diff --git a/metrics/alertmanager/pd.rules.yml b/metrics/alertmanager/pd.rules.yml
index b746f94db95..680df4966d8 100644
--- a/metrics/alertmanager/pd.rules.yml
+++ b/metrics/alertmanager/pd.rules.yml
@@ -153,6 +153,20 @@ groups:
       value: '{{ $value }}'
       summary: PD leader lease dropped without failover
 
+  - alert: PD_leader_service_stuck
+    expr: |
+      (service_member_role{job="pd",service="PD"} == 0)
+      and on(instance,job) (etcd_server_is_leader{job="pd"} == 1)
+    for: 1m
+    labels:
+      env: ENV_LABELS_ENV
+      level: critical
+      expr: (service_member_role{job="pd",service="PD"} == 0) and on(instance,job) (etcd_server_is_leader{job="pd"} == 1)
+    annotations:
+      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, PD service is not the PD leader while being the embedded etcd leader; values:{{ $value }}'
+      value: '{{ $value }}'
+      summary: PD leader service is stuck in non-leader state
+
   - alert: PD_cluster_store_space_used_more_than_80%
     expr: sum(pd_cluster_status{type="storage_size"}) / sum(pd_cluster_status{type="storage_capacity"}) * 100  > 80
     for: 1m
diff --git a/tests/alertmanager/pd.rules.test.yml b/tests/alertmanager/pd.rules.test.yml
index cfd87ff4ac7..820e83bb7d0 100644
--- a/tests/alertmanager/pd.rules.test.yml
+++ b/tests/alertmanager/pd.rules.test.yml
@@ -114,3 +114,112 @@ tests:
       - eval_time: 12m
         alertname: PD_leader_lease_drop_without_failover
         exp_alerts: []
+
+  - interval: 15s
+    name: pd-leader-service-stuck
+    input_series:
+      # PD service layer drops at minute 3 and never recovers.
+      - series: 'service_member_role{job="pd",service="PD",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
+
+      # Embedded etcd leader stays stable on pd-1 throughout.
+      - series: 'etcd_server_is_leader{job="pd",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1'
+
+    alert_rule_test:
+      # Condition first true at 3m; for:1m means PENDING until 4m.
+      - eval_time: 3m45s
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 4m
+        alertname: PD_leader_service_stuck
+        exp_alerts:
+          - exp_labels:
+              env: ENV_LABELS_ENV
+              level: critical
+              job: pd
+              service: PD
+              instance: pd-1
+              expr: '(service_member_role{job="pd",service="PD"} == 0) and on(instance,job) (etcd_server_is_leader{job="pd"} == 1)'
+            exp_annotations:
+              summary: 'PD leader service is stuck in non-leader state'
+              description: 'cluster: ENV_LABELS_ENV, instance: pd-1, PD service is not the PD leader while being the embedded etcd leader; values:0'
+              value: '0'
+      - eval_time: 6m
+        alertname: PD_leader_service_stuck
+        exp_alerts:
+          - exp_labels:
+              env: ENV_LABELS_ENV
+              level: critical
+              job: pd
+              service: PD
+              instance: pd-1
+              expr: '(service_member_role{job="pd",service="PD"} == 0) and on(instance,job) (etcd_server_is_leader{job="pd"} == 1)'
+            exp_annotations:
+              summary: 'PD leader service is stuck in non-leader state'
+              description: 'cluster: ENV_LABELS_ENV, instance: pd-1, PD service is not the PD leader while being the embedded etcd leader; values:0'
+              value: '0'
+
+  - interval: 15s
+    name: pd-leader-service-stuck-suppressed-by-failover
+    input_series:
+      # Normal failover: pd-1 loses both PD and etcd leadership at minute 5.
+      - series: 'service_member_role{job="pd",service="PD",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
+      - series: 'etcd_server_is_leader{job="pd",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
+
+      # pd-2 takes over etcd leadership; service_member_role is absent on pd-2
+      # before it wins PD leadership (the metric is only initialized on first win).
+      - series: 'etcd_server_is_leader{job="pd",instance="pd-2"}'
+        values: '0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1'
+
+    alert_rule_test:
+      # Both roles transfer simultaneously; condition is immediately false after minute 5.
+      - eval_time: 5m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 5m30s
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 6m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 12m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+
+  - interval: 15s
+    name: pd-leader-service-stuck-staggered-failover
+    input_series:
+      # pd-1 PD service drops at minute 5 (30s before etcd leadership transfers).
+      - series: 'service_member_role{job="pd",service="PD",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
+      # pd-1 etcd leadership transfers 30s later at minute 5:30.
+      - series: 'etcd_server_is_leader{job="pd",instance="pd-1"}'
+        values: '1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
+
+      # pd-2 takes over both roles at minute 5:30; service_member_role is absent
+      # on pd-2 before it wins PD leadership.
+      - series: 'etcd_server_is_leader{job="pd",instance="pd-2"}'
+        values: '0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1'
+
+    alert_rule_test:
+      # During 5:00–5:15 pd-1 has service=0 and etcd=1 (transient 30s window).
+      # for:1m requires 4 consecutive evals (4×15s); only 2 pass before etcd flips.
+      - eval_time: 5m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 5m15s
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      # At 5:30 pd-1 etcd drops; transient was too short to satisfy for:1m.
+      - eval_time: 5m30s
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 6m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []
+      - eval_time: 12m
+        alertname: PD_leader_service_stuck
+        exp_alerts: []