Arnike/watcher
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Added prometheus instance cpu and ram metrics

Browse Source
This commit is contained in:
Nikolay Tatarinov
2025-10-03 11:57:15 +03:00
parent 03073a1b0d
commit 8eab47e107
2 changed files with 26 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
watcher/decision_engine/datasources/prometheus_base.py
View File

@@ -40,8 +40,8 @@ class PrometheusBase(base.DataSourceBase):
host_inlet_temp=None, host_inlet_temp=None,
host_airflow=None, host_airflow=None,
host_power=None, host_power=None,
instance_cpu_usage='ceilometer_cpu', instance_cpu_usage='libvirt_domain_info_cpu_time_seconds_total',
instance_ram_usage='ceilometer_memory_usage', instance_ram_usage='libvirt_domain_info_memory_usage_bytes',
instance_ram_allocated='instance.memory', instance_ram_allocated='instance.memory',
instance_l3_cache_usage=None, instance_l3_cache_usage=None,
instance_root_disk_size='instance.disk', instance_root_disk_size='instance.disk',
@@ -311,6 +311,18 @@ class PrometheusBase(base.DataSourceBase):
'agg': aggregate, 'meter': meter, 'period': period, 'agg': aggregate, 'meter': meter, 'period': period,
'vcpus': vcpus} 'vcpus': vcpus}
) )
elif meter == 'libvirt_domain_info_cpu_time_seconds_total':
query_args = (
"(rate(%(meter)s{%(label)s='%(label_value)s'}[%(period)ss]))"
% {'label': uuid_label_key, 'label_value': instance_label,
'meter': meter, 'period': period}
)
elif meter == 'libvirt_domain_info_memory_usage_bytes':
query_args = (
"%(agg)s_over_time(%(meter)s{%(label)s='%(label_value)s'}[%(period)ss]) / 1024"
% {'label': uuid_label_key, 'label_value': instance_label,
'meter': meter, 'period': period, 'agg': aggregate}
)
else: else:
raise exception.InvalidParameter( raise exception.InvalidParameter(
message=(_("Cannot process prometheus meter %s") % meter) message=(_("Cannot process prometheus meter %s") % meter)

12
watcher/decision_engine/strategy/strategies/host_maintenance.py
View File

@@ -89,6 +89,14 @@ class HostMaintenance(base.HostMaintenanceBaseStrategy):
"will backup the maintenance node.", "will backup the maintenance node.",
"type": "string", "type": "string",
}, },
{
"disabled_reason": {
"description": "Describe reason for moving host"
"into maintenance mode.",
"type": "string",
"default": False,
}
}
"disable_live_migration": { "disable_live_migration": {
"description": "Disable live migration in maintenance. " "description": "Disable live migration in maintenance. "
"If True, active instances will be cold " "If True, active instances will be cold "
@@ -295,6 +303,10 @@ class HostMaintenance(base.HostMaintenanceBaseStrategy):
maintenance_node = self.input_parameters.get('maintenance_node') maintenance_node = self.input_parameters.get('maintenance_node')
backup_node = self.input_parameters.get('backup_node') backup_node = self.input_parameters.get('backup_node')
disabled_reason = self.input_parameters.get('disabled_reason')
if disabled_reason:
self.REASON_FOR_MAINTAINING = disabled_reason
# if no VMs in the maintenance_node, just maintain the compute node # if no VMs in the maintenance_node, just maintain the compute node
src_node = self.compute_model.get_node_by_name(maintenance_node) src_node = self.compute_model.get_node_by_name(maintenance_node)