Clarity around PaaS environments, charging and hosts
Hi,
I was hoping to get some details around PaaS environments specifically and how they match up to our terminology of hosts and as such pricing.
For example:
VMWare: Host = a Virtual Machine
OpenShift Dedicated: Host = Compute Node (I believe)
Azure Web Apps: Host = Azure Instance
AWS: Host = AWS Instance
Can I get these confirmed as well as the same workings for other environments (OpenStack, CloudFoundry etc.) as well as any other useful terminology conversion like is a Kubernetes Pod normally the same as a process group.
The next part of this then comes into pricing of PaaS. We are deploying a PaaS agent to most of these rather than full stack as there is no access to the host. As such where does the pricing work from:
Is it what gets seen as the host, e.g. OpenShift Dedicated Compute Node and normal RAM sizing applies. If this is the case, when multiple PaaS agents are deployed on the same Compute Node, is the pricing worked out per agent or per reported host/Compute Node?
If it is per agent, this would be extremely expensive for customers and yet they would be receiving reduced visibility as PaaS agent does not have log monitoring, network monitoring etc.
Or is it done on the process the agent is actually injected into, e.g. the Jetty process within the Docker container. This is the one I seem to get told/implied but looking at the account consumption pages, it does not seem to tally up.
Best Answer
This answer above is outdated. We now calculate "application only" models (on K8s) using memory available to a container. If there is a limit on the container, this will be used.
Later in 2021 we will calculate based on memory utilization instead of limits.
What about GCP Kubernetes clusters?
In principle we charge the same way as for normal hosts. per host unit meaning memory.
in PaaS environments we usually have a paas memory limit which we understand in CloudFoundry and Azure WebSite.
an Azure WebSite instance in dedicated mode means that the host is actually fully reserved for the Azure WebSite app and also shown as a single "host" in the Dynatrace UI, so memory limit and the memory shown in the chart are the same.
If no memory limit can be detected it falls back to the memory seen of the host. This is a problem for Openshift and kubernetes if one does not run in FullStack mode. We will need to add support here to read out the memory limits from the kubernetes master. @Daniela R. is aware of this.
To recap.
* Pricing in CloudFoundry works fine in Fullstack and buildpack mode.
* Pricing in Azure WebSite work fine as long as the instance is dedicated
* Pricing for Openshift works only fine for FullStack mode. In paas mode we would currently not take memory limit into account, we will tackle this in the future.
