I like when people ask if you can change the #CNI in a #kubernetes cluster on the fly, without destroying and recreating the cluster. Well, you *can* but is it simple, non-expensive, and not time-consuming? No. It's also very dependent on your specific application deployments.

https://cilium.io/blog/2023/09/07/db-schenker-migration-to-cilium/

#epbf #cilium

I updated the benchmarks for #epbf overhead in ebpf_exporter.

Overhead for an empty probe for getpid() syscall that immediately returns:

* 15ns for a tracepoint (tp_btf) 🚀
* 24ns for an fentry 🏎️
* 137ns for kprobe 🐌

@train I know it is useful for ~3 things.
1. Defining quirks for ALSA
2. Defining quirks for Hid/++
3. Exposing highly semantic interfaces to userspace.

For these former two, #ePBF brings benefits by allowing userspace to quickly iterate over necessary fixes. Also there is a good chance you can then backport fixes easily to LTS.
For the latter exposing interfaces via ioctl or sysfs is cumbersome in highly semantic scenarios. Here #eBPF could make sense as long as latency, etc. is not a concern.