Merge pull request #13176 from jessicamack/13019_add_node_type

Add node_type to instance info and capacity metrics
This commit is contained in:
jessicamack 2022-11-15 12:43:53 -05:00 committed by GitHub
commit 0933a96d60
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 13 additions and 5 deletions

View File

@ -238,7 +238,9 @@ def instance_info(since, include_hostnames=False, **kwargs):
info = {} info = {}
# Use same method that the TaskManager does to compute consumed capacity without querying all running jobs for each Instance # Use same method that the TaskManager does to compute consumed capacity without querying all running jobs for each Instance
active_tasks = models.UnifiedJob.objects.filter(status__in=['running', 'waiting']).only('task_impact', 'controller_node', 'execution_node') active_tasks = models.UnifiedJob.objects.filter(status__in=['running', 'waiting']).only('task_impact', 'controller_node', 'execution_node')
tm_instances = TaskManagerInstances(active_tasks, instance_fields=['uuid', 'version', 'capacity', 'cpu', 'memory', 'managed_by_policy', 'enabled']) tm_instances = TaskManagerInstances(
active_tasks, instance_fields=['uuid', 'version', 'capacity', 'cpu', 'memory', 'managed_by_policy', 'enabled', 'node_type']
)
for tm_instance in tm_instances.instances_by_hostname.values(): for tm_instance in tm_instances.instances_by_hostname.values():
instance = tm_instance.obj instance = tm_instance.obj
instance_info = { instance_info = {
@ -251,6 +253,7 @@ def instance_info(since, include_hostnames=False, **kwargs):
'enabled': instance.enabled, 'enabled': instance.enabled,
'consumed_capacity': tm_instance.consumed_capacity, 'consumed_capacity': tm_instance.consumed_capacity,
'remaining_capacity': instance.capacity - tm_instance.consumed_capacity, 'remaining_capacity': instance.capacity - tm_instance.consumed_capacity,
'node_type': instance.node_type,
} }
if include_hostnames is True: if include_hostnames is True:
instance_info['hostname'] = instance.hostname instance_info['hostname'] = instance.hostname

View File

@ -57,6 +57,7 @@ def metrics():
[ [
'hostname', 'hostname',
'instance_uuid', 'instance_uuid',
'node_type',
], ],
registry=REGISTRY, registry=REGISTRY,
) )
@ -84,6 +85,7 @@ def metrics():
[ [
'hostname', 'hostname',
'instance_uuid', 'instance_uuid',
'node_type',
], ],
registry=REGISTRY, registry=REGISTRY,
) )
@ -111,6 +113,7 @@ def metrics():
[ [
'hostname', 'hostname',
'instance_uuid', 'instance_uuid',
'node_type',
], ],
registry=REGISTRY, registry=REGISTRY,
) )
@ -120,6 +123,7 @@ def metrics():
[ [
'hostname', 'hostname',
'instance_uuid', 'instance_uuid',
'node_type',
], ],
registry=REGISTRY, registry=REGISTRY,
) )
@ -180,12 +184,13 @@ def metrics():
instance_data = instance_info(None, include_hostnames=True) instance_data = instance_info(None, include_hostnames=True)
for uuid, info in instance_data.items(): for uuid, info in instance_data.items():
hostname = info['hostname'] hostname = info['hostname']
INSTANCE_CAPACITY.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['capacity']) node_type = info['node_type']
INSTANCE_CAPACITY.labels(hostname=hostname, instance_uuid=uuid, node_type=node_type).set(instance_data[uuid]['capacity'])
INSTANCE_CPU.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['cpu']) INSTANCE_CPU.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['cpu'])
INSTANCE_MEMORY.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['memory']) INSTANCE_MEMORY.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['memory'])
INSTANCE_CONSUMED_CAPACITY.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['consumed_capacity']) INSTANCE_CONSUMED_CAPACITY.labels(hostname=hostname, instance_uuid=uuid, node_type=node_type).set(instance_data[uuid]['consumed_capacity'])
INSTANCE_REMAINING_CAPACITY.labels(hostname=hostname, instance_uuid=uuid).set(instance_data[uuid]['remaining_capacity']) INSTANCE_REMAINING_CAPACITY.labels(hostname=hostname, instance_uuid=uuid, node_type=node_type).set(instance_data[uuid]['remaining_capacity'])
INSTANCE_INFO.labels(hostname=hostname, instance_uuid=uuid).info( INSTANCE_INFO.labels(hostname=hostname, instance_uuid=uuid, node_type=node_type).info(
{ {
'enabled': str(instance_data[uuid]['enabled']), 'enabled': str(instance_data[uuid]['enabled']),
'managed_by_policy': str(instance_data[uuid]['managed_by_policy']), 'managed_by_policy': str(instance_data[uuid]['managed_by_policy']),