[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [tracecompass-dev] [diamon-discuss] Parsing external datasources in trace compass for regions of interest in a trace
|
Hi Matthew
See below for some comments.
/Bernd
On 02/09/2015 01:49 PM, Julien Desfossez wrote:
On 15-02-09 01:41 PM, Matthew Khouzam wrote:
Great feedback. I think for a first draft, this will work out swimmingly.
Right now, I was actually finding the first group of [], splitting with
"," [1], then for every timestamp, creating a bookmark.
That's a good first start I think.
so [12:20:23 , 12:23:22] some range
would give
bookmark 1-> 12:20:23 -> some range start
bookmark 2-> 12:23:22 -> some range end
I'm not sure exactly how clear it will be with overlapping ranges in the
TraceCompass list of bookmarks, but I guess it will be handled in a next
phase.
I think, having 2 bookmarks for a time range would not be very intuitive
for the user. There should be only one bookmark for each area of
interest regardless if it is for a single timestamp or a time range. In
the bookmark implementation you can save various attributes. You can
store the start and end time for a range. If the user opens a bookmark
with end time, Trace Compass will open the trace, seek to the start time
and then set the current active time range between start time and end
time. If the end time isn't set the last step is not done. The Trace
Compass framework has all the means to do these steps.
and
[22:33:44 ] [key : value] [name = value] [id -> ref ] ...
would parse as everything after the timestamp would be read as a string.
text [22:33:44] text2
would discard text.
Is this ok?
Yes, let's start with that. As we discussed, if we need to output
differently from the analyses, we will make the changes.
[1]note, it will not work with https://github.com/lttng/lttng-analyses
where dash is used.
Oh if you are referring to the "Work in progress" section, you are
right, don't worry about this, it is a prototype and cannot even be used
from the scripts in the master branch. When this is ready it will use
the standard format.
Thanks,
Julien
On 15-02-06 06:03 PM, Julien Desfossez wrote:
On 15-02-06 05:22 PM, Mathieu Desnoyers wrote:
----- Original Message -----
From: "Matthew Khouzam" <matthew.khouzam@xxxxxxxxxxxx>
To: "tracecompass developer discussions" <tracecompass-dev@xxxxxxxxxxx>, diamon-discuss@xxxxxxxxxxxxxxxxxxxxxxxxx,
lttng-dev@xxxxxxxxxxxxxxx
Sent: Friday, February 6, 2015 5:13:42 PM
Subject: [diamon-discuss] Parsing external datasources in trace compass for regions of interest in a trace
Hello,
I am right now looking getting external datasources to supply
information for a trace, like a region of interest.
Background: we can run an analysis on a trace manually, in python, excel
or what have you and come up with something interesting, like: around
12:30pm something happened to a trace. Let's dig deeper with tracecompass.
To make this work: we need to find a way to import the data, so we
should agree on a standard data standard. I personally like the LTTng
analysis outputs so I am picking that for now. :)
Here are some formats that could be acceptable as regions of interest
* [2015-01-15 12:18:37.216484041, 2015-01-15 12:18:53.821580313] name
* [2015-01-15 12:18:37.216484041, 2015-01-15 12:18:53.821580313] name
boookmark
* [ 2015-01-15 12:18:37.216484041] name boookmark
* [12:18:37.216484041] name boookmark
* [ 12:18:37.216484041, 12:18:53.821580313] name boookmark
* [ 2014-12-12 17:29:43.802588035] name with space
* [ 2014-12-12 17:29:43] irrational title
* [17:29:43.802588035] rational title
* [ 17:29:43] test test test
* [17:29:43,17:29:44] thing
* [12:18:37.216484041 ] (no text)
When there is one timestamp, we will create one bookmark.
When there is a range, we would create two bookmarks in a given trace,
the first one being appended with " start" the second with " end"
This should allow tracecompass to marginally benefit from the lttng
analyses that are looking pretty nice, and later perhaps from tcp
analyses from tshark and syslog stuff.
Does anyone have an objection to this way of working?
Do you intend on supporting timestamps that appear elsewhere than
the beginning of lines ? Do you consider all text at the right of
the timestamp or timestamp range to be part of the naming of the
region of interest linked to the timestamp/timestamp range ?
Julien, did we want to reserve [] for only timestamp range or
we also want them for single timestamps ?
Right now the [] are only valid with time ranges (they are not accepted
with --begin and --end). I don't mind allowing the brackets for all
timestamps (like in dmesg), we still have the "," to identify if it's a
range or not.
But for convenience, they will remain optional for --begin and --end.
We just have to keep in mind, that [32,64] could be a range of
values/frequency and not necessarily timestamps.
Julien
_______________________________________________
diamon-discuss mailing list
diamon-discuss@xxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/diamon-discuss
