During a few short performance test I used CloudWatch for the first time.
I don’t like it.
I was looking around and checked other metrics tools but that didn’t help because I don’t like them either.
So it’s not really a CloudWatch problem. It’s me.
I believe I want both the raw data as well as fast aggregated data. I I would like to be able to get CPU usage in the last thirty minutes per ten seconds and I would like to get all the information about all individual calls to a certain service in the last 24 months. And I don’t want to use Cloud Watch for the former and logstash or similar for the latter. I want to have both in one interface. (Though I also see an addition need for logstash, but that is another story)
I am wondering if I am wrong about my expectations or if my understanding of the tools that are around is wrong.
OpenTSDB seems to be up my alley in terms of providing both, but supposedly does not perform well in some scenarios. With Netflix Atlas, I don’t like that it does not support other backends and I also not sure how flexible it is in the set-up but I believe if it was possible to connect Atlas to OpenTSDB it would be what I am after. With Graphite I don’t like the store mechanism since it isn’t covering my long term storage “needs”. With InfluxDB I am not happy about the lack of clustering (there is a experimental version) but I also find it is doing too much even though I think I would like it’s query interface.
There are several paid options, but my understanding is too limited to understand what metrics I want in one year to be able to decide this now.
Well, there is no way around it, I need to try things out.