+++ This bug was initially created as a clone of Bug #1067 +++
The analogous issue has long existed and been long reported in the proc pmda.
A large amount of extra /proc scanning work is done for queries related to
individual metrics / individual processes. This makes it more likely that the
pmda interferes with the kernel (competing for process-table locks), and times
out. That in turn in effect kills the pmda. (See also bug #1036, bug #1079).
The phenomenon is easily observed with strace.
For example: a % pmval -i $$ proc.psinfo.ppid
involves reads to /proc/$$/stat, /wchan, /environ, only one of which was
needed.
For example: an active hotproc pmda scans multiple proc files for every
process, even if hotproc metrics are not being fetched; even involving files
not necessary to evaluate the hotproc selection criteria.