From 29cd4e3af108ec421aaad02774c1ca1c21811bea Mon Sep 17 00:00:00 2001
From: Zexi Li <zexi.li@icloud.com>
Date: Thu, 20 Jul 2023 11:51:40 +0800
Subject: [PATCH] fix(monitor): return error if metric is nil (#17575)

---
 pkg/monitor/controller/balancer/balancer.go | 13 ++++++++++++-
 pkg/monitor/controller/balancer/cpu.go      |  3 +++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/pkg/monitor/controller/balancer/balancer.go b/pkg/monitor/controller/balancer/balancer.go
index 4e7d664948..c42fb324b9 100644
--- a/pkg/monitor/controller/balancer/balancer.go
+++ b/pkg/monitor/controller/balancer/balancer.go
@@ -266,6 +266,8 @@ func findGuestsOfHost(drv IMetricDriver, host IHost, ds *tsdb.DataSource, ms *mo
 	}
 
 	ret := make([]ICandidate, 0)
+	found := false
+	errs := []error{}
 	for _, obj := range objs {
 		gHostId, err := obj.GetString("host_id")
 		if err != nil {
@@ -291,15 +293,24 @@ func findGuestsOfHost(drv IMetricDriver, host IHost, ds *tsdb.DataSource, ms *mo
 			}
 			c, err := drv.GetCandidate(obj, host, ds)
 			if err != nil {
-				return nil, errors.Wrapf(err, "drv.GetCandidate of guest %s", obj)
+				errs = append(errs, errors.Wrapf(err, "drv.GetCandidate of guest %s", obj))
+				continue
+
 			}
 			if c.GetScore() == 0 {
 				log.Debugf("ignore guest %s cause %s score is 0", c.GetName(), drv.GetType())
 				continue
 			}
 			ret = append(ret, c)
+			found = true
 		}
 	}
+	if !found {
+		return nil, errors.NewAggregate(errs)
+	}
+	if len(errs) != 0 {
+		log.Warningf("not all guests found: %s", errors.NewAggregate(errs))
+	}
 	return ret, nil
 }
 
diff --git a/pkg/monitor/controller/balancer/cpu.go b/pkg/monitor/controller/balancer/cpu.go
index b732bbc889..c1cfaf1d86 100644
--- a/pkg/monitor/controller/balancer/cpu.go
+++ b/pkg/monitor/controller/balancer/cpu.go
@@ -127,6 +127,9 @@ func newCPUCandidate(gst jsonutils.JSONObject, host *HostResource, ds *tsdb.Data
 	}
 
 	metric := metrics.Get(res.GetId())
+	if metric == nil {
+		return nil, errors.Errorf("not found resource %q metric from %#v", res.GetId(), metrics.indexes)
+	}
 	usage := metric.Values["usage_active"]
 	return &cpuCandidate{
 		guestResource: res,