htrace-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Colin P. McCabe" <cmcc...@apache.org>
Subject htrace gui meeting minutes
Date Sat, 04 Apr 2015 01:57:22 GMT
We had: Abe, Colin, May, Masatake in attendance.

We talked about more closely integrating the search page with the
swimlanes view.  In particular, would it be possible to have them be
one view?  Maybe.  We need zoom buttons, time delineation bars, and
some other minor features in the swimlanes view.   We talked about
having parent/child relationships shown on this view.

Pagination has been a trouble spot for the search page.  We need to
use entire spans as continuation tokens, and correctly handle cases
where we are sorting by a key other than span id.  It would be nice if
the web UI could fetch 4 pages worth of results, in order to give the
user an idea of whether there were many results, or only a few.  This
may be difficult to do with backgrid.js.

It would also be nice to be able to click on a span in the view to see
its children (May's suggestion).  This may also be difficult with
backgrid, but it would be really cool.

We talked about having tabs to move between different activities.  The
ones discussed were server info, configuration, search, and
aggregrates.

server info: The server info tab must contain an owl.  We already have
a serverinfo JSON RPC, so this could be used.  We might also want
server stats on this page eventually.

configuration: It would be nice to have the configuration page
accessible via /conf, for familiarity to hadoop admins.  This would
show the htraced configuration.

aggregates: this view would provide all the same predicates as the
search view, but would allow users to choose an aggregate from a
drop-down list.  Examples include: average, median, std deviation,
max, min, count.  We also would want a bucket size.  One really great
example is taking the sum of the number of spans within each 10 minute
interval in the last 24 hours, to see activity spikes.  We may need to
think about whether this puts too much load on the server.  if so,
we'd have to precompute a few aggregates, or do this outside htraced.
The histogram view is a later target than the other views.

best,
Colin

Mime
View raw message