diff --git a/watcher/decision_engine/datasources/prometheus.py b/watcher/decision_engine/datasources/prometheus.py
index 9b803b6be..7b7572f3b 100644
--- a/watcher/decision_engine/datasources/prometheus.py
+++ b/watcher/decision_engine/datasources/prometheus.py
@@ -264,8 +264,8 @@ class PrometheusHelper(base.DataSourceBase):
         This function builds and returns the string query that will be sent
         to the Prometheus server /query endpoint. For host cpu usage we use:
 
-        100 - (avg by (instance)(rate(node_cpu_seconds_total{mode='idle',
-                                       instance='some_host'}[300s])) * 100)
+        100 - (avg by (fqdn)(rate(node_cpu_seconds_total{mode='idle',
+                                       fqdn='some_host'}[300s])) * 100)
 
         so using prometheus rate function over the specified period, we average
         per instance (all cpus) idle time and then 'everything else' is cpu
@@ -307,7 +307,7 @@ class PrometheusHelper(base.DataSourceBase):
 
         if meter == 'node_cpu_seconds_total':
             query_args = (
-                "100 - (%(agg)s by (instance)(rate(%(meter)s"
+                "100 - (%(agg)s by (%(label)s)(rate(%(meter)s"
                 "{mode='idle',%(label)s='%(label_value)s'}[%(period)ss])) "
                 "* 100)"
                 % {'label': self.prometheus_fqdn_label,
@@ -464,8 +464,8 @@ class PrometheusHelper(base.DataSourceBase):
         This calculates the host cpu usage and returns it as a percentage
         The calculation is made by using the cpu 'idle' time, per
         instance (so all CPUs are included). For example the query looks like
-        (100 - (avg by (instance)(rate(node_cpu_seconds_total
-            {mode='idle',instance='localhost:9100'}[300s])) * 100))
+        (100 - (avg by (fqdn)(rate(node_cpu_seconds_total
+            {mode='idle',fqdn='compute1.example.com'}[300s])) * 100))
         """
         aggregate = self._invert_max_min_aggregate(aggregate)
         cpu_usage = self.statistic_aggregation(
diff --git a/watcher/tests/decision_engine/datasources/test_prometheus_helper.py b/watcher/tests/decision_engine/datasources/test_prometheus_helper.py
index 1c4b0d662..5aaccb0ec 100644
--- a/watcher/tests/decision_engine/datasources/test_prometheus_helper.py
+++ b/watcher/tests/decision_engine/datasources/test_prometheus_helper.py
@@ -146,7 +146,7 @@ class TestPrometheusHelper(base.BaseTestCase):
         )
         self.assertEqual(expected_cpu_usage, result)
         mock_prometheus_query.assert_called_once_with(
-            "100 - (avg by (instance)(rate(node_cpu_seconds_total"
+            "100 - (avg by (fqdn)(rate(node_cpu_seconds_total"
             "{mode='idle',fqdn='marios-env.controlplane.domain'}[300s]))"
             " * 100)")
 
@@ -575,7 +575,7 @@ class TestPrometheusHelper(base.BaseTestCase):
 
     def test_build_prometheus_query_node_cpu_avg_agg(self):
         expected_query = (
-            "100 - (avg by (instance)(rate(node_cpu_seconds_total"
+            "100 - (avg by (fqdn)(rate(node_cpu_seconds_total"
             "{mode='idle',fqdn='a_host'}[111s])) * 100)")
         result = self.helper._build_prometheus_query(
             'avg', 'node_cpu_seconds_total', 'a_host', '111')
@@ -583,7 +583,7 @@ class TestPrometheusHelper(base.BaseTestCase):
 
     def test_build_prometheus_query_node_cpu_max_agg(self):
         expected_query = (
-            "100 - (max by (instance)(rate(node_cpu_seconds_total"
+            "100 - (max by (fqdn)(rate(node_cpu_seconds_total"
             "{mode='idle',fqdn='b_host'}[444s])) * 100)")
         result = self.helper._build_prometheus_query(
             'max', 'node_cpu_seconds_total', 'b_host', '444')
@@ -610,7 +610,7 @@ class TestPrometheusHelper(base.BaseTestCase):
     def test_build_prometheus_query_node_cpu_avg_agg_custom_label(self):
         self.helper.prometheus_fqdn_label = 'custom_fqdn_label'
         expected_query = (
-            "100 - (avg by (instance)(rate(node_cpu_seconds_total"
+            "100 - (avg by (custom_fqdn_label)(rate(node_cpu_seconds_total"
             "{mode='idle',custom_fqdn_label='a_host'}[111s])) * 100)")
         result = self.helper._build_prometheus_query(
             'avg', 'node_cpu_seconds_total', 'a_host', '111')