<@U6EFFT5FG> I brought this up in office hours las...
# ebpf
z
@alessandrogario I brought this up in office hours last week: What are the reasons for the newer ebpf functionality being still considered experimental? When we discussed in office hours we thought it was mostly about compatibility with older kernels? We all agreed that if this is the primary reason it may be time to consider this the default version. What are your thoughts?
a
There are some codegen issues that need to be fixed, and it’s probably broken or unreliable on new kernels
I’ve retained the runtime codegen since one of the (future) milestones was to load user scripts as tables
it’s also missing fields, and the ones that are there can be enhanced
just to clarify, it is true that it will not work on old kernels due to lack of features, but it’s also broken on the newest kernels now
z
Hmm, broken on new kernels definitely sounds like a problem to me. What is required to get it working on those newer kernels?
a
I’m not sure what’s the problem specifically, but the codegen step is failing
either invalid types and/or some other breakage when using them
You might be able to use it on specific kernel versions, i think ~5.20 was working fine
s
Zach — What lead to this question? is it from you, or a customer of Fleet’s?
z
This was a customer who says that newer version gives them much better performance but they'll get questions internally if they enable "experimental" functionality in prod.