Incorrect capacity for remote execution nodes 14051 (#14315)

awx/main/models/ha.py

@@ -309,8 +309,8 @@ class Instance(HasPolicyEditsMixin, BaseModel):
             self.cpu_capacity = 0
             self.mem_capacity = 0  # formula has a non-zero offset, so we make sure it is 0 for hop nodes
         else:
-            self.cpu_capacity = get_cpu_effective_capacity(self.cpu)
+            self.cpu_capacity = get_cpu_effective_capacity(self.cpu, is_control_node=bool(self.node_type in (Instance.Types.CONTROL, Instance.Types.HYBRID)))
-            self.mem_capacity = get_mem_effective_capacity(self.memory)
+            self.mem_capacity = get_mem_effective_capacity(self.memory, is_control_node=bool(self.node_type in (Instance.Types.CONTROL, Instance.Types.HYBRID)))
         self.set_capacity_value()
 
     def save_health_data(self, version=None, cpu=0, memory=0, uuid=None, update_last_seen=False, errors=''):
@@ -333,12 +333,17 @@ class Instance(HasPolicyEditsMixin, BaseModel):
             self.version = version
             update_fields.append('version')
 
-        new_cpu = get_corrected_cpu(cpu)
+        if self.node_type == Instance.Types.EXECUTION:
+            new_cpu = cpu
+            new_memory = memory
+        else:
+            new_cpu = get_corrected_cpu(cpu)
+            new_memory = get_corrected_memory(memory)
+
         if new_cpu != self.cpu:
             self.cpu = new_cpu
             update_fields.append('cpu')
 
-        new_memory = get_corrected_memory(memory)
         if new_memory != self.memory:
             self.memory = new_memory
             update_fields.append('memory')

awx/main/tasks/system.py

@@ -466,7 +466,6 @@ def execution_node_health_check(node):
     data = worker_info(node)
 
     prior_capacity = instance.capacity
-
     instance.save_health_data(
         version='ansible-runner-' + data.get('runner_version', '???'),
         cpu=data.get('cpu_count', 0),

awx/main/tests/functional/test_jobs.py

@@ -38,8 +38,8 @@ def test_orphan_unified_job_creation(instance, inventory):
 
 @pytest.mark.django_db
 @mock.patch('awx.main.tasks.system.inspect_execution_and_hop_nodes', lambda *args, **kwargs: None)
-@mock.patch('awx.main.models.ha.get_cpu_effective_capacity', lambda cpu: 8)
+@mock.patch('awx.main.models.ha.get_cpu_effective_capacity', lambda cpu, is_control_node: 8)
-@mock.patch('awx.main.models.ha.get_mem_effective_capacity', lambda mem: 62)
+@mock.patch('awx.main.models.ha.get_mem_effective_capacity', lambda mem, is_control_node: 62)
 def test_job_capacity_and_with_inactive_node():
     i = Instance.objects.create(hostname='test-1')
     i.save_health_data('18.0.1', 2, 8000)

awx/main/tests/unit/settings/test_k8s_resource_setttings.py

@@ -36,7 +36,9 @@ def test_SYSTEM_TASK_ABS_MEM_conversion(value, converted_value, mem_capacity):
         mock_settings.IS_K8S = True
         assert convert_mem_str_to_bytes(value) == converted_value
         assert get_corrected_memory(-1) == converted_value
-        assert get_mem_effective_capacity(-1) == mem_capacity
+        assert get_mem_effective_capacity(1, is_control_node=True) == mem_capacity
+        # SYSTEM_TASK_ABS_MEM should not effect memory and capacity for execution nodes
+        assert get_mem_effective_capacity(2147483648, is_control_node=False) == 20
 
 
 @pytest.mark.parametrize(
@@ -58,4 +60,6 @@ def test_SYSTEM_TASK_ABS_CPU_conversion(value, converted_value, cpu_capacity):
         mock_settings.SYSTEM_TASK_FORKS_CPU = 4
         assert convert_cpu_str_to_decimal_cpu(value) == converted_value
         assert get_corrected_cpu(-1) == converted_value
-        assert get_cpu_effective_capacity(-1) == cpu_capacity
+        assert get_cpu_effective_capacity(-1, is_control_node=True) == cpu_capacity
+        # SYSTEM_TASK_ABS_CPU should not effect cpu count and capacity for execution nodes
+        assert get_cpu_effective_capacity(2.0, is_control_node=False) == 8

awx/main/utils/common.py

@@ -768,14 +768,13 @@ def get_corrected_cpu(cpu_count):  # formerlly get_cpu_capacity
     return cpu_count  # no correction
 
 
-def get_cpu_effective_capacity(cpu_count):
+def get_cpu_effective_capacity(cpu_count, is_control_node=False):
     from django.conf import settings
 
-    cpu_count = get_corrected_cpu(cpu_count)
-
     settings_forkcpu = getattr(settings, 'SYSTEM_TASK_FORKS_CPU', None)
     env_forkcpu = os.getenv('SYSTEM_TASK_FORKS_CPU', None)
-
+    if is_control_node:
+        cpu_count = get_corrected_cpu(cpu_count)
     if env_forkcpu:
         forkcpu = int(env_forkcpu)
     elif settings_forkcpu:
@@ -834,6 +833,7 @@ def get_corrected_memory(memory):
 
     # Runner returns memory in bytes
     # so we convert memory from settings to bytes as well.
+
     if env_absmem is not None:
         return convert_mem_str_to_bytes(env_absmem)
     elif settings_absmem is not None:
@@ -842,14 +842,13 @@ def get_corrected_memory(memory):
     return memory
 
 
-def get_mem_effective_capacity(mem_bytes):
+def get_mem_effective_capacity(mem_bytes, is_control_node=False):
     from django.conf import settings
 
-    mem_bytes = get_corrected_memory(mem_bytes)
-
     settings_mem_mb_per_fork = getattr(settings, 'SYSTEM_TASK_FORKS_MEM', None)
     env_mem_mb_per_fork = os.getenv('SYSTEM_TASK_FORKS_MEM', None)
-
+    if is_control_node:
+        mem_bytes = get_corrected_memory(mem_bytes)
     if env_mem_mb_per_fork:
         mem_mb_per_fork = int(env_mem_mb_per_fork)
     elif settings_mem_mb_per_fork: