When Onload is being used in an interrupt-driven mode (see Interrupt Handling - Using Onload) interrupts should affinitized to the same NUMA node running the Onload application, but not on the same CPU core as the application.
When Onload is spinning (busy-wait) there will be few (if any) interrupts, so it is not a real concern where these are handled.
If you are using a SHRUB controller in interrupt-driven mode, this raises interrupts at a rate of about one interrupt per 512 packets received on the Express datapath. This rate is again low enough that it is typically not a real concern where these are handled.