[RFE] PoC: top cpu/memory

[blanquicet]

PoC: top cpu/memory

Several people in the community is asking why we don’t have a gadget like top block-io, that shows the pods consuming more CPU/memory.

Q&A from IG meeting:

• Isn't it already provided by kubectl top? Yes, but:
  • IG would not depend on the metrics server like kubectl top does.
  • People already implementing the integration of IG for other use cases will get memory and CPU info for free. They won't need to integrate also kubectl top.
• As far as we know, there is no way to implement this using eBPF. So, how can we feed the IG data pipeline?
  • One of the proposals is that the gadget image doesn't contain an eBPF program but a binary which collect the CPU/Memory info. Then, IG will be in charge of creating a datasource from it which will feed the data pipeline. Is it feasible? We are not sure, so the proposal is to create a PoC with this idea to evaluate its feasibility.

[flyth]

If it's easy enough to support it from the IG core and provides enough value for many use cases, we can introduce it as a helper DataSource that can be activated from just the gadget.yaml by adding an annotation to the gadget. In this case (CPU/mem) it could just take an additional parameter for the interval whenever activated and then create a new DataSource and emit the values for every interval.

[blanquicet]

But we were considering make it a gadget, not a complement for other gadgets. Or did I misunderstand what you said?

[flyth]

Sure, but does it matter where information is coming from? If it's easy to maintain and can be useful in different scenarios, IMHO providing such functionality from the core is okay (WASM or spawning other binaries always increases overhead and in the latter case could also raise security concerns.)