From 9f269ed825b7602025ce6dcce709cf577949a806 Mon Sep 17 00:00:00 2001
From: Matthew Johnson <matthewj@netflix.com>
Date: Thu, 1 Feb 2024 18:29:14 -0500
Subject: [PATCH] add net.perf.conntrackAllowanceAvailable metric

In a recent incident, the net.perf.conntrackAllowanceExceeded was a strong
signal for the cause. Adding the conntrackAllowanceAvailable metric will
allow it to be used as an input to auto-scaling policies.
---
 lib/ethtool.h       | 13 +++++++++++++
 lib/ethtool_test.cc |  8 ++++++--
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/lib/ethtool.h b/lib/ethtool.h
index f6b3f81..049a1a9 100644
--- a/lib/ethtool.h
+++ b/lib/ethtool.h
@@ -89,6 +89,19 @@ class Ethtool {
         continue;
       }
 
+      found = stat_line.find("conntrack_allowance_available:");
+      if (found != std::string::npos) {
+        auto metric = registry_->GetGauge(id_for("net.perf.conntrackAllowanceAvailable", iface, nullptr, net_tags_));
+        std::vector<std::string> stat_fields = absl::StrSplit(stat_line, ':');
+        try {
+          auto number = std::stoll(stat_fields[1]);
+          metric->Set(number);
+        } catch (const std::invalid_argument& e) {
+          atlasagent::Logger()->error("Unable to parse {} as a number: {}", stat_fields[1], e.what());
+        }
+        continue;
+      }
+
       found = stat_line.find("linklocal_allowance_exceeded:");
       if (found != std::string::npos) {
         auto metric = registry_->GetMonotonicCounter(id_for("net.perf.linklocalAllowanceExceeded", iface, nullptr, net_tags_));
diff --git a/lib/ethtool_test.cc b/lib/ethtool_test.cc
index 01d2994..5b088be 100644
--- a/lib/ethtool_test.cc
+++ b/lib/ethtool_test.cc
@@ -35,13 +35,15 @@ TEST(Ethtool, Stats) {
       "    bw_out_allowance_exceeded: 0\n",
       "    pps_allowance_exceeded: 0\n",
       "    conntrack_allowance_exceeded: 0\n",
+      "    conntrack_allowance_available: 100\n",
       "    linklocal_allowance_exceeded: 0\n",
       "    queue_0_tx_cnt: 368940\n",
       "    queue_0_tx_bytes: 126196057\n"};
 
   ethtool.stats(first_sample, "eth0");
   auto ms = registry.Measurements();
-  EXPECT_EQ(ms.size(), 0);
+  // one gauge, the rest mono counters
+  EXPECT_EQ(ms.size(), 1);
 
   std::vector<std::string> second_sample = {
       "NIC statistics:\n",
@@ -50,6 +52,7 @@ TEST(Ethtool, Stats) {
       "    bw_in_allowance_exceeded: 5\n",
       "    bw_out_allowance_exceeded: 10\n",
       "    conntrack_allowance_exceeded: 15\n",
+      "    conntrack_allowance_available: 110\n",
       "    linklocal_allowance_exceeded: 20\n",
       "    pps_allowance_exceeded: 25\n",
       "    queue_0_tx_cnt: 368940\n",
@@ -58,13 +61,14 @@ TEST(Ethtool, Stats) {
   // we need two samples, because these are all monotonic counters
   ethtool.stats(second_sample, "eth0");
   ms = registry.Measurements();
-  EXPECT_EQ(ms.size(), 5);
+  EXPECT_EQ(ms.size(), 6);
 
   auto map = measurements_to_map(ms, "");
   std::unordered_map<std::string, double> expected = {
       {"net.perf.bwAllowanceExceeded|count|in", 5},
       {"net.perf.bwAllowanceExceeded|count|out", 10},
       {"net.perf.conntrackAllowanceExceeded|count", 15},
+      {"net.perf.conntrackAllowanceAvailable|gauge", 110},
       {"net.perf.linklocalAllowanceExceeded|count", 20},
       {"net.perf.ppsAllowanceExceeded|count", 25}};
   EXPECT_EQ(map, expected);