Hi guys,
----- Original Message -----
> > [...]
> On my machine, this is:
>
> % ls -al /etc/pcp/pmcd/pmcd.conf /etc/pcp/pmcd/pmcd.options
> -rw-r--r--. 1 root pcpqa 770 Apr 23 21:26 /etc/pcp/pmcd/pmcd.conf
> -rw-r--r--. 1 pcpqa pcpqa 606 Apr 7 21:58 /etc/pcp/pmcd/pmcd.options
>
> It indicates that running the pcpqa suite leaves stains, as it were,
> on the well-bleached defaults that come with the base package. (I've
I want to cover my eyes and pretend I didn't just see those file
timestamps. 5 months since running QA there? Ouch!
> long dreamed of a pcpqa implementation that leaves the /etc/pcp/*
> files and system daemons well alone, and runs private copies on
> alternate ports etc. only.)
I can't imagine this ever working in pcp/qa, its assumptions are so
deeply entrenched in this regard.
The pcp/qa goals are somewhat different to "standalone" software.
It tests with installed services like memcached, rsyslogd, ds389,
and on and on, and its not shy in reconfiguring them to exercise
core PCP behaviour. It intentionally tests the installed packages
as well (so pcp upgrade paths from one version to another are also
able to be tested). Effort required far outweighs the benefits of
changing this approach after >20 years of effort now.
So, I agree with Kens synopsis that disciplined QA test writing is
the key to tackling this in the core pcp/qa testsuite. Ken, maybe
we could check after every test finishes (so, inside ./check) that
the permissions are not modified, using the qa/994 mechanism? (and
report immediately on any test that breaks it, as it happens?)
However, Frank, I do think it is an achievable goal for the separate
JVM, web, and other trees (like pcolby has shown us already, and as
I've been trying to convey). By a happy coincidence, I implemented
it yesterday in the new web tree - I'll push this for review shortly.
In the one test we have for all of the pcp-web-manager code combined
(sob! one test! again, I can't help but feel partially responsible
- oh well, let's make the most of it now) this conversion was doable
but it still took me an hour to do and test properly. Multiply that
by hundreds of core pcp/qa tests - many of which are not amenable to
conversion at all - the effort vs reward is not favourable IMO - we
would be better off investing that effort in other ways.
Please make use of VMs/containers/whatever throwaway-hosts if you're
concerned about this for local system testing.
cheers.
--
Nathan
|