Ah, I wanted to talk to this change a bit more, since the initial
commit message (for c97946c9ed24ae6d71a271fab0495dbd9013f4fd) was
a bit terse.
----- Original Message -----
> Changes committed to git://oss.sgi.com/kenj/pcp.git dev
> ...
> If we're going to change the arguments to pcp(1) as per commit
> 741f7b6 (not sure why this was a good idea), then need this
> change to make the usage message match the code.
>
The change referred to here is to remove one of two ways in which
a hostname can be specified to the pcp(1) script. Its questionable
because changing it means anyone/anything relying on old behaviour
would regress. I've punted that's unlikely, but it could happen -
I'd expect everyone would be using the widely known -h option, and
not relying on this quirky alternate specification.
Stepping back a bit, the missing link here is of course why do this
at all. The context is around doing something like pmlaunch (for
those who remember it! - it was basically a shell framework for one
tool to launch another, with state, usually from the gui tools).
What I've been considering is starting a layered package (similar to
the pcp-gui model) of non-PCP tools that act like the original tool,
but use PCP services. The plan being to provide functionality that
people know how to interpret (e.g. iostat output) with the additional
features that PCP provides (like archive mode). One difficulty I've
seen in the pmcollectl tool is that new options need to be added to
enable PCP features. Subtle issues like differences in command line
syntax for time window specification are problematic there for example;
and back to iostat - it has -h, -t, -z options already, again making
life a bit more difficult as these conflict with the "standard" PCP
options.
So, I was thinking of extending the pcp(1) command to have additional
capabilities around being able to run the pcp-variants of these other
tools - something like the git(1) frontend wrapper command, e.g.
$ pcp iostat -nh
$ pcp -h foo.bar.com iostat -nh
$ pcp -a /tmp/weekly iostat -nh
Anyway, this'll require extending the command line of the pcp.sh script
to cover the standard set, convert into env vars (ala pmlaunch) and run
the wrapped commands - which then use the environment to enable PCP-fu.
And the wrapped command ("iostat" above) conflicts with the alternate
hostname specification, and hence this change.
cheers.
--
Nathan
|