Current situation
As far as I know, we don't have tests for BPF integrated in the test/release pipeline.

Impact
Software using BPF might break if kernel bugs are introduced in new releases and it can more easily go unnoticed without automatic testing. As an example, Inspektor Gadget uses various BPF tools (from traceloop, tcptracer-bpf, BCC). We've had an issue with the BCC opensnoop tool in a Flatcar Edge release (see iovisor/bcc#2726).

Ideal future situation
Some basic tests of common BPF tools should be automatically executed before each release. Sometime, it can break because of a kernel bug, or because of a bug in the tool. It should be assessed to decide whether the release can go.

Implementation options
Some existing tests could be reused:

• BCC tests. I would like additional tests for opensnoop and execsnoop because the previous regression bug was in those tools.
• tcptracer-bpf test
• traceloop manual test with systemd-run