I believe the current thinking is that if there's a fix on OVN side for this scenario, it would involve OVN controller splitting the list of ports for the UNKNOWN flow into pieces, each of which would be below the kernel/netlink limit. Then call to controller() at the end of each chunk, which would reinject the packet back into the table, for the next chunk, until all chunks are completed.

If so, the description here should be adjusted to reflect this.

(In reply to Ihar Hrachyshka from comment #2)
> I believe the current thinking is that if there's a fix on OVN side for this
> scenario, it would involve OVN controller splitting the list of ports for
> the UNKNOWN flow into pieces, each of which would be below the
> kernel/netlink limit. Then call to controller() at the end of each chunk,
> which would reinject the packet back into the table, for the next chunk,
> until all chunks are completed.
> 
> If so, the description here should be adjusted to reflect this.

Ihar, you're right but for our team's tracking it's better if we open a new bug to track the alternate approach.  I opened bug 2232152 for that and will be closing this one.