Re: [tracecompass-dev] uuid of the trace clock

[Geneviève Bastien <gbastien@xxxxxxxxxxxx>]

Hi Chris,


On 2017-11-13 04:01 AM, Mansky Christian wrote:
> Dear tracecompassers.
> I have a question regarding the uuid of the trace-clock (in the
> metadata). I only did a very quick research of the following, so
> please correct me if I'm wrong. However, it is my understanding that
> Lttng takes the boot_id (in /proc/sys/kernel/random) as the uuid of
> the clock. Since Lttng uses the CLOCK_MONOTONIC this makes sense, as
> every combination of the monotonic timestamp and the boot_id creates a
> unique timestamp.
> If I now have a tracing project with multiple traces from different
> boot-ups, I would like to identify traces that belong together (for
> creating an experiment). Uuid-wise; hence, from the same bootup and
> timestamp-wise; meaning overlapping traces). Is there a quick way to
> do this?
You mean programmatically or a for a user looking at the Trace Compass GUI?

The former is the easiest. As you said, traces from the same bootup have
the same clock-id. That's why we use the clock-id as the host ID for CTF
traces (using hostname have more chances of name clashes, like a few VMs
named ubuntu, but clock-id is unique per machine). That's how we know a
UST and kernel traces go together for instance. Then we use
trace1.getHostId() in code to identify the host. Most analyses in Trace
Compass are host-aware.

In the Trace Compass GUI, if you select an opened CTF trace and look at
the Properties view, you will see under the "Trace Properties", a
property called "Host ID", which corresponds to the clock id.

I hope this helps,
Geneviève