On 07/07/2014 11:01 AM, Nathan Scott wrote:
Hi Mark,
----- Original Message -----
[...] A new QA test is brewing
to check this and a new pmie rule with syslog alerts for production
environments might be warranted too.
Sounds good. I've reverted 3c479d028 to give some minimal coverage
until then - you might find either pmprobe with a derived metric to
be a useful vehicle for doing that verification, or perhaps pmie.
(I'm not a huge fan of $PCP_QA_ESTIMATE_MEMAVAILABLE, FWIW - seems a
roundabout kind of way to check the value, the above two suggestions
might be simpler? - *shrug* - your call of course).
I was planning to just do something like this :
grep -q MemAvailable /proc/meminfo || notrun
k=$(pmprobe -Lv mem.util.available | awk '{print $NF}')
u=$(PCP_QA_ESTIMATE_MEMAVAILABLE=1 pmprobe -Lv mem.util.available |\
awk '{print $NF}')
and then check $k and $u are within 1% (which is near enough?), but
if you think the envvar to too hacky, then we can check it some other
way.
Would you recommend updating the pmchart "Overview" view to use this
metric?
yep good idea. mem.util.{dirty,writeback} would be useful too.
(this metric will be automatically included in the logged
set BTW, via src/pmlogconf/kernel/memory-linux - so nothing to do in
that neck of the woods).
ah good,
cheers
|