Allow partial program load failure in NewCollectionWithOptions

[i-Pear]

In NewCollectionWithOptions [1], if a program fails to load, the entire collection will be discarded. However, there are scenarios that require a "best-effort" approach, where the failed program is skipped, and a collection containing only the successfully loaded programs is returned.

Our use case involves a monitoring program for various LSM tracepoints. Before loading the collection, we do not know if the current kernel supports all the tracepoints. We want to skip those programs not supported by the kernel and continue using the remaining ones.

To achieve this, it might be necessary to add a field "AcceptFailure" to the CollectionOptions. This field would indicate whether each program in the CollectionSpec can be skipped if it fails to load.

What do you think about this?

Cc: @alban
[1] https://github.com/cilium/ebpf/blob/main/collection.go#L385

[ti-mo]

AcceptFailure invites silent breakage, so on the surface that doesn't sound like a good idea. Do you have an example set of programs that would only partially load? Do they have an lsm/ prefix in the ELF? I haven't tried using those, so I don't know the common failure scenarios.

A middle ground could be using LoadAndAssign to indicate which subset of programs you're interested in loading. Programs that don't appear in the input struct won't be handed to the kernel. I assume your application depends on having at least some programs successfully loaded. You could progressively fall back to a smaller and smaller subset of programs depending on what the kernel will accept.