mirror of
https://git.selfprivacy.org/SelfPrivacy/selfprivacy-rest-api.git
synced 2024-11-18 00:19:15 +00:00
dettlaff
391e4802b2
Co-authored-by: nhnn <nhnn@disroot.org> Co-authored-by: Inex Code <inex.code@selfprivacy.org> Reviewed-on: https://git.selfprivacy.org/SelfPrivacy/selfprivacy-rest-api/pulls/140 Reviewed-by: Inex Code <inex.code@selfprivacy.org> Co-authored-by: dettlaff <dettlaff@riseup.net> Co-committed-by: dettlaff <dettlaff@riseup.net>
430 lines
14 KiB
Python
430 lines
14 KiB
Python
"""Prometheus monitoring queries."""
|
|
|
|
# pylint: disable=too-few-public-methods
|
|
import requests
|
|
|
|
import strawberry
|
|
|
|
from dataclasses import dataclass
|
|
from typing import Optional, Annotated, Union, List, Tuple
|
|
from datetime import datetime, timedelta
|
|
|
|
PROMETHEUS_URL = "http://localhost:9001"
|
|
|
|
|
|
@strawberry.type
|
|
@dataclass
|
|
class MonitoringValue:
|
|
timestamp: datetime
|
|
value: str
|
|
|
|
|
|
@strawberry.type
|
|
@dataclass
|
|
class MonitoringMetric:
|
|
metric_id: str
|
|
values: List[MonitoringValue]
|
|
|
|
|
|
@strawberry.type
|
|
class MonitoringQueryError:
|
|
error: str
|
|
|
|
|
|
@strawberry.type
|
|
class MonitoringValues:
|
|
values: List[MonitoringValue]
|
|
|
|
|
|
@strawberry.type
|
|
class MonitoringMetrics:
|
|
metrics: List[MonitoringMetric]
|
|
|
|
|
|
MonitoringValuesResult = Annotated[
|
|
Union[MonitoringValues, MonitoringQueryError],
|
|
strawberry.union("MonitoringValuesResult"),
|
|
]
|
|
|
|
|
|
MonitoringMetricsResult = Annotated[
|
|
Union[MonitoringMetrics, MonitoringQueryError],
|
|
strawberry.union("MonitoringMetricsResult"),
|
|
]
|
|
|
|
|
|
class MonitoringQueries:
|
|
@staticmethod
|
|
def _send_range_query(
|
|
query: str, start: int, end: int, step: int, result_type: Optional[str] = None
|
|
) -> Union[dict, MonitoringQueryError]:
|
|
try:
|
|
response = requests.get(
|
|
f"{PROMETHEUS_URL}/api/v1/query_range",
|
|
params={
|
|
"query": query,
|
|
"start": start,
|
|
"end": end,
|
|
"step": step,
|
|
},
|
|
timeout=0.8,
|
|
)
|
|
if response.status_code != 200:
|
|
return MonitoringQueryError(
|
|
error=f"Prometheus returned unexpected HTTP status code. Error: {response.text}. The query was {query}"
|
|
)
|
|
json = response.json()
|
|
if result_type and json["data"]["resultType"] != result_type:
|
|
return MonitoringQueryError(
|
|
error="Unexpected resultType returned from Prometheus, request failed"
|
|
)
|
|
return json["data"]
|
|
except Exception as error:
|
|
return MonitoringQueryError(
|
|
error=f"Prometheus request failed! Error: {str(error)}"
|
|
)
|
|
|
|
@staticmethod
|
|
def _send_query(
|
|
query: str, result_type: Optional[str] = None
|
|
) -> Union[dict, MonitoringQueryError]:
|
|
try:
|
|
response = requests.get(
|
|
f"{PROMETHEUS_URL}/api/v1/query",
|
|
params={
|
|
"query": query,
|
|
},
|
|
timeout=0.8,
|
|
)
|
|
if response.status_code != 200:
|
|
return MonitoringQueryError(
|
|
error=f"Prometheus returned unexpected HTTP status code. Error: {response.text}. The query was {query}"
|
|
)
|
|
json = response.json()
|
|
if result_type and json["data"]["resultType"] != result_type:
|
|
return MonitoringQueryError(
|
|
error="Unexpected resultType returned from Prometheus, request failed"
|
|
)
|
|
return json["data"]
|
|
except Exception as error:
|
|
return MonitoringQueryError(
|
|
error=f"Prometheus request failed! Error: {str(error)}"
|
|
)
|
|
|
|
@staticmethod
|
|
def _prometheus_value_to_monitoring_value(x: Tuple[int, str]):
|
|
return MonitoringValue(timestamp=datetime.fromtimestamp(x[0]), value=x[1])
|
|
|
|
@staticmethod
|
|
def _clean_slice_id(slice_id: str, clean_id: bool) -> str:
|
|
"""Slices come in form of `/slice_name.slice`, we need to remove the `.slice` and `/` part."""
|
|
if clean_id:
|
|
parts = slice_id.split(".")[0].split("/")
|
|
if len(parts) > 1:
|
|
return parts[1]
|
|
else:
|
|
raise ValueError(f"Incorrect format slice_id: {slice_id}")
|
|
return slice_id
|
|
|
|
@staticmethod
|
|
def _prometheus_response_to_monitoring_metrics(
|
|
response: dict, id_key: str, clean_id: bool = False
|
|
) -> List[MonitoringMetric]:
|
|
if response["resultType"] == "vector":
|
|
return list(
|
|
map(
|
|
lambda x: MonitoringMetric(
|
|
metric_id=MonitoringQueries._clean_slice_id(
|
|
x["metric"].get(id_key, "/unknown.slice"),
|
|
clean_id=clean_id,
|
|
),
|
|
values=[
|
|
MonitoringQueries._prometheus_value_to_monitoring_value(
|
|
x["value"]
|
|
)
|
|
],
|
|
),
|
|
response["result"],
|
|
)
|
|
)
|
|
else:
|
|
return list(
|
|
map(
|
|
lambda x: MonitoringMetric(
|
|
metric_id=MonitoringQueries._clean_slice_id(
|
|
x["metric"].get(id_key, "/unknown.slice"), clean_id=clean_id
|
|
),
|
|
values=list(
|
|
map(
|
|
MonitoringQueries._prometheus_value_to_monitoring_value,
|
|
x["values"],
|
|
)
|
|
),
|
|
),
|
|
response["result"],
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def _calculate_offset_and_duration(
|
|
start: datetime, end: datetime
|
|
) -> Tuple[int, int]:
|
|
"""Calculate the offset and duration for Prometheus queries.
|
|
They mast be in seconds.
|
|
"""
|
|
offset = int((datetime.now() - end).total_seconds())
|
|
duration = int((end - start).total_seconds())
|
|
return offset, duration
|
|
|
|
@staticmethod
|
|
def cpu_usage_overall(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
step: int = 60, # seconds
|
|
) -> MonitoringValuesResult:
|
|
"""
|
|
Get CPU information.
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
step (int): Interval in seconds for querying disk usage data.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
start_timestamp = int(start.timestamp())
|
|
end_timestamp = int(end.timestamp())
|
|
|
|
query = '100 - (avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100)'
|
|
|
|
data = MonitoringQueries._send_range_query(
|
|
query, start_timestamp, end_timestamp, step, result_type="matrix"
|
|
)
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringValues(
|
|
values=list(
|
|
map(
|
|
MonitoringQueries._prometheus_value_to_monitoring_value,
|
|
data["result"][0]["values"],
|
|
)
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def memory_usage_overall(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
step: int = 60, # seconds
|
|
) -> MonitoringValuesResult:
|
|
"""
|
|
Get memory usage.
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
step (int): Interval in seconds for querying memory usage data.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
start_timestamp = int(start.timestamp())
|
|
end_timestamp = int(end.timestamp())
|
|
|
|
query = "100 - (100 * (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes))"
|
|
|
|
data = MonitoringQueries._send_range_query(
|
|
query, start_timestamp, end_timestamp, step, result_type="matrix"
|
|
)
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringValues(
|
|
values=list(
|
|
map(
|
|
MonitoringQueries._prometheus_value_to_monitoring_value,
|
|
data["result"][0]["values"],
|
|
)
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def memory_usage_max_by_slice(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
) -> MonitoringMetricsResult:
|
|
"""
|
|
Get maximum memory usage for each service (i.e. systemd slice).
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
offset, duration = MonitoringQueries._calculate_offset_and_duration(start, end)
|
|
|
|
if offset == 0:
|
|
query = f'max_over_time((container_memory_rss{{id!~".*slice.*slice", id=~".*slice"}}+container_memory_swap{{id!~".*slice.*slice", id=~".*slice"}})[{duration}s:])'
|
|
else:
|
|
query = f'max_over_time((container_memory_rss{{id!~".*slice.*slice", id=~".*slice"}}+container_memory_swap{{id!~".*slice.*slice", id=~".*slice"}})[{duration}s:] offset {offset}s)'
|
|
|
|
data = MonitoringQueries._send_query(query, result_type="vector")
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringMetrics(
|
|
metrics=MonitoringQueries._prometheus_response_to_monitoring_metrics(
|
|
data, "id", clean_id=True
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def memory_usage_average_by_slice(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
) -> MonitoringMetricsResult:
|
|
"""
|
|
Get average memory usage for each service (i.e. systemd slice).
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
offset, duration = MonitoringQueries._calculate_offset_and_duration(start, end)
|
|
|
|
if offset == 0:
|
|
query = f'avg_over_time((container_memory_rss{{id!~".*slice.*slice", id=~".*slice"}}+container_memory_swap{{id!~".*slice.*slice", id=~".*slice"}})[{duration}s:])'
|
|
else:
|
|
query = f'avg_over_time((container_memory_rss{{id!~".*slice.*slice", id=~".*slice"}}+container_memory_swap{{id!~".*slice.*slice", id=~".*slice"}})[{duration}s:] offset {offset}s)'
|
|
|
|
data = MonitoringQueries._send_query(query, result_type="vector")
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringMetrics(
|
|
metrics=MonitoringQueries._prometheus_response_to_monitoring_metrics(
|
|
data, "id", clean_id=True
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def disk_usage_overall(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
step: int = 60, # seconds
|
|
) -> MonitoringMetricsResult:
|
|
"""
|
|
Get disk usage information.
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
step (int): Interval in seconds for querying disk usage data.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
start_timestamp = int(start.timestamp())
|
|
end_timestamp = int(end.timestamp())
|
|
|
|
query = """100 - (100 * sum by (device) (node_filesystem_avail_bytes{fstype!="rootfs",fstype!="ramfs",fstype!="tmpfs",mountpoint!="/efi"}) / sum by (device) (node_filesystem_size_bytes{fstype!="rootfs",fstype!="ramfs",fstype!="tmpfs",mountpoint!="/efi"}))"""
|
|
|
|
data = MonitoringQueries._send_range_query(
|
|
query, start_timestamp, end_timestamp, step, result_type="matrix"
|
|
)
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringMetrics(
|
|
metrics=MonitoringQueries._prometheus_response_to_monitoring_metrics(
|
|
data, "device"
|
|
)
|
|
)
|
|
|
|
@staticmethod
|
|
def network_usage_overall(
|
|
start: Optional[datetime] = None,
|
|
end: Optional[datetime] = None,
|
|
step: int = 60, # seconds
|
|
) -> MonitoringMetricsResult:
|
|
"""
|
|
Get network usage information for both download and upload.
|
|
|
|
Args:
|
|
start (datetime, optional): The start time.
|
|
Defaults to 20 minutes ago if not provided.
|
|
end (datetime, optional): The end time.
|
|
Defaults to current time if not provided.
|
|
step (int): Interval in seconds for querying network data.
|
|
"""
|
|
|
|
if start is None:
|
|
start = datetime.now() - timedelta(minutes=20)
|
|
|
|
if end is None:
|
|
end = datetime.now()
|
|
|
|
start_timestamp = int(start.timestamp())
|
|
end_timestamp = int(end.timestamp())
|
|
|
|
query = """
|
|
label_replace(rate(node_network_receive_bytes_total{device!="lo"}[5m]), "direction", "receive", "device", ".*")
|
|
or
|
|
label_replace(rate(node_network_transmit_bytes_total{device!="lo"}[5m]), "direction", "transmit", "device", ".*")
|
|
"""
|
|
|
|
data = MonitoringQueries._send_range_query(
|
|
query, start_timestamp, end_timestamp, step, result_type="matrix"
|
|
)
|
|
|
|
if isinstance(data, MonitoringQueryError):
|
|
return data
|
|
|
|
return MonitoringMetrics(
|
|
metrics=MonitoringQueries._prometheus_response_to_monitoring_metrics(
|
|
data, "direction"
|
|
)
|
|
)
|