Does the tool support per-cpu array / maps?

[raghu-nandan-bs]

looking at this it seems like reading and processing values from per-cpu arrays/maps are not supported. Can you please confirm if this is correct?

if they are indeed not supported, was it a design decision to not consider per-cpu maps?

[bobrik]

It is not supported because we didn't need it. I planned to do this eventually.

Do you have a use-case that requires per-cpu maps?

[raghu-nandan-bs]

Thanks @bobrik

collected data from functions like netif_receive_skb requires that we use per cpu counters so that we dont run into race conditions. I wish we could export this as per-cpu metric... Also, the intention is to track "microburst"s so another problem I'm running into is tracking max per "x" millisecond values over a scrape interval.

Please let me know if you have a solution in mind.

[bobrik]

It seems like there are two use-cases:

1. Collect a metric per cpu and export it per cpu.
2. Collect a metric per cpu, but export it as a combined counter.

It sounds like you want the former. I can see some people wanting the latter, so we should probably support both.

Also, the intention is to track "microburst"s so another problem I'm running into is tracking max per "x" millisecond values over a scrape interval.
Please let me know if you have a solution in mind.

This sounds like an interesting non-trivial example we might want to have in the repo. I don't have a solution for this off the top of my head, but please file a separate issue for this. If you already have a non-ebpf_exporter based solution, it would be good to attach it there for a reference.

[raghu-nandan-bs]

Thanks for the response @bobrik , will get back

[bobrik]

Please take a look at #226 for per cpu map support.

[raghu-nandan-bs]

thanks @bobrik ! will check