alarms: refactor the alarms subsystem
also include eotk alarms now
This commit is contained in:
parent
a935055083
commit
e2ce24bf3b
17 changed files with 288 additions and 152 deletions
54
app/terraform/alarms/eotk_aws.py
Normal file
54
app/terraform/alarms/eotk_aws.py
Normal file
|
@ -0,0 +1,54 @@
|
|||
from typing import Tuple, Optional
|
||||
|
||||
import boto3
|
||||
from sqlalchemy import func
|
||||
|
||||
from app import app
|
||||
from app.alarms import get_or_create_alarm
|
||||
from app.extensions import db
|
||||
from app.models.base import Group
|
||||
from app.models.alarms import AlarmState
|
||||
from app.models.onions import Eotk
|
||||
from app.terraform import BaseAutomation
|
||||
|
||||
|
||||
def alarms_in_region(region: str, prefix: str, aspect: str) -> None:
|
||||
cloudwatch = boto3.client('cloudwatch',
|
||||
aws_access_key_id=app.config['AWS_ACCESS_KEY'],
|
||||
aws_secret_access_key=app.config['AWS_SECRET_KEY'],
|
||||
region_name=region)
|
||||
dist_paginator = cloudwatch.get_paginator('describe_alarms')
|
||||
page_iterator = dist_paginator.paginate(AlarmNamePrefix=prefix)
|
||||
for page in page_iterator:
|
||||
for cw_alarm in page['MetricAlarms']:
|
||||
eotk_id = cw_alarm["AlarmName"][len(prefix):].split("-")
|
||||
group: Optional[Group] = Group.query.filter(func.lower(Group.group_name) == eotk_id[1]).first()
|
||||
if group is None:
|
||||
print("Unable to find group for " + cw_alarm['AlarmName'])
|
||||
continue
|
||||
eotk = Eotk.query.filter(
|
||||
Eotk.group_id == group.id,
|
||||
Eotk.region == region
|
||||
).first()
|
||||
if eotk is None:
|
||||
print("Skipping unknown instance " + cw_alarm['AlarmName'])
|
||||
continue
|
||||
alarm = get_or_create_alarm(eotk.brn, aspect)
|
||||
if cw_alarm['StateValue'] == "OK":
|
||||
alarm.update_state(AlarmState.OK, "CloudWatch alarm OK")
|
||||
elif cw_alarm['StateValue'] == "ALARM":
|
||||
alarm.update_state(AlarmState.CRITICAL, "CloudWatch alarm ALARM")
|
||||
else:
|
||||
alarm.update_state(AlarmState.UNKNOWN, f"CloudWatch alarm {cw_alarm['StateValue']}")
|
||||
|
||||
|
||||
class AlarmEotkAwsAutomation(BaseAutomation):
|
||||
short_name = "monitor_eotk_aws"
|
||||
description = "Import alarms for AWS EOTK instances"
|
||||
|
||||
def automate(self, full: bool = False) -> Tuple[bool, str]:
|
||||
for region in ["us-east-2", "eu-central-1"]:
|
||||
alarms_in_region(region, "eotk-bw-out-high-", "bandwidth-out-high")
|
||||
alarms_in_region(region, "eotk-cpu-high-", "instance-cpu")
|
||||
db.session.commit()
|
||||
return True, ""
|
Loading…
Add table
Add a link
Reference in a new issue