Increase health check channel buffer (vitessio#17821)

Signed-off-by: Manan Gupta <manan@planetscale.com>

go/vt/discovery/healthcheck.go

@@ -69,6 +69,9 @@ var (
 	hcPrimaryPromotedCounters = stats.NewCountersWithMultiLabels("HealthcheckPrimaryPromoted", "Primary promoted in keyspace/shard name because of health check errors", []string{"Keyspace", "ShardName"})
 	healthcheckOnce           sync.Once
 
+	// counter that tells us how many healthcheck messages have been dropped
+	hcChannelFullCounter = stats.NewCounter("HealthCheckChannelFullErrors", "Number of times the healthcheck broadcast channel was full")
+
 	// TabletURLTemplateString is a flag to generate URLs for the tablets that vtgate discovers.
 	TabletURLTemplateString = "http://{{.GetTabletHostPort}}"
 	tabletURLTemplate       *template.Template
@@ -632,7 +635,7 @@ func (hc *HealthCheckImpl) recomputeHealthy(key KeyspaceShardTabletType) {
 func (hc *HealthCheckImpl) Subscribe() chan *TabletHealth {
 	hc.subMu.Lock()
 	defer hc.subMu.Unlock()
-	c := make(chan *TabletHealth, 2)
+	c := make(chan *TabletHealth, 2048)
 	hc.subscribers[c] = struct{}{}
 	return c
 }
@@ -651,6 +654,9 @@ func (hc *HealthCheckImpl) broadcast(th *TabletHealth) {
 		select {
 		case c <- th:
 		default:
+			// If the channel is full, we drop the message.
+			hcChannelFullCounter.Add(1)
+			log.Warningf("HealthCheck broadcast channel is full, dropping message for %s", topotools.TabletIdent(th.Tablet))
 		}
 	}
 }

go/vt/discovery/healthcheck_test.go

@@ -23,6 +23,7 @@ import (
 	"io"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -1475,6 +1476,45 @@ func TestDebugURLFormatting(t *testing.T) {
 	require.Contains(t, wr.String(), expectedURL, "output missing formatted URL")
 }
 
+// TestConcurrentUpdates tests that concurrent updates from the HealthCheck implementation aren't dropped.
+// Added in response to https://github.com/vitessio/vitess/issues/17629.
+func TestConcurrentUpdates(t *testing.T) {
+	ctx := utils.LeakCheckContext(t)
+	// reset error counters
+	hcErrorCounters.ResetAll()
+	ts := memorytopo.NewServer(ctx, "cell")
+	defer ts.Close()
+	hc := createTestHc(ctx, ts)
+	// close healthcheck
+	defer hc.Close()
+
+	// Subscribe to the healthcheck
+	// Make the receiver keep track of the updates received.
+	ch := hc.Subscribe()
+	var totalCount atomic.Int32
+	go func() {
+		for range ch {
+			totalCount.Add(1)
+			// Simulate a somewhat slow consumer.
+			time.Sleep(100 * time.Millisecond)
+		}
+	}()
+
+	// Run multiple updates really quickly
+	// one after the other.
+	totalUpdates := 10
+	for i := 0; i < totalUpdates; i++ {
+		hc.broadcast(&TabletHealth{})
+	}
+	// Unsubscribe from the healthcheck
+	// and verify we process all the updates eventually.
+	hc.Unsubscribe(ch)
+	defer close(ch)
+	require.Eventuallyf(t, func() bool {
+		return totalUpdates == int(totalCount.Load())
+	}, 5*time.Second, 100*time.Millisecond, "expected all updates to be processed")
+}
+
 func tabletDialer(ctx context.Context, tablet *topodatapb.Tablet, _ grpcclient.FailFast) (queryservice.QueryService, error) {
 	connMapMu.Lock()
 	defer connMapMu.Unlock()