From 923bbd696fabd6a2a11f380cf4368974461d9690 Mon Sep 17 00:00:00 2001
From: James Tucker <james@tailscale.com>
Date: Wed, 11 Jun 2025 13:56:46 -0700
Subject: [PATCH] prober: record DERP dropped packets as they occur

Record dropped packets as soon as they time out, rather than after tx
record queues spill over, this will more accurately capture small
amounts of packet loss in a timely fashion.

Updates tailscale/corp#24522

Signed-off-by: James Tucker <james@tailscale.com>
---
 prober/derp.go | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/prober/derp.go b/prober/derp.go
index e21c8ce76..c7a82317d 100644
--- a/prober/derp.go
+++ b/prober/derp.go
@@ -425,6 +425,24 @@ type txRecord struct {
 	txRecords := make([]txRecord, 0, packetsPerSecond*int(packetTimeout.Seconds()))
 	var txRecordsMu sync.Mutex
 
+	// applyTimeouts walks over txRecords and expires any records that are older
+	// than packetTimeout, recording in metrics that they were removed.
+	applyTimeouts := func() {
+		txRecordsMu.Lock()
+		defer txRecordsMu.Unlock()
+
+		now := time.Now()
+		recs := txRecords[:0]
+		for _, r := range txRecords {
+			if now.Sub(r.at) > packetTimeout {
+				packetsDropped.Add(1)
+			} else {
+				recs = append(recs, r)
+			}
+		}
+		txRecords = recs
+	}
+
 	// Send the packets.
 	sendErrC := make(chan error, 1)
 	// TODO: construct a disco CallMeMaybe in the same fashion as magicsock, e.g. magic bytes, src pub, seal payload.
@@ -445,10 +463,12 @@ type txRecord struct {
 			case <-ctx.Done():
 				return
 			case <-t.C:
+				applyTimeouts()
 				txRecordsMu.Lock()
 				if len(txRecords) == cap(txRecords) {
 					txRecords = slices.Delete(txRecords, 0, 1)
 					packetsDropped.Add(1)
+					log.Printf("unexpected: overflow in txRecords")
 				}
 				txRecords = append(txRecords, txRecord{time.Now(), seq})
 				txRecordsMu.Unlock()