balance: fix TiDB regionMiss backoff causes TiProxy to incorrectly migrate connections (#832) (#834)

Co-authored-by: djshow832 <[email protected]>

Author: ti-chi-bot

pkg/balance/factor/factor_balance.go

@@ -279,18 +279,17 @@ func (fbb *FactorBasedBalance) BackendsToBalance(backends []policy.BackendCtx) (
 			if score1 >= score2 {
 				// The factors with higher priorities are ordered, so this factor shouldn't violate them.
 				// E.g. if the CPU usage of A is higher than B, don't migrate from B to A even if A is preferred in location.
-				var advice BalanceAdvice
-				var fields []zap.Field
-				advice, balanceCount, fields = factor.BalanceCount(scoredBackends[i], scoredBackends[0])
+				advice, count, fields := factor.BalanceCount(scoredBackends[i], scoredBackends[0])
 				if advice == AdviceNegtive {
 					// If the factor will be unbalanced after migration, skip the rest factors.
 					// E.g. if the CPU usage of A will be much higher than B after migration,
 					// don't migrate from B to A even if A is preferred in location.
 					break
 				}
 				logFields = append(logFields, fields...)
-				if score1 > score2 && advice == AdvicePositive && balanceCount > 0.0001 {
+				if score1 > score2 && advice == AdvicePositive && count > 0.0001 {
 					from, to = scoredBackends[i].BackendCtx, scoredBackends[0].BackendCtx
+					balanceCount = count
 					reason = factor.Name()
 					logFields = append(logFields, zap.String("factor", reason),
 						zap.String("from_total_score", strconv.FormatUint(scoredBackends[i].scoreBits, 16)),

pkg/balance/factor/factor_cpu.go

@@ -41,6 +41,12 @@ var (
 		Names:     []string{"process_cpu_seconds_total", "tidb_server_maxprocs"},
 		Retention: 1 * time.Minute,
 		Metric2Value: func(mfs map[string]*dto.MetricFamily) model.SampleValue {
+			if mfs["process_cpu_seconds_total"] == nil || len(mfs["process_cpu_seconds_total"].Metric) == 0 {
+				return model.SampleValue(math.NaN())
+			}
+			if mfs["tidb_server_maxprocs"] == nil || len(mfs["tidb_server_maxprocs"].Metric) == 0 {
+				return model.SampleValue(math.NaN())
+			}
 			cpuTotal := mfs["process_cpu_seconds_total"].Metric[0].Untyped
 			maxProcs := mfs["tidb_server_maxprocs"].Metric[0].Untyped
 			if cpuTotal == nil || maxProcs == nil {