mirror of
https://github.com/rocky-linux/infrastructure
synced 2024-12-22 19:08:30 +00:00
63 lines
1.9 KiB
Python
63 lines
1.9 KiB
Python
|
#!/usr/bin/env python
|
||
|
|
||
|
from diagrams import Diagram, Cluster, Edge
|
||
|
from diagrams.aws.compute import EC2
|
||
|
from diagrams.aws.general import General
|
||
|
from diagrams.aws.network import ELB
|
||
|
from diagrams.onprem.compute import Server
|
||
|
from diagrams.onprem.iac import Ansible
|
||
|
from diagrams.onprem.monitoring import Grafana, Prometheus
|
||
|
from diagrams.saas.alerting import Pushover
|
||
|
from diagrams.saas.chat import Slack
|
||
|
|
||
|
graph_attr = {
|
||
|
}
|
||
|
|
||
|
node_attr = {
|
||
|
}
|
||
|
|
||
|
with Diagram("Prometheus MVP",
|
||
|
show=False,
|
||
|
direction="TB",
|
||
|
outformat="png",
|
||
|
graph_attr=graph_attr,
|
||
|
node_attr=node_attr):
|
||
|
|
||
|
with Cluster("Rocky VPC"):
|
||
|
with Cluster("AWS services"):
|
||
|
aws_group = [
|
||
|
EC2("service01"),
|
||
|
EC2("service02"),
|
||
|
]
|
||
|
with Cluster("metrics host"):
|
||
|
metrics = Prometheus("metrics")
|
||
|
alertmanager = Prometheus("alertmanager")
|
||
|
dashboard = Grafana("monitoring")
|
||
|
metrics << dashboard
|
||
|
metrics >> alertmanager
|
||
|
|
||
|
Ansible("ansible") >> metrics
|
||
|
metrics >> Edge(style="dashed",
|
||
|
label="ec2 read permissions") >> General("AWS API")
|
||
|
|
||
|
alertmanager >> Edge(style="dashed",
|
||
|
label="non-critical") >> Slack("rocky-alerts")
|
||
|
alertmanager >> Edge(style="dashed",
|
||
|
label="critical") >> Pushover("tbd")
|
||
|
ELB("metrics.rockylinux.org") >> Edge(label="TCP3000") >> dashboard
|
||
|
with Cluster("Cloudvider"):
|
||
|
cloudvider_group = [
|
||
|
Server("server01"),
|
||
|
Server("server02"),
|
||
|
]
|
||
|
|
||
|
with Cluster("Spry Servers"):
|
||
|
spry_group = [
|
||
|
Server("server01"),
|
||
|
Server("server02"),
|
||
|
]
|
||
|
|
||
|
metrics >> aws_group
|
||
|
metrics >> spry_group
|
||
|
metrics >> cloudvider_group
|