Hey All,
I've pushed my work on this so far to lberk/dev on
sourceware.org/git/pcpfans.git
I've focused the changes within pcp.spec.in to start, so to try out the
changes, simply ./Makepkgs on a fedora or centos box.
Just about every pmda except core ones have been split into their own
pcp-pmda-<name> package, and on systems that already have a stable pcp
release (fedora 22 and older or centos 7 and older) there is a
pcp-compat package. This pcp-compat will ensure all the new subpackages will
be installed by default. This way there is no broken/missing elements
after an upgrade.
I've been careful to test qa after splitting off pmdas of each language
(python, perl, and C, respectively), and there is various updated qa
bits to match that. As it currently stands, I have no regressions on my
local runs.
I've also moved the python user tools under src/pcp, and created their
own sub package: pcp-system-tools. This was required, as now both
python and perl bindings are no longer installed by default (in an
attempt to lower the default footprint). As with the pmdas, this
package will be dragged in on existing releases. I've also added
symlinks to their previous commands, so no change is required when
invoking the tools.
I've also added a %post hook to build the pmns (and removes the
.NeedRebuild file if successful). I explicitly tested this by making
sure to nuke /var/lib/pcp* and running the full testsuite (also manually
checked the pmns before and after adding that hook, no difference).
Overall these changes have reduced the default footprint from 51mb to
7mb.
Any comments are appreciated. Frank has already pointed out on IRC, a
pcp-full-collection style package, which requires and installs the full
pcp set of packages would be nice to have. Especially as a convienience
to the user. I think this is a great idea, and unless there is strong
opposition to it, will be adding that asap.
Cheers,
Lukas
|