Awkward NIC Metric To OpMon Assignment

[aeoranday]

dpdklibs/src/IfaceWrapper.cpp

Lines 341 to 361 in a78a6d8

	if ( std::regex_match(name, match, queue_regex) ) {
	auto queue_name = match[1].str() + '-' + match[2].str();
	auto & entry = xq[queue_name];
	try {
	opmonlib::set_value( entry, match[3], m_iface_xstats.m_xstats_values[i] );
	} catch ( const ers::Issue & e ) {
	ers::warning( MetricPublishFailed( ERS_HERE, name, e) );
	}
	continue;
	}
	google::protobuf::Message * metric_p = nullptr;
	static std::regex err_regex(R"(.+error.*)");
	if ( std::regex_match( name, err_regex ) ) metric_p = & xerrs;
	else metric_p = & xinfos;
	try {
	opmonlib::set_value(*metric_p, name, m_iface_xstats.m_xstats_values[i]);
	} catch ( const ers::Issue & e ) {
	ers::warning( MetricPublishFailed( ERS_HERE, name, e) );
	}

The NIC metrics grabbed by rte_ethdev.h are dependent on the NIC installed to a machine, so this is expected to vary with upgrades, detector servers, etc. However, the assignment to OpMon is direct and is not necessarily tuned to an arbitrary NIC. This results in missed OpMon metric assignment in the case of slight naming differences or a regular emission of warnings for a NIC metric that is not defined in OpMon.

We are likely not interested in tracking all possible NIC metrics, so we should define the set of metrics we are interested in and define them in the OpMon metrics. The translation between NIC and OpMon metrics would then need to be robust to slight naming differences.

An immediate quality-of-life improvement would be to emit these warnings only once instead of every metric poll.