There was some last minute discussion last week about posting a
Ganglia-web tarball alongside the full 3.3.5 release tarball

There is no major technical challenge involved in doing that, but I have
some comments about it:

- someone (maybe me) should have produced it much earlier in the release
cycle for testing: I wouldn't be comfortable producing it at the last
minute and instantly making it official.  Even though the contents of
the tarball would be identical to the contents of the web directory in
the full tarball, we still need to maintain a policy of testing every
tarball for some number of days.  There is always human error even in
the simplest of tasks, and the `cooling off' period catches such things.

- several times, I've raised the possibility that ganglia-web should
diverge again - but without bringing back the old web interface.  Future
monitor-core releases would have no included web/ component.  I think
the web team are doing great work, certainly a lot more than I've ever
contributed.  That is why I was so keen to see their work merge into the
main release and become the official web interface.  Now I can see that
may not be the perfect scenario either, so I think that the best way to
let the web releases live up to their full potential is to start making
releases at a version number higher than where we are now, e.g. the
earlier releases were versioned gweb2-2.20, but now they would just
release something like this:

ganglia-web-3.4.0.tar.gz
ganglia-web-3.4.1.tar.gz (bug fix)
ganglia-web-3.5.0.tar.gz (new features)

In fact, there could be a minor release every 1-2 months if the
development pace continues at the rate it has been going, so we could
see 3.8.0 or 3.10.0 by Christmas and that would be perfectly fine.
There is no reason not to reach for 3.50.0 or beyond, and there is no
expectation that monitor-core must keep up or catch up.

The only expectation that I do have is that the releases should not
cause confusion, so they should not go back to version numbers < 3.4.
It must be unambiguous that there is now only one web/

Based on my experience with the 3.3.5 release, I think that is the best
approach for web:

a) it is good for packaging, because someone who has ganglia-web 3.3.5
on Debian or Fedora can update to 3.4.0 or 3.5.0 automatically

b) it avoids confusion: users always look for the highest release
number, they don't have to choose between default web and web2

c) it allows a different release process (e.g. autotools is not needed
for releasing PHP code, as there is nothing to compile)

There are two issues I can see, these are not major obstacles, just
things to be aware of:

- released Linux distributions typically accept bug fix revisions (e.g.
if they are carrying 3.3.5, they'll accept 3.3.6 with a bug fix or
security fix, but they won't accept 3.4.0) - so if ganglia-web 3.5
becomes part of Debian 7.0, then it could be in that catalog for a good
18 months, and the web team may need to keep a branch for backporting
the most essential bug fixes to 3.4

- as web will no longer be tied to monitor-core, web should pay close
attention to the gmetad version, gmond versions (if the version metric
is present), etc and refuse to run with a version that is too old, if we
ever reach that scenario.  For example, if gmetad 3.5 has some new XML
attributes, and if web 3.10 absolutely depends on those attributes, then
it will need to check the version.  If web and monitor-core are
together, then maybe we could be more casual about that and just tell
people to use everything from the same tarball.

Does anyone have any other alternative to put on the table, or do we
just start going down this path, maybe with a web-3.4 release this month?



------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second 
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Ganglia-developers mailing list
Ganglia-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-developers

Reply via email to