Answer a panicking verb with an error, say flag errors in the answer, and leave refused logins out of D15

Read verbs now run in the serving process, where a panic would end every
call; refused logins are nobody's reconnect loop (review of hq issue 327).
This commit is contained in:
jochen
2026-10-08 21:08:34 +02:00
parent 1e04670052
commit cc7fb99f29
9 changed files with 110 additions and 33 deletions
+11 -1
View File
@@ -64,7 +64,12 @@ func probeReconnects(ctx context.Context, d *doctor) ([]conditions.Observation,
if len(on) == 0 {
return nil, nil // no bus module assigned: a mesh whose bus is not the mesh's module
}
read, err := askClosed(ctx, d.js.Conn(), on[0])
return reconnectsOn(ctx, d.js.Conn(), on[0])
}
// reconnectsOn asks the bus module on its machine and says who reconnects in a loop.
func reconnectsOn(ctx context.Context, conn *nats.Conn, node string) ([]conditions.Observation, error) {
read, err := askClosed(ctx, conn, node)
if isNothingServes(err) {
// A bus module older than its tool has nothing it can say, which is not a failure of the probe.
return nil, nil
@@ -100,6 +105,11 @@ func reconnecting(read closedConnections) []conditions.Observation {
}
var out []conditions.Observation
for _, u := range read.Users {
if u.User == "" || strings.HasPrefix(u.User, "(") {
// Refused before it logged in: no client of the mesh's, so nobody's reconnect loop. A login
// refused again and again is a question of its own, not this probe's.
continue
}
perHour := float64(u.Dropped) / hours
if perHour <= reconnectBound {
continue