So...
imo, what is a dimension and what is a metric.
well to me a metric is a value that we want to measure on across all the
different permutations of that metric. The metric has to be generic in
the sense that it exposes no detail about the value it is measuring.
Metrics are like dry oats. Flavorless, bland, colorless but yet useful.
i.e *region.operation.count* - just described the count of all
operations that have been made against a region.
Dimensions provide "color" to the metric. i.e
*region.operation.count*... could have the following metrics:
* Availability zone *optional*
* distributed system id (unique DS ID, limit 255)
* server name (options are endless here)
* region name (the options are endless here)
* region type (partition,replicate,normal,replicate-proxy)
* operation type (put, get, invalidate,destroy
What we have is still a single metric, "*region.operation.count*" and
this metric will work... Anyone can read it and know what it is
measuring. The "slicing and dicing" comes from the color that is
provided by the dimensions and the combinations of any -> all of them.
Now, dimensions will change from metric to metric. i.e all the
dimensions for regions, might not all be applicable for other metrics...
but that goes without saying...
I hope that helps.
--Udo
On 1/15/19 13:44, Nicholas Vallely wrote:
FYI
I spoke with Udo about tagging specifically, not necessarily hierarchy
or naming. He said that adding tags that are unique, don't change
very often, and that you might want to group on seemed like the best
course of action. So for instance we discussed IP-PORT and he
mentioned that this might be a bad one to use because it could change
as a VM goes up/down and wouldn't be able to show a trend or be able
to easily group on in the same way you would get with a ServerName.
Other than that one, everything else that I have in the spreadsheet
meets this criterion, he also mentioned 'Region Type' as something
else we could tag.
Nick
On Tue, Jan 15, 2019 at 11:58 AM Dale Emery <dem...@pivotal.io
<mailto:dem...@pivotal.io>> wrote:
Hi Udo and all,
> On Jan 15, 2019, at 10:06 AM, Udo Kohlmeyer <u...@apache.org
<mailto:u...@apache.org>> wrote:
>
> It would be good to see the new Micrometer stats have a logical
grouping, that makes it easier for users to search for metrics.
Do you know of any useful guidelines or conventions for creating a
hierarchy of metrics, and criteria for evaluating the goodness of
the hierarchy?
And for which details to represent in the meter name hierarchy vs
tags/dimensions?
Dale
—
Dale Emery
dem...@pivotal.io <mailto:dem...@pivotal.io>
