Frank,
On 5/7/2015 4:27 PM, Frank Ch. Eigler wrote:
[...]
Not sure what they are called, but for lack of a better name I
can't get the "metric operations" to work [...]
Yes, this is documented in pmwebapi(3) and
http://oss.sgi.com/bugzilla/show_bug.cgi?id=1094 .
OK, Thanks. Missed that part. This functionality would be very
interesting to us. We will try to find time to work on this, but that
probably wont happen until the fall, so let me know if you start looking
at this at some point so we don't duplicate effort.
My test case is about 1 month worth of a single node's archives with
~200 metrics collected at 30 sec intervals (3 GB), roughly 1 file
per day, ~30 files. [...] When editing the metrics it takes a while
for the field to populate. [...] but it would be great to get some
feedback that something is happening. [...]
We ship generally -unmodified- webapps, so if the normal
graphite/grafana webapp doesn't have a 'please wait ...' kind of
blinkenlight, it's not there in the pcp-webjs copy either.
OK, makes sense. I'm just starting to figure out how these pieces fit
together. I'll see if I can find some grafana supported solution.
3.
I am starting to do my own testing, but has anybody done scalability
studies? My largest dimension is going to be number of hosts. So is
it reasonable that I could plot ~5 metrics over a day but across 100
host archives? 1000 host archives? [...]
The largest views I've handled involved some dozens of hosts (split
over some hundreds of time-sliced archives). With the pmwebd -M
(multithreaded mode), it's behaved reasonably quickly; make sure
you're on pcp 3.10.4. How well it scales is also a function of the
web browser - so try the "png" (server-side) as well as "flot"
(browser-side) rendering options.
Thanks for the information, will continue to do testing.
Martins
|