Yeah, I believe that's correct. It would be cool if we persisted this to
a table, but it may be better to just use another metrics
gathering/storage system here rather than reimplement the wheel since we
have nothing here presently.
Eric did write an Accumulo backend for OpenTSDB. I don't think that
would give enough insight to all the types of stuff we track, but I
could be mistaken :)
On 5/19/14, 12:45 PM, Arshak Navruzyan wrote:
Josh,
Thanks for the clarification. In terms of the current implementation,
guessing the full stats are not stored anywhere. The monitor code just
polls each tserver and keep the info it collects in memory? Is that
right?
Thanks,
Arshak
On Mon, May 19, 2014 at 9:09 AM, Josh Elser <[email protected]> wrote:
I haven't seen one myself.
The rest interfaces for collecting metrics is, sadly, known to be lacking.
This aligns with some stated desires to make the rest interface for
Accumulo metrics a standardized API (which we assert stability guarantees
on) and then change the monitor to use those directly (instead of doing
goofy things in the implementation. We have a ticket somewhere for this (I
think Al Krinker has stated some interest in getting involved here).
I've been curious about what we could get with Graphite. This would be a
fun experiment.
On 5/7/14, 9:09 AM, Arshak Navruzyan wrote:
I was wondering, is there a plugin for Accumulo for collectd? I noticed
the current json / xml interfaces don't provide all the stats that the
monitor app shows. Perhaps collectd can be used to collect and store the
full set of stats.
