Re: [dpdk-dev] [PATCH v3 33/33] doc: add trace library guide

[Jerin Jacob <jerinjacobk@gmail.com>]

On Wed, Apr 1, 2020 at 8:27 PM Thomas Monjalon <thomas@monjalon.net> wrote:
>
> Hi,
>
> I did not review the code.
> I want to understand the design first,

I don't intend to write this as a low-level design document.
I would suggest please look into API documentation in the first trace patch.
http://patches.dpdk.org/patch/67322/


> and I start with the user documentation.

OK.

>
> John, Marko, who is going to review this new documentation?
> I think we need some help for syntax and phrasing here. Thanks
>
>
> > +Trace Library
> > +=============
> > +
> > +DPDK provides a tracing library that gives the ability to add tracepoints
> > +in application to get runtime trace/debug information for control and fast
> > +APIs with minimum impact on fast path performance. Typical trace overhead is
> > +~20 cycles and instrumentation overhead is 1 cycle.
>
> Please could you start this doc with explaining what is a tracepoint?

OK. I will pull of the what is tracepoint from the wiki.

>
> How is it different of a log?

OK.

>
> My main question is to understand why it is not implemented as a CTF output
> of rte_log. Reminder we have rte_openlog_stream(FILE *).
> Is there a way to adapt/extend this API to specify using CTF formatter?
> What are the other gaps?

Log and trace quite different. Trace will have a directory, which contains
the trace files for all the cores, quite different from the log.

See rte_trace_save() API.

Another difference is how fast trace can be written. log is snprint()
style stuff
(pretty slow), Where trace deals with fixed primitives like rte_trace_ctf_*

See lttng library API to understand it better. We have similar API
like lttng( based
on community feedback)


>
> > +
> > +Library mainly caters below mentioned use cases:
> > +
> > +- The DPDK provider will not have access to the DPDK customer applications.
>
> There is no provider/customer.
> DPDK is available on dpdk.org and users are not paying.

I will reword.

>
> > +  Inbuilt tracer support will us enable to debug/analyze the slow path and
> > +  fast path DPDK API usage.
> > +
> > +- Provides a low overhead fast path multi-core PMD driver's debugging/analysis
> > +  infrastructure to fix the functional and performance issue(s).
> > +
> > +- Post trace analysis tools can provide various status across the system such
> > +  as cpu_idle() using the timestamp added in the trace.
> > +
> > +Below sections will provide detailed information about:
> > +
> > + - Trace a user application
> > + - View and analyze the recorded events
> > +
> > +Trace a user application
> > +------------------------
> > +
> > +This section steps you through a simple example to trace an application.
> > +A trace can be achieved using below mentioned steps:
> > +
> > +Define and register a tracepoint
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +The application can define and register tracepoints either existing C file or
> > +create a new file (say xyz_app_trace_point.c). Also, all the tracepoints must be
> > +resolved before rte_eal_init i.e. tracepoints must be registered as constructor
> > +using RTE_INIT interface.
>
> I think you want to explain the design before,
> so usage constraints will be better understood.

I think, first usage and then design. Not all care about internal
design.(In the context of DPDK programming guide)

>
> > +
> > +Following are the MACRO definition exposed by the trace Library to define and
> > +register a tracepoint.
> > +
> > +.. code-block:: c
> > +
> > + #define RTE_TRACE_POINT_DEFINE(tp)\
> > +        uint64_t __attribute__((section("__rte_trace_point"))) __##tp
> > +
> > + #define RTE_TRACE_POINT_REGISTER(trace, name, level)\
> > +       __rte_trace_point_register(&__##trace, RTE_STR(name), RTE_LOG_ ## level,\
> > +                                  (void (*)(void)) trace)
>
> Showing the macro definitions does not help to understand
> neither the concept of define and register, nor the usage.

Ack. I will remove.

>
> > +Example tracepoint definition and registration
> > +
> > +.. code-block:: c
> > +
> > + RTE_TRACE_POINT_DEFINE(rte_trace_lib_eal_generic_str); /* Definition */
>
> What is a definition for?
> Please explain a very simple use case by stating the goal and the output
> of the example clearly.

OK.


>
> > +
> > + RTE_INIT(eal_trace_init)
> > + {
> > +     /* Registration */
> > +     RTE_TRACE_POINT_REGISTER(rte_trace_lib_eal_generic_str,
> > +                              lib.eal.generic.str, INFO);
> > + }
>
> Here we just see some function calls without understanding what it does.

I think, one needs to go through code for that. I am NOT adding
low-level design here.

>
> > +
> > +For more details refer trace API documentation.
>
> Yes the details must be in the API documentation.
> This guide should explain the design and goals.
>
> > +Defined tracepoint must be exported into corresponding .map file.
> > +
> > +.. Note::
> > +
> > +    A tracepoint is defined like __##tp i.e. __rte_trace_lib_eal_generic_str
> > +    for above example. Same must be updated into corresponding .map file.
>
> We don't know what is __##tp. That's so cryptic :-)

I will fix it.

>
> > +
> > +Define trace function to write events
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +After a successful tracepoint registration, the application must define a
> > +trace function which solves three purposes:
> > +
> > + - Calculates the size of the event.
> > + - Generate CTF metadata field string for the event.
>
> The format of the trace is the responsibility of the application?

No.

>
> > + - Emit the event to trace memory.
> > +
> > +A tracepoint can be classified as either a data path or a slow path tracepoint.
> > +So based on that, the application must define tracepoint function using one of
> > +the mentioned MACRO
> > +
> > +.. code-block:: c
> > +
> > + /* Define tracepoint function for slow path */
> > + #define RTE_TRACE_POINT(tp, args, ...)\
> > +        __RTE_TRACE_POINT(generic, tp, args, __VA_ARGS__)
> > +
> > + /* Define tracepoint function for data path */
> > + #define RTE_TRACE_POINT_DP(tp, args, ...)\
> > +        __RTE_TRACE_POINT(dp, tp, args, __VA_ARGS__)
>
> No need to copy/paste the code (which will be outdated at some point).
> I think naming the macros is enough.

OK

>
> > +RTE_TRACE_POINT_DP is compiled out by default and can be enabled using
> > +CONFIG_RTE_ENABLE_TRACE_DP configuration parameter. Also application can use
> > +``rte_trace_is_dp_enabled`` to get current status of RTE_TRACE_POINT_DP.
> > +For more details, refer DPDK Trace API documentation.
> > +
> > +Example tracepoint function definition
> > +
> > +.. code-block:: c
> > +
> > + /* Slow path tracepoint */
> > + RTE_TRACE_POINT(
> > +        rte_trace_lib_eal_generic_str,
> > +        RTE_TRACE_POINT_ARGS(const char *str),
> > +        rte_trace_ctf_string(str);
> > + )
> > +
> > + /* Data path tracepoint */
> > + RTE_TRACE_POINT_DP(
> > +        rte_trace_lib_eal_generic_str,
> > +        RTE_TRACE_POINT_ARGS(const char *str),
> > +        rte_trace_ctf_string(str);
> > + )
> > +
> > +Emit events to trace memory
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +After trace function definition is ready to emit tracepoints.
> > +To emit the event application needs to invoke tracepoint function, as defined
> > +in the above steps, at the desired location.
> > +
> > +Below examples emit tracepoints in ``rte_eth_dev_configure`` to print a test
> > +string:
> > +
> > +.. code-block:: c
> > +
> > + int
> > + rte_eth_dev_configure(uint16_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
> > +        const struct rte_eth_conf *dev_conf)
> > + {
> > +        struct rte_eth_dev *dev;
> > +        struct rte_eth_dev_info dev_info;
> > +        struct rte_eth_conf orig_conf;
> > +        int diag;
> > +        int ret;
> > +
> > +        RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -EINVAL);
> > +
> > +        dev = &rte_eth_devices[port_id];
> > +
> > +        RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->dev_configure, -ENOTSUP);
> > +
> > +        ...
> > +
> > +        rte_trace_lib_eal_generic_str("tp_test_string");
> > +        return ret;
> > + }
>
> Please do not copy real code. Hello-world style is enough.
> We do not want to be disturbed by the context of the code in such guide.

OK. Ack.

>
> > +
> > +Generate CTF formatted metadata
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +As of now emitted events just specify the debug information written by the
> > +application but to view/analyze these events must be formatted into Common Trace
> > +Format(CTF) so that any CTF compliant trace analysis tool can view those traces.
> > +
> > +Trace library exposes below API to write events to CTF formatted metadata file.
> > +
> > +.. code-block:: c
> > +
> > + int rte_trace_save(void);
> > +
> > +Currently library invokes this API implicitly during tear down and metadata file
> > +is generated at either ``/root/dpdk-traces/rte-yyyy-mm-dd-[AP]M-hh-mm-ss/`` or
> > +at location if user has passed during command line(``say /tmp``) then
> > +``/tmp/rte-yyyy-mm-dd-[AP]M-hh-mm-ss/``
>
> The trace is written in a file when the application is closing?

rte_trace_save() invoked in rte_eal_cleanup()

> How is it working if application crashes?

Segfault should call rte_eal_cleanup() and rte_trace_save() will be
called internally.


> How can we debug the application in real-time?

The application can call rte_trace_save() any point in time to save
the trace buffers to the filesystem,

>
> > +
> > +For more information, refer :doc:`../linux_gsg/linux_eal_parameters` for trace.
> > +
> > +View and analyze the recorded events
> > +------------------------------------
> > +Once ``Trace a user application`` is completed, the user can view/inspect the
> > +recorded events.
> > +
> > +There are many tools you can use to read DPDK traces:
> > +
> > + - ``babeltrace`` is a command-line utility that converts trace formats; it
> > +   supports the format that DPDK trace library produces, CTF, as well as a
> > +   basic text output that can be grep ed. The babeltrace command is part of the
> > +   opensource ``Babeltrace`` project.
> > +
> > + - ``Trace Compass`` is a graphical user interface for viewing and analyzing any
> > +   type of logs or traces, including DPDK traces.
> > +
> > +.. Note::
> > +
> > +   This section assumes that the trace library saved the traces, it recorded
> > +   during the previous tutorials, to their specified location.
> > +
> > +
> > +Use the babeltrace command-line tool
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +The simplest way to list all the recorded events of a trace is to pass its path
> > +to babeltrace with no options::
> > +
> > +    babeltrace </path-to-trace-events/rte-yyyy-mm-dd-[AP]M-hh-mm-ss/>
> > +
> > +``babeltrace`` finds all traces recursively within the given path and prints all
> > +their events, merging them in chronological order.
> > +
> > +You can pipe the output of the babeltrace into a tool like grep(1) for further
> > +filtering. Below example grep the events for ``ethdev`` only::
> > +
> > +    babeltrace /tmp/my-dpdk-trace | grep ethdev
> > +
> > +You can pipe the output of babeltrace into a tool like wc(1) to count the
> > +recorded events. Below example count the number of ``ethdev`` events::
> > +
> > +    babeltrace /tmp/my-dpdk-trace | grep ethdev | wc --lines
> > +
> > +Use the tracecompass GUI tool
> > +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > +``Tracecompass`` is another tool to view/analyze the DPDK traces which gives
> > +a graphical view of events. Like ``babeltrace``, tracecompass also provides
> > +an interface to search for a particular event. To use ``tracecompass``, following are
> > +the minimum required steps:
> > +
> > + - Install ``tracecompass`` to the localhost. Variants are available for Linux,
> > +   Windows, and OS-X.
> > + - Launch ``tracecompass`` which will open a graphical window with trace
> > +   management interfaces.
> > + - Open a trace using ``File->Open Trace`` option and select metadata file
> > +   which is to be viewed/analyzed.
> > +
> > +For more details, refer `Trace Compass <https://www.eclipse.org/tracecompass/>`_
>
> I don't think we need to explain external tools here. They can evolve.
> Giving their names and URL is enough.

It takes time for someone to understand the tool usage model. So added
minimum information such as
"
 - Open a trace using ``File->Open Trace`` option and select metadata file
   which is to be viewed/analyzed.

"

> Maybe just with a short description (text vs graphical).
>
> > +
> > +Core Concepts
> > +-------------
>
> I think the concepts must be explained at the beginning.
>
> > +As DPDK trace library is designed to generate traces that uses Common Trace
> > +Format(CTF). CTF specification consist of following units to create a trace.
> > +
> > + - ``Stream`` Sequence of packets.
> > + - ``Packet`` Header and one or more events.
> > + - ``Event`` Header and payload.
> > +
> > +For detailed information, refer `Common Trace Format <https://diamon.org/ctf/>`_
> > +
> > +Channel and trace memory
> > +~~~~~~~~~~~~~~~~~~~~~~~~
> > +A channel is an object which is responsible for holding the trace memory.
> > +The trace library creates the trace memory per thread to enable the lock-less
> > +scheme to emit the event. When a DPDK tracer emits an event, it will be recorded
> > +to the trace buffers that associated with that thread.
> > +
> > +Event record mode
> > +~~~~~~~~~~~~~~~~~
> > +Event record mode is an attribute of trace buffers. Trace library exposes two
> > +modes:
> > +
> > + - ``Overwrite`` This mode enables trace buffers to wrap around when trace buffer memory is full.
> > + - ``Discard`` This mode enables trace buffers to discard when trace buffer memory is full.
> > +
> > +This mode can be enabled/disabled either using eal command line parameters or
> > +DPDK trace library API to configure the mode.
> > +Refer :doc:`../linux_gsg/linux_eal_parameters` and trace API documentation more
> > +details.
> > +
> > +Metadata
> > +~~~~~~~~
> > +Metadata defines the layout of event records so that trace analysis tool can
> > +read the streams and show into the relevant format.
>
> I don't understand metadata from the explanation above.
> Is it useful to understand when using this library?

Some understanding of CTF will be required to consume the output of tracer.
More details about CTF metadata is here:

https://diamon.org/ctf/

> What is the impact for the user?

The last section has the implementation details.

The format followed in general in the doc:
- API usage model
- Implementation details( in case some would like to contribute to the
trace subsystem)

>
> > +For more details, refer `Common Trace Format <https://diamon.org/ctf/>`_.
>
>
> Thanks for the work.
> Please let's improve the doc while discussing the design.

Sure.

>
>