
6851 . <<creating-destroying-tracing-sessions,Create a tracing session>>,
6852 <<enabling-disabling-events,create an event rule>> matching all the
6853 user space tracepoints, and
6854 <<basic-tracing-session-control,start tracing>>:
6859 $ lttng create my-session
6860 $ lttng enable-event --userspace --all
6865 If you create the channel manually with the man:lttng-enable-channel(1)
6866 command, you can control how frequently are the current values of the
6867 channel's properties sampled to evaluate user conditions with the
6868 opt:lttng-enable-channel(1):--monitor-timer option.
6870 . Run the `notif-app` application. This program accepts the
6871 <<tracing-session,tracing session>> name and the user space channel
6872 name as its two first arguments. The channel which LTTng automatically
6873 creates with the man:lttng-enable-event(1) command above is named
6879 $ ./notif-app my-session channel0
6883 . In another terminal, run an application with a very high event
6884 throughput so that the 75{nbsp}% buffer usage condition is reached.
6886 In the first terminal, the application should print lines like this:
6889 Buffer usage is 81.45197 % in tracing session "my-session", user space
6893 If you don't see anything, try modifying the condition in
6894 path:{notif-app.c} to a lower value (0.1, for example), rebuilding it
6895 (step 2) and running it again (step 4).
6902 [[lttng-modules-ref]]
6903 === noch:{LTTng-modules}
6907 [[lttng-tracepoint-enum]]
6908 ==== `LTTNG_TRACEPOINT_ENUM()` usage
6910 Use the `LTTNG_TRACEPOINT_ENUM()` macro to define an enumeration:
6914 LTTNG_TRACEPOINT_ENUM(name, TP_ENUM_VALUES(entries))
6919 * `name` with the name of the enumeration (C identifier, unique
6920 amongst all the defined enumerations).
6921 * `entries` with a list of enumeration entries.
6923 The available enumeration entry macros are:
6925 +ctf_enum_value(__name__, __value__)+::
6926 Entry named +__name__+ mapped to the integral value +__value__+.
6928 +ctf_enum_range(__name__, __begin__, __end__)+::
6929 Entry named +__name__+ mapped to the range of integral values between
6930 +__begin__+ (included) and +__end__+ (included).
6932 +ctf_enum_auto(__name__)+::
6933 Entry named +__name__+ mapped to the integral value following the
6934 last mapping's value.
6936 The last value of a `ctf_enum_value()` entry is its +__value__+
6939 The last value of a `ctf_enum_range()` entry is its +__end__+ parameter.
6941 If `ctf_enum_auto()` is the first entry in the list, its integral
6944 Use the `ctf_enum()` <<lttng-modules-tp-fields,field definition macro>>
6945 to use a defined enumeration as a tracepoint field.
6947 .Define an enumeration with `LTTNG_TRACEPOINT_ENUM()`.
6951 LTTNG_TRACEPOINT_ENUM(
6954 ctf_enum_auto("AUTO: EXPECT 0")
6955 ctf_enum_value("VALUE: 23", 23)
6956 ctf_enum_value("VALUE: 27", 27)
6957 ctf_enum_auto("AUTO: EXPECT 28")
6958 ctf_enum_range("RANGE: 101 TO 303", 101, 303)
6959 ctf_enum_auto("AUTO: EXPECT 304")
6967 [[lttng-modules-tp-fields]]
6968 ==== Tracepoint fields macros (for `TP_FIELDS()`)
6970 [[tp-fast-assign]][[tp-struct-entry]]The available macros to define
6971 tracepoint fields, which must be listed within `TP_FIELDS()` in
6972 `LTTNG_TRACEPOINT_EVENT()`, are:
6974 [role="func-desc growable",cols="asciidoc,asciidoc"]
6975 .Available macros to define LTTng-modules tracepoint fields
6977 |Macro |Description and parameters
6980 +ctf_integer(__t__, __n__, __e__)+
6982 +ctf_integer_nowrite(__t__, __n__, __e__)+
6984 +ctf_user_integer(__t__, __n__, __e__)+
6986 +ctf_user_integer_nowrite(__t__, __n__, __e__)+
6988 Standard integer, displayed in base 10.
6991 Integer C type (`int`, `long`, `size_t`, ...).
6997 Argument expression.
7000 +ctf_integer_hex(__t__, __n__, __e__)+
7002 +ctf_user_integer_hex(__t__, __n__, __e__)+
7004 Standard integer, displayed in base 16.
7013 Argument expression.
7015 |+ctf_integer_oct(__t__, __n__, __e__)+
7017 Standard integer, displayed in base 8.
7026 Argument expression.
7029 +ctf_integer_network(__t__, __n__, __e__)+
7031 +ctf_user_integer_network(__t__, __n__, __e__)+
7033 Integer in network byte order (big-endian), displayed in base 10.
7042 Argument expression.
7045 +ctf_integer_network_hex(__t__, __n__, __e__)+
7047 +ctf_user_integer_network_hex(__t__, __n__, __e__)+
7049 Integer in network byte order, displayed in base 16.
7058 Argument expression.
7061 +ctf_enum(__N__, __t__, __n__, __e__)+
7063 +ctf_enum_nowrite(__N__, __t__, __n__, __e__)+
7065 +ctf_user_enum(__N__, __t__, __n__, __e__)+
7067 +ctf_user_enum_nowrite(__N__, __t__, __n__, __e__)+
7072 Name of a <<lttng-tracepoint-enum,previously defined enumeration>>.
7075 Integer C type (`int`, `long`, `size_t`, ...).
7081 Argument expression.
7084 +ctf_string(__n__, __e__)+
7086 +ctf_string_nowrite(__n__, __e__)+
7088 +ctf_user_string(__n__, __e__)+
7090 +ctf_user_string_nowrite(__n__, __e__)+
7092 Null-terminated string; undefined behavior if +__e__+ is `NULL`.
7098 Argument expression.
7101 +ctf_array(__t__, __n__, __e__, __s__)+
7103 +ctf_array_nowrite(__t__, __n__, __e__, __s__)+
7105 +ctf_user_array(__t__, __n__, __e__, __s__)+
7107 +ctf_user_array_nowrite(__t__, __n__, __e__, __s__)+
7109 Statically-sized array of integers.
7112 Array element C type.
7118 Argument expression.
7124 +ctf_array_bitfield(__t__, __n__, __e__, __s__)+
7126 +ctf_array_bitfield_nowrite(__t__, __n__, __e__, __s__)+
7128 +ctf_user_array_bitfield(__t__, __n__, __e__, __s__)+
7130 +ctf_user_array_bitfield_nowrite(__t__, __n__, __e__, __s__)+
7132 Statically-sized array of bits.
7134 The type of +__e__+ must be an integer type. +__s__+ is the number
7135 of elements of such type in +__e__+, not the number of bits.
7138 Array element C type.
7144 Argument expression.
7150 +ctf_array_text(__t__, __n__, __e__, __s__)+
7152 +ctf_array_text_nowrite(__t__, __n__, __e__, __s__)+
7154 +ctf_user_array_text(__t__, __n__, __e__, __s__)+
7156 +ctf_user_array_text_nowrite(__t__, __n__, __e__, __s__)+
7158 Statically-sized array, printed as text.
7160 The string does not need to be null-terminated.
7163 Array element C type (always `char`).
7169 Argument expression.
7175 +ctf_sequence(__t__, __n__, __e__, __T__, __E__)+
7177 +ctf_sequence_nowrite(__t__, __n__, __e__, __T__, __E__)+
7179 +ctf_user_sequence(__t__, __n__, __e__, __T__, __E__)+
7181 +ctf_user_sequence_nowrite(__t__, __n__, __e__, __T__, __E__)+
7183 Dynamically-sized array of integers.
7185 The type of +__E__+ must be unsigned.
7188 Array element C type.
7194 Argument expression.
7197 Length expression C type.
7203 +ctf_sequence_hex(__t__, __n__, __e__, __T__, __E__)+
7205 +ctf_user_sequence_hex(__t__, __n__, __e__, __T__, __E__)+
7207 Dynamically-sized array of integers, displayed in base 16.
7209 The type of +__E__+ must be unsigned.
7212 Array element C type.
7218 Argument expression.
7221 Length expression C type.
7226 |+ctf_sequence_network(__t__, __n__, __e__, __T__, __E__)+
7228 Dynamically-sized array of integers in network byte order (big-endian),
7229 displayed in base 10.
7231 The type of +__E__+ must be unsigned.
7234 Array element C type.
7240 Argument expression.
7243 Length expression C type.
7249 +ctf_sequence_bitfield(__t__, __n__, __e__, __T__, __E__)+
7251 +ctf_sequence_bitfield_nowrite(__t__, __n__, __e__, __T__, __E__)+
7253 +ctf_user_sequence_bitfield(__t__, __n__, __e__, __T__, __E__)+
7255 +ctf_user_sequence_bitfield_nowrite(__t__, __n__, __e__, __T__, __E__)+
7257 Dynamically-sized array of bits.
7259 The type of +__e__+ must be an integer type. +__s__+ is the number
7260 of elements of such type in +__e__+, not the number of bits.
7262 The type of +__E__+ must be unsigned.
7265 Array element C type.
7271 Argument expression.
7274 Length expression C type.
7280 +ctf_sequence_text(__t__, __n__, __e__, __T__, __E__)+
7282 +ctf_sequence_text_nowrite(__t__, __n__, __e__, __T__, __E__)+
7284 +ctf_user_sequence_text(__t__, __n__, __e__, __T__, __E__)+
7286 +ctf_user_sequence_text_nowrite(__t__, __n__, __e__, __T__, __E__)+
7288 Dynamically-sized array, displayed as text.
7290 The string does not need to be null-terminated.
7292 The type of +__E__+ must be unsigned.
7294 The behaviour is undefined if +__e__+ is `NULL`.
7297 Sequence element C type (always `char`).
7303 Argument expression.
7306 Length expression C type.
7312 Use the `_user` versions when the argument expression, `e`, is
7313 a user space address. In the cases of `ctf_user_integer*()` and
7314 `ctf_user_float*()`, `&e` must be a user space address, thus `e` must
7317 The `_nowrite` versions omit themselves from the session trace, but are
7318 otherwise identical. This means the `_nowrite` fields won't be written
7319 in the recorded trace. Their primary purpose is to make some
7320 of the event context available to the
7321 <<enabling-disabling-events,event filters>> without having to
7322 commit the data to sub-buffers.
7328 Terms related to LTTng and to tracing in general:
7331 The http://diamon.org/babeltrace[Babeltrace] project, which includes
7332 the cmd:babeltrace command, some libraries, and Python bindings.
7334 <<channel-buffering-schemes,buffering scheme>>::
7335 A layout of sub-buffers applied to a given channel.
7337 <<channel,channel>>::
7338 An entity which is responsible for a set of ring buffers.
7340 <<event,Event rules>> are always attached to a specific channel.
7343 A reference of time for a tracer.
7345 <<lttng-consumerd,consumer daemon>>::
7346 A process which is responsible for consuming the full sub-buffers
7347 and write them to a file system or send them over the network.
7349 <<channel-overwrite-mode-vs-discard-mode,discard mode>>:: The event loss
7350 mode in which the tracer _discards_ new event records when there's no
7351 sub-buffer space left to store them.
7354 The consequence of the execution of an instrumentation
7355 point, like a tracepoint that you manually place in some source code,
7356 or a Linux kernel KProbe.
7358 An event is said to _occur_ at a specific time. Different actions can
7359 be taken upon the occurrence of an event, like record the event's payload
7362 <<channel-overwrite-mode-vs-discard-mode,event loss mode>>::
7363 The mechanism by which event records of a given channel are lost
7364 (not recorded) when there is no sub-buffer space left to store them.
7366 [[def-event-name]]event name::
7367 The name of an event, which is also the name of the event record.
7368 This is also called the _instrumentation point name_.
7371 A record, in a trace, of the payload of an event which occured.
7373 <<event,event rule>>::
7374 Set of conditions which must be satisfied for one or more occuring
7375 events to be recorded.
7377 `java.util.logging`::
7379 https://docs.oracle.com/javase/7/docs/api/java/util/logging/package-summary.html[core logging facilities].
7381 <<instrumenting,instrumentation>>::
7382 The use of LTTng probes to make a piece of software traceable.
7384 instrumentation point::
7385 A point in the execution path of a piece of software that, when
7386 reached by this execution, can emit an event.
7388 instrumentation point name::
7389 See _<<def-event-name,event name>>_.
7392 A http://logging.apache.org/log4j/1.2/[logging library] for Java
7393 developed by the Apache Software Foundation.
7396 Level of severity of a log statement or user space
7397 instrumentation point.
7400 The _Linux Trace Toolkit: next generation_ project.
7402 <<lttng-cli,cmd:lttng>>::
7403 A command-line tool provided by the LTTng-tools project which you
7404 can use to send and receive control messages to and from a
7408 The https://github.com/lttng/lttng-analyses[LTTng analyses] project,
7409 which is a set of analyzing programs that are used to obtain a
7410 higher level view of an LTTng trace.
7412 cmd:lttng-consumerd::
7413 The name of the consumer daemon program.
7416 A utility provided by the LTTng-tools project which can convert
7417 ring buffer files (usually
7418 <<persistent-memory-file-systems,saved on a persistent memory file system>>)
7421 LTTng Documentation::
7424 <<lttng-live,LTTng live>>::
7425 A communication protocol between the relay daemon and live viewers
7426 which makes it possible to see events "live", as they are received by
7429 <<lttng-modules,LTTng-modules>>::
7430 The https://github.com/lttng/lttng-modules[LTTng-modules] project,
7431 which contains the Linux kernel modules to make the Linux kernel
7432 instrumentation points available for LTTng tracing.
7435 The name of the relay daemon program.
7437 cmd:lttng-sessiond::
7438 The name of the session daemon program.
7441 The https://github.com/lttng/lttng-tools[LTTng-tools] project, which
7442 contains the various programs and libraries used to
7443 <<controlling-tracing,control tracing>>.
7445 <<lttng-ust,LTTng-UST>>::
7446 The https://github.com/lttng/lttng-ust[LTTng-UST] project, which
7447 contains libraries to instrument user applications.
7449 <<lttng-ust-agents,LTTng-UST Java agent>>::
7450 A Java package provided by the LTTng-UST project to allow the
7451 LTTng instrumentation of `java.util.logging` and Apache log4j 1.2
7454 <<lttng-ust-agents,LTTng-UST Python agent>>::
7455 A Python package provided by the LTTng-UST project to allow the
7456 LTTng instrumentation of Python logging statements.
7458 <<channel-overwrite-mode-vs-discard-mode,overwrite mode>>::
7459 The event loss mode in which new event records overwrite older
7460 event records when there's no sub-buffer space left to store them.
7462 <<channel-buffering-schemes,per-process buffering>>::
7463 A buffering scheme in which each instrumented process has its own
7464 sub-buffers for a given user space channel.
7466 <<channel-buffering-schemes,per-user buffering>>::
7467 A buffering scheme in which all the processes of a Unix user share the
7468 same sub-buffer for a given user space channel.
7470 <<lttng-relayd,relay daemon>>::
7471 A process which is responsible for receiving the trace data sent by
7472 a distant consumer daemon.
7475 A set of sub-buffers.
7477 <<lttng-sessiond,session daemon>>::
7478 A process which receives control commands from you and orchestrates
7479 the tracers and various LTTng daemons.
7481 <<taking-a-snapshot,snapshot>>::
7482 A copy of the current data of all the sub-buffers of a given tracing
7483 session, saved as trace files.
7486 One part of an LTTng ring buffer which contains event records.
7489 The time information attached to an event when it is emitted.
7492 A set of files which are the concatenations of one or more
7493 flushed sub-buffers.
7496 The action of recording the events emitted by an application
7497 or by a system, or to initiate such recording by controlling
7501 The http://tracecompass.org[Trace Compass] project and application.
7504 An instrumentation point using the tracepoint mechanism of the Linux
7505 kernel or of LTTng-UST.
7507 tracepoint definition::
7508 The definition of a single tracepoint.
7511 The name of a tracepoint.
7513 tracepoint provider::
7514 A set of functions providing tracepoints to an instrumented user
7517 Not to be confused with a _tracepoint provider package_: many tracepoint
7518 providers can exist within a tracepoint provider package.
7520 tracepoint provider package::
7521 One or more tracepoint providers compiled as an object file or as
7525 A software which records emitted events.
7527 <<domain,tracing domain>>::
7528 A namespace for event sources.
7530 <<tracing-group,tracing group>>::
7531 The Unix group in which a Unix user can be to be allowed to trace the
7534 <<tracing-session,tracing session>>::
7535 A stateful dialogue between you and a <<lttng-sessiond,session
7539 An application running in user space, as opposed to a Linux kernel
7540 module, for example.