Re: When the trace buffers are saved to disk?

[Thomas Monjalon <thomas@monjalon.net>]

29/11/2024 21:39, Adel Belkhiri:
> Thank you for the clarification, Thomas. Indeed, the documentation for the
> trace library is kind of limited. If you don't mind, I have another
> question: Would it be useful to have an API to register a callback (to save
> trace data) when the buffer is full?

I suppose yes, the problem being which thread is running file writing.

I leave it to the maintainers of the trace library.


> On Fri, Nov 29, 2024 at 6:44 AM Thomas Monjalon <thomas@monjalon.net> wrote:
> > 28/11/2024 20:17, Adel Belkhiri:
> > > Hi all,
> > >
> > > Recently, while tracing applications from the apps and examples
> > > directories, I became confused about when the trace buffer is written to
> > > disk. Is the trace data saved only when rte_save_trace() is called, or
> > does
> >
> > It is rte_trace_save()
> >
> > > it also automatically save when the buffer becomes full?
> >
> > No, DPDK is not doing such thing without user agreement.
> >
> > > From my understanding, rte_save_trace() is invoked when the application
> > > executes rte_eal_cleanup(). Does this mean the target application needs
> > to
> > > explicitly support tracing by calling rte_save_trace()—perhaps at regular
> > > intervals—to dump the trace buffer to disk? Otherwise, will we only get a
> > > fragment of the trace saved during rte_eal_cleanup() execution?
> >
> > Yes you get it right.
> >
> > > Thank you for clarifying this point.
> >
> > Thanks for asking.
> >
> > If you think the doc below is not clear enough,
> > do not hesitate to submit a patch to make the doc better:
> >
> >         https://doc.dpdk.org/guides/prog_guide/trace_lib.html