[lttng-dev] RFC Streaming and reading traces over the network

Stefan Hajnoczi stefanha at gmail.com
Fri Mar 30 06:22:21 EDT 2012

On Thu, Mar 29, 2012 at 5:00 PM, Julien Desfossez
<jdesfossez at efficios.com> wrote:
> On 27/03/12 06:10 AM, Stefan Hajnoczi wrote:
>> On Mon, Mar 26, 2012 at 1:27 PM, Julien Desfossez
>> <jdesfossez at efficios.com> wrote:
>>> - indexes are sent over the control data connection
>> What exactly is an index?  I guess it's a lookup data structure that
>> tells you where in the trace data to start reading?
>> Sorry if this is outside the scope of network streaming, maybe this is
>> basic Babeltrace/CTF background knowledge that I missed.
> Indeed the index is a concept in babeltrace, I should have mentionned it.
> When you open a trace with Babeltrace (or with the API), before printing
> the trace it generates an index in memory in order to have a mapping
> between the trace packets and their position in the trace files.
> I hope it makes this proposal clearer.

Thanks, I think I understand the general scheme and it makes sense.  I
guess the tricky part is buffering - whether to send a tiny packet per
trace event (useful in case machine A crashes) or to buffer up several
packets to reduce CPU and network consumption.  When machine A isn't
going to crash buffering is nice, but when machine A crashes buffering
is bad :).  With Linux in particular it might be okay - I've had good
results using netconsole to capture kernel panics.


More information about the lttng-dev mailing list