A rough summary of kstats which would be nice to have, since these are handled a little inconsistently and I think we want parity here for something we can surface to the control plane:
opte$n:
in_uft_flows, in_uft_capacity, out_uft_flows, out_uft_capacity. LFTs have equivalent stats already.in_uft_evictions, out_uft_evictions. We will not directly hit these today, but may do so in future if we make, e.g., all UFTs subject to LRU at capacity.
opte$n_$layer
evictions. We will actively hit this based on TCP state.
opte$n_routecache (separate kstat? Currently Port owns the top-level kstat name, rather than the driver).
A rough summary of kstats which would be nice to have, since these are handled a little inconsistently and I think we want parity here for something we can surface to the control plane:
opte$n:in_uft_flows,in_uft_capacity,out_uft_flows,out_uft_capacity. LFTs have equivalent stats already.in_uft_evictions,out_uft_evictions. We will not directly hit these today, but may do so in future if we make, e.g., all UFTs subject to LRU at capacity.opte$n_$layerevictions. We will actively hit this based on TCP state.opte$n_routecache(separate kstat? CurrentlyPortowns the top-level kstat name, rather than the driver).capacityoccupancy(not flows, because with Recast route caching as an LPM, make multipath decisions locally #539 we will track destinations)hitmissmiss_table_full