You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
jumpserver/apps/terminal/utils/components.py

177 lines
6.8 KiB

# -*- coding: utf-8 -*-
#
from itertools import groupby
from common.utils import get_logger
from terminal.const import ComponentLoad, TerminalType
logger = get_logger(__name__)
class ComputeLoadUtil:
# system status
@staticmethod
def _common_compute_system_status(value, thresholds):
if thresholds[0] <= value <= thresholds[1]:
return ComponentLoad.normal.value
elif thresholds[1] < value <= thresholds[2]:
return ComponentLoad.high.value
else:
return ComponentLoad.critical.value
@classmethod
def _compute_system_stat_status(cls, stat):
system_stat_thresholds_mapper = {
'cpu_load': [0, 5, 20],
'memory_used': [0, 85, 95],
'disk_used': [0, 80, 99]
}
system_status = {}
for stat_key, thresholds in system_stat_thresholds_mapper.items():
stat_value = getattr(stat, stat_key)
if stat_value is None:
msg = 'stat: {}, stat_key: {}, stat_value: {}'
logger.debug(msg.format(stat, stat_key, stat_value))
stat_value = 0
status = cls._common_compute_system_status(stat_value, thresholds)
system_status[stat_key] = status
return system_status
@classmethod
def compute_load(cls, stat, terminal_type=None):
if not stat:
# TODO The core component and celery component will return true for the time being.
if terminal_type in [TerminalType.core, TerminalType.celery]:
return ComponentLoad.normal
else:
return ComponentLoad.offline
system_status_values = cls._compute_system_stat_status(stat).values()
if ComponentLoad.critical in system_status_values:
return ComponentLoad.critical
elif ComponentLoad.high in system_status_values:
return ComponentLoad.high
else:
return ComponentLoad.normal
class TypedComponentsStatusMetricsUtil(object):
def __init__(self):
self.components = []
self.grouped_components = []
self.get_components()
def get_components(self):
from ..models import Terminal
components = Terminal.objects.filter(is_deleted=False).order_by('type')
grouped_components = groupby(components, lambda c: c.type)
grouped_components = [(i[0], list(i[1])) for i in grouped_components]
self.grouped_components = grouped_components
self.components = components
def get_metrics(self):
metrics = []
for _tp, components in self.grouped_components:
metric = {
'total': 0,
'type': _tp,
'session_active': 0,
ComponentLoad.high: [],
ComponentLoad.normal: [],
ComponentLoad.offline: [],
ComponentLoad.critical: [],
}
for component in components:
metric[component.load].append(component.name)
metric['total'] += 1
metric['session_active'] += component.get_online_session_count()
metrics.append(metric)
return metrics
class ComponentsPrometheusMetricsUtil(TypedComponentsStatusMetricsUtil):
def __init__(self):
super().__init__()
self.metrics = self.get_metrics()
@staticmethod
def convert_status_metrics(metrics):
return {
'any': metrics['total'],
'normal': metrics['normal'],
'high': metrics['high'],
'critical': metrics['critical'],
'offline': metrics['offline']
}
def get_component_status_metrics(self):
prometheus_metrics = list()
# 各组件状态个数汇总
prometheus_metrics.append('# JumpServer 各组件状态个数汇总')
status_metric_text = 'jumpserver_components_status_total{component_type="%s", status="%s"} %s'
for metric in self.metrics:
tp = metric['type']
prometheus_metrics.append(f'## 组件: {tp}')
status_metrics = self.convert_status_metrics(metric)
for status, value in status_metrics.items():
metric_text = status_metric_text % (tp, status, value)
prometheus_metrics.append(metric_text)
return prometheus_metrics
def get_component_session_metrics(self):
prometheus_metrics = list()
# 各组件在线会话数汇总
prometheus_metrics.append('# JumpServer 各组件在线会话数汇总')
session_active_metric_text = 'jumpserver_components_session_active_total{component_type="%s"} %s'
for metric in self.metrics:
tp = metric['type']
prometheus_metrics.append(f'## 组件: {tp}')
metric_text = session_active_metric_text % (tp, metric['session_active'])
prometheus_metrics.append(metric_text)
return prometheus_metrics
def get_component_stat_metrics(self):
prometheus_metrics = list()
# 各组件节点指标
prometheus_metrics.append('# JumpServer 各组件一些指标')
state_metric_text = 'jumpserver_components_%s{component_type="%s", component="%s"} %s'
stats_key = [
'cpu_load', 'memory_used', 'disk_used', 'session_online'
]
old_stats_key = [
'system_cpu_load_1', 'system_memory_used_percent',
'system_disk_used_percent', 'session_active_count'
]
old_stats_key_mapper = dict(zip(stats_key, old_stats_key))
for stat_key in stats_key:
prometheus_metrics.append(f'## 指标: {stat_key}')
for component in self.components:
if not component.is_alive:
continue
component_stat = component.last_stat
if not component_stat:
continue
metric_text = state_metric_text % (
stat_key, component.type, component.name, getattr(component_stat, stat_key)
)
prometheus_metrics.append(metric_text)
old_stat_key = old_stats_key_mapper.get(stat_key)
old_metric_text = state_metric_text % (
old_stat_key, component.type, component.name, getattr(component_stat, stat_key)
)
prometheus_metrics.append(old_metric_text)
return prometheus_metrics
def get_prometheus_metrics_text(self):
prometheus_metrics = list()
for method in [
self.get_component_status_metrics,
self.get_component_session_metrics,
self.get_component_stat_metrics
]:
prometheus_metrics.extend(method())
prometheus_metrics.append('\n')
prometheus_metrics_text = '\n'.join(prometheus_metrics)
return prometheus_metrics_text