05 Apr 2023 06:51 AM - last edited on 05 Apr 2023 07:30 AM by Ana_Kuzmenchuk
I have some questions which are not 100% clear to me. Therefore I would appreciate any help.
1. In order to reduce HU consumption does it make any difference if we follow the Classic full-stack injection or Cloud-native full-stack injection? ( we assumed with CNFS we could limit monitoring to certain namespaces/pods and would need only HU for the memory used by the namespaces and pods. )
2. If we use application only to achieve the above scenario then would we only need Host Units for the memory used by the pods?
3. And would we still be able to have End-To-End traces from the UI to the database in App-Only mode?
Thanks in advanced 🙂
As far I know, the HU are calculated based on Memory detected, not used.
When you run CNFS or CFS, the HU calculation will be based on your KS Nodes memory.
For App-only, I am not sure, and really interested on this. Let's wait for someone with the right answer 😁
edit: I think I got it here: https://www.dynatrace.com/support/help/manage/monitoring-consumption/application-and-infrastructure-...
For these technologies, host units are calculated based on memory detected at the operating system instance or container level. Calculations take into account the detected memory limit, such as container memory limits. If no memory limits are detected, calculations may use the underlying host memory, which may reflect a higher number of host units.
My understanding is:
1. Classic full-stack injection or Cloud-native full-stack injection will be based on your KS Nodes memory as @dannemca mentioned above.
2. Application only would consume the maximum amount of resources per monitored application(per monitored pod or namespace)
3. With application only you still get the "full-stack mode", so services and traces are available as well as UX etc.
Limitations: When deployed in application-only mode, OneAgent monitors the memory, disk, CPU, and networking of processes within the container only. Host metrics aren't monitored. Topology is limited to pods and containers.
If someone has a different understanding, please correct me 😉
There is another limitation of App-only. Native DT log monitoring does not work in app-only mode. In case of App-only you have to ingest to the logs via the log ingest API eg. with fluentd or fluentbit...
And @dannemca has found the right asnwer for the app only HU calculation. App-only worth using only if there are memory limits.