log sampled invalid server ids

Summary: when analyzing a quic_cid_real0_pkts increase issue, I found it is useful if we know what are invalid server ids. I am sampling it in bpf program via quic_stats and log it in katranlb.

Reviewed By: lima1756

Differential Revision: D48018555

fbshipit-source-id: 3bb120a1f823a9f91069e7a4d3fa598763f088ed

katran/lib/BalancerStructs.h

@@ -117,6 +117,7 @@ struct lb_quic_packets_stats {
   uint64_t ch_routed;
   uint64_t cid_initial;
   uint64_t cid_invalid_server_id;
+  uint64_t cid_invalid_server_id_sample;
   uint64_t cid_routed;
   uint64_t cid_unknown_real_dropped;
   uint64_t cid_v0;

katran/lib/KatranLb.cpp

@@ -38,6 +38,7 @@ namespace katran {
 
 namespace {
 using EventId = monitoring::EventId;
+constexpr int kMaxInvalidServerIds = 10000;
 } // namespace
 
 KatranLb::KatranLb(
@@ -2058,6 +2059,17 @@ lb_quic_packets_stats KatranLb::getLbQuicPacketsStats() {
         sum_stat.ch_routed += stat.ch_routed;
         sum_stat.cid_initial += stat.cid_initial;
         sum_stat.cid_invalid_server_id += stat.cid_invalid_server_id;
+        if (stat.cid_invalid_server_id_sample &&
+            (invalidServerIds_.find(stat.cid_invalid_server_id_sample) ==
+             invalidServerIds_.end()) &&
+            invalidServerIds_.size() < kMaxInvalidServerIds) {
+          LOG(ERROR) << "Invalid server id "
+                     << stat.cid_invalid_server_id_sample << " in quic packet";
+          invalidServerIds_.insert(stat.cid_invalid_server_id_sample);
+          if (invalidServerIds_.size() == kMaxInvalidServerIds) {
+            LOG(ERROR) << "Too many invalid server ids, will skip logging";
+          }
+        }
         sum_stat.cid_routed += stat.cid_routed;
         sum_stat.cid_unknown_real_dropped += stat.cid_unknown_real_dropped;
         sum_stat.cid_v0 += stat.cid_v0;

katran/lib/KatranLb.h

@@ -1156,6 +1156,12 @@ class KatranLb {
    * Callback to be notified when a real is added or deleted
    */
   RealsIdCallback* realsIdCallback_{nullptr};
+
+  /**
+   * set of invalid server ids sampled from katran
+   */
+
+  std::unordered_set<uint64_t> invalidServerIds_;
 };
 
 } // namespace katran

katran/lib/bpf/balancer_kern.c

@@ -776,6 +776,7 @@ process_packet(struct xdp_md* xdp, __u64 off, bool is_ipv6) {
             // pos 0 means the entry for the server id is not initialized.
             // fallback to ch
             quic_packets_stats->cid_invalid_server_id += 1;
+            quic_packets_stats->cid_invalid_server_id_sample = qpr.server_id;
             quic_packets_stats->ch_routed += 1;
           } else {
             pckt.real_index = key;

katran/lib/bpf/balancer_structs.h

@@ -143,6 +143,7 @@ struct lb_quic_packets_stats {
   __u64 ch_routed;
   __u64 cid_initial;
   __u64 cid_invalid_server_id;
+  __u64 cid_invalid_server_id_sample;
   __u64 cid_routed;
   __u64 cid_unknown_real_dropped;
   __u64 cid_v0;