yanet_get_nic_counters allocates count handlers and sets list->count = count, but only fills entries that match the query. If there are fewer (or zero) matches, the remaining handlers are left uninitialized and will be encoded/returned as garbage. Compute match_count first (like the existing query-based counter listing logic earlier in this file), allocate based on match_count, and set list->count = match_count (return NULL when match_count == 0).

cp_config_gen_get_nic_counter_storage() currently just aliases cp_config_gen_get_device_counter_storage(), but NIC counters are named separately ("nic_*") and aren’t registered in the device counter registry by default. As-is, NIC lookups will return the same storage as device counters and won’t contain NIC metrics. Introduce a dedicated NIC counter storage entry in the counter storage registry (or explicitly extend device counters with NIC fields) so this function returns the correct storage.

Initializer for the worker_counter_info entry "remote_tx" has an extra trailing comma inside the struct initializer (..., (const size_t[]){...},},). This provides too many initializers for the anonymous struct and should fail compilation; remove the extra comma so the entry has exactly 3 fields like the others.

iterations is a local variable initialized to 0 on each worker_loop_round() call, so the (iterations + 1) % NIC_STATS_FREAQUENCY counter resets every round. Since the inner while (1) typically exits once the packet fronts are empty, this likely never reaches 0 and worker_unload_nic_stats() won’t run (NIC stats won’t be updated). Persist the counter across rounds (e.g., store it in struct dataplane_worker/dp_worker, or use a time-based check).

Macro name NIC_STATS_FREAQUENCY is misspelled; rename to NIC_STATS_FREQUENCY to avoid propagating the typo into call sites and future code.

Large blocks of commented-out rte_eth_stats_get/fprintf logic were left in stat_thread(). If this logging is no longer needed, remove the dead code; if it is still useful for debugging, consider guarding it behind a build flag (to keep the runtime path and source clean).

Comment uses informal abbreviations ("bcs", "unload not often") and is hard to parse. Please rephrase in clear English (e.g., explain the cache-line placement trade-off explicitly).

yanet_get_nic_counters looks up counter storage via cp_config_gen_get_nic_counter_storage(), which currently aliases cp_config_gen_get_device_counter_storage(). Device counter registries only register "rx/tx/rx_bytes/tx_bytes" (see lib/controlplane/config/cp_device.c), so the NIC-prefixed counters won’t exist in this storage and the NIC RPC will return no real data. Either register these NIC counters in the device counter registry/storage, or change this lookup to point at the actual storage where NIC stats are written.

The new rpc Nic is added to the public gRPC service, but there is no corresponding handler implemented in controlplane/internal/gateway/counters_service.go (so clients will get Unimplemented unless/ until it’s wired up). Please add the server-side method that calls DPConfig.NicCounters() (or remove the RPC until it’s supported end-to-end).

struct nic_stat is currently unused (all related logic is commented out). Removing the unused type and commented code will keep the dataplane hot-path file easier to maintain.