2024-06-15 18:17:08 +00:00
|
|
|
"""Prometheus monitoring queries."""
|
|
|
|
|
|
|
|
# pylint: disable=too-few-public-methods
|
2024-06-22 12:12:29 +00:00
|
|
|
import requests
|
2024-07-10 12:53:56 +00:00
|
|
|
|
2024-07-08 15:00:49 +00:00
|
|
|
import strawberry
|
2024-07-10 12:53:56 +00:00
|
|
|
from strawberry.scalars import JSON
|
|
|
|
|
2024-07-08 15:00:49 +00:00
|
|
|
from dataclasses import dataclass
|
|
|
|
from typing import Optional
|
2024-06-25 17:25:31 +00:00
|
|
|
from datetime import datetime, timedelta
|
2024-06-15 18:17:08 +00:00
|
|
|
|
2024-06-17 18:56:58 +00:00
|
|
|
PROMETHEUS_URL = "http://localhost:9001"
|
2024-06-15 18:17:08 +00:00
|
|
|
|
|
|
|
|
2024-07-08 15:00:49 +00:00
|
|
|
@strawberry.type
|
|
|
|
@dataclass
|
|
|
|
class PrometheusQueryResult:
|
|
|
|
result_type: str
|
|
|
|
result: JSON
|
2024-06-15 18:17:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
class PrometheusQueries:
|
|
|
|
@staticmethod
|
2024-07-08 15:00:49 +00:00
|
|
|
def _send_query(query: str, start: int, end: int, step: int):
|
2024-06-15 18:17:08 +00:00
|
|
|
try:
|
2024-07-08 15:00:49 +00:00
|
|
|
response = requests.get(
|
|
|
|
f"{PROMETHEUS_URL}/api/v1/query",
|
|
|
|
params={
|
|
|
|
"query": query,
|
|
|
|
"start": start,
|
|
|
|
"end": end,
|
|
|
|
"step": step,
|
|
|
|
},
|
2024-06-15 18:17:08 +00:00
|
|
|
)
|
2024-07-08 15:00:49 +00:00
|
|
|
if response.status_code != 200:
|
|
|
|
raise Exception("Prometheus returned unexpected HTTP status code")
|
|
|
|
json = response.json()
|
|
|
|
return PrometheusQueryResult(
|
|
|
|
result_type=json.result_type, result=json.result
|
2024-06-15 18:17:08 +00:00
|
|
|
)
|
2024-07-08 15:00:49 +00:00
|
|
|
except Exception as error:
|
|
|
|
raise Exception("Prometheus request failed! " + str(error))
|
2024-06-15 18:17:08 +00:00
|
|
|
|
|
|
|
@staticmethod
|
2024-06-21 16:33:37 +00:00
|
|
|
def cpu_usage(
|
2024-07-07 12:33:15 +00:00
|
|
|
start: Optional[int] = None,
|
|
|
|
end: Optional[int] = None,
|
|
|
|
step: int = 60, # seconds
|
2024-07-08 15:00:49 +00:00
|
|
|
) -> PrometheusQueryResult:
|
|
|
|
"""
|
|
|
|
Get CPU information.
|
2024-07-07 12:33:15 +00:00
|
|
|
|
|
|
|
Args:
|
|
|
|
start (int, optional): Unix timestamp indicating the start time.
|
|
|
|
Defaults to 20 minutes ago if not provided.
|
|
|
|
end (int, optional): Unix timestamp indicating the end time.
|
|
|
|
Defaults to current time if not provided.
|
|
|
|
step (int): Interval in seconds for querying disk usage data.
|
|
|
|
"""
|
|
|
|
|
|
|
|
if not start:
|
|
|
|
start = int((datetime.now() - timedelta(minutes=20)).timestamp())
|
|
|
|
|
|
|
|
if not end:
|
|
|
|
end = int(datetime.now().timestamp())
|
2024-06-21 16:33:37 +00:00
|
|
|
|
2024-06-15 18:17:08 +00:00
|
|
|
query = '100 - (avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100)'
|
|
|
|
|
2024-07-08 15:00:49 +00:00
|
|
|
return PrometheusQueries._send_query(query, start, end, step)
|
2024-07-08 15:18:07 +00:00
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def memory_usage(
|
|
|
|
start: Optional[int] = None,
|
|
|
|
end: Optional[int] = None,
|
|
|
|
step: int = 60, # seconds
|
|
|
|
) -> PrometheusQueryResult:
|
|
|
|
"""
|
|
|
|
Get memory usage.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
start (int, optional): Unix timestamp indicating the start time.
|
|
|
|
Defaults to 20 minutes ago if not provided.
|
|
|
|
end (int, optional): Unix timestamp indicating the end time.
|
|
|
|
Defaults to current time if not provided.
|
|
|
|
step (int): Interval in seconds for querying memory usage data.
|
|
|
|
"""
|
|
|
|
|
|
|
|
if not start:
|
|
|
|
start = int((datetime.now() - timedelta(minutes=20)).timestamp())
|
|
|
|
|
|
|
|
if not end:
|
|
|
|
end = int(datetime.now().timestamp())
|
|
|
|
|
|
|
|
query = "100 - (100 * (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes))"
|
|
|
|
|
|
|
|
return PrometheusQueries._send_query(query, start, end, step)
|
2024-06-15 18:17:08 +00:00
|
|
|
|
|
|
|
@staticmethod
|
2024-06-21 16:33:37 +00:00
|
|
|
def disk_usage(
|
2024-07-07 12:33:15 +00:00
|
|
|
start: Optional[int] = None,
|
|
|
|
end: Optional[int] = None,
|
|
|
|
step: int = 60, # seconds
|
2024-07-08 15:00:49 +00:00
|
|
|
) -> PrometheusQueryResult:
|
2024-07-07 12:33:15 +00:00
|
|
|
"""
|
|
|
|
Get disk usage information.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
start (int, optional): Unix timestamp indicating the start time.
|
|
|
|
Defaults to 20 minutes ago if not provided.
|
|
|
|
end (int, optional): Unix timestamp indicating the end time.
|
|
|
|
Defaults to current time if not provided.
|
|
|
|
step (int): Interval in seconds for querying disk usage data.
|
|
|
|
"""
|
|
|
|
|
|
|
|
if not start:
|
|
|
|
start = int((datetime.now() - timedelta(minutes=20)).timestamp())
|
|
|
|
|
|
|
|
if not end:
|
|
|
|
end = int(datetime.now().timestamp())
|
|
|
|
|
2024-06-16 19:06:40 +00:00
|
|
|
query = '100 - (100 * ((node_filesystem_avail_bytes{mountpoint="/",fstype!="rootfs"} ) / (node_filesystem_size_bytes{mountpoint="/",fstype!="rootfs"}) ))'
|
2024-06-16 19:01:25 +00:00
|
|
|
|
2024-07-08 15:00:49 +00:00
|
|
|
return PrometheusQueries._send_query(query, start, end, step)
|