Improving graphs in console
Here are three things that I think would improve the bucket-level graphs in the Couchbase Console.
1) When you're looking at the live graphs for a data bucket, you can change the granularity between "Minute", "Hour", "Day", etc. In "Minute", the units are ops/second. In "Hour", they're ops/minute. In "Day", they're ops/hour, etc. However, all of these are mislabeled as ops/second.
2) Add more granularity to the graphs. Imagine you have a process that generates heavy load at the beginning of the hour, every hour. If you looked at the hourly chart, this peak would be hidden away because the graph only plots one point for the whole hour. If you had more points per unit of time, it would be easier to see these peaks.
3) Historical data. It'd be nice to have the ability to see last week's metrics, or last month's.
The graph heading is correct. It is always showing ops/s. In the more "zoomed out" views it starts averaging it, but it always shows it as operations per second. For example in the hour view, one hour stretches across the entire graph width, so has lots of datapoints. Even in the "Day" view, there are still a meaningful number of datapoints we use (but not sure of the exact windows of averaging for the different modes).
So this should address your second point as well, in the hour view you are able to see load spikes that are shorter than an hour, but you do get some aliasing of course, due to the averaging window, so Nyquist applies and very brief spikes won't be visible.
Good idea about the historic data. Right now what you could do is use the REST interface to interface with another monitoring/graphing system (e.g. use the existing nagios plugin) and then you can use outside solutions to archive those metrics for historic analysis.
Hope this helped!
Hope you are having fun with Couchbase, what kind of project are you working on?
Cheers,
Frank
Thanks for the feedback-- will make sure these get to the right person!