forked from 3wordchant/capsul-flask
237 lines
7.0 KiB
Python
237 lines
7.0 KiB
Python
import matplotlib.ticker as ticker
|
|
import matplotlib.pyplot as pyplot
|
|
import matplotlib.dates as mdates
|
|
from functools import reduce
|
|
import requests
|
|
#import json
|
|
from datetime import datetime
|
|
from io import BytesIO
|
|
from flask import Blueprint
|
|
from flask import current_app
|
|
from flask import session
|
|
from flask import render_template, make_response
|
|
from werkzeug.exceptions import abort
|
|
|
|
from capsulflask.db import get_model
|
|
from capsulflask.auth import account_required
|
|
|
|
bp = Blueprint("metrics", __name__, url_prefix="/metrics")
|
|
|
|
durations = dict(
|
|
_5m=[60*5, 15],
|
|
_1h=[60*60, 60],
|
|
_1d=[60*60*24, 60*20],
|
|
_30d=[60*60*24*30, 60*300]
|
|
)
|
|
sizes = dict(
|
|
s=[0.77, 0.23, 4],
|
|
m=[1, 1, 2],
|
|
l=[6, 4, 1],
|
|
)
|
|
|
|
green = (121/255, 240/255, 50/255)
|
|
blue = (70/255, 150/255, 255/255)
|
|
red = (255/255, 50/255, 8/255)
|
|
|
|
@bp.route("/html/<string:metric>/<string:capsulid>/<string:duration>")
|
|
@account_required
|
|
def display_metric(metric, capsulid, duration):
|
|
vm = get_model().get_vm_detail(session["account"], capsulid)
|
|
|
|
return render_template(
|
|
"display-metric.html",
|
|
vm=vm,
|
|
duration=duration,
|
|
durations=list(map(lambda x: x.strip("_"), durations.keys())),
|
|
metric=metric
|
|
)
|
|
|
|
|
|
@bp.route("/<string:metric>/<string:capsulid>/<string:duration>/<string:size>")
|
|
@account_required
|
|
def metric_png(metric, capsulid, duration, size):
|
|
result = get_plot_bytes(metric, capsulid, duration, size)
|
|
|
|
if result[0] != 200:
|
|
abort(result[0])
|
|
|
|
response = make_response(result[1])
|
|
response.headers.set('Content-Type', 'image/png')
|
|
return response
|
|
|
|
|
|
def get_plot_bytes(metric, capsulid, duration, size):
|
|
|
|
duration = f"_{duration}"
|
|
|
|
if duration not in durations:
|
|
return (404, None)
|
|
|
|
if size not in sizes:
|
|
return (404, None)
|
|
|
|
vm = get_model().get_vm_detail(session["account"], capsulid)
|
|
|
|
if not vm:
|
|
return (404, None)
|
|
|
|
now_unix = int(datetime.strftime(datetime.now(), "%s"))
|
|
duration_seconds = durations[duration][0]
|
|
interval_seconds = durations[duration][1] * sizes[size][2]
|
|
if interval_seconds < 30:
|
|
interval_seconds = 30
|
|
|
|
# Prometheus queries to pull metrics for VMs
|
|
metric_queries = dict(
|
|
cpu=f"irate(libvirtd_domain_info_cpu_time_seconds_total{{domain='{capsulid}'}}[30s])",
|
|
memory=f"libvirtd_domain_info_memory_usage_bytes{{domain='{capsulid}'}}",
|
|
network_in=f"rate(libvirtd_domain_interface_stats_receive_bytes_total{{domain='{capsulid}'}}[{interval_seconds}s])",
|
|
network_out=f"rate(libvirtd_domain_interface_stats_transmit_bytes_total{{domain='{capsulid}'}}[{interval_seconds}s])",
|
|
disk=f"rate(libvirtd_domain_block_stats_read_bytes_total{{domain='{capsulid}'}}[{interval_seconds}s])%2Brate(libvirtd_domain_block_stats_write_bytes_total{{domain='{capsulid}'}}[{interval_seconds}s])",
|
|
)
|
|
|
|
# These represent the top of the graph for graphs that are designed to be viewed at a glance.
|
|
# they are also used to colorize the graph at any size.
|
|
scales = dict(
|
|
cpu=vm["vcpus"],
|
|
memory=vm["memory_mb"]*1024*1024,
|
|
network_in=1024*1024*2,
|
|
network_out=1024*200,
|
|
disk=1024*1024*4,
|
|
)
|
|
|
|
if metric not in metric_queries:
|
|
return (404, None)
|
|
|
|
range_and_interval = f"start={now_unix-duration_seconds}&end={now_unix}&step={interval_seconds}"
|
|
|
|
prometheus_range_url = f"{current_app.config['PROMETHEUS_URL']}/api/v1/query_range"
|
|
|
|
#print(f"{prometheus_range_url}?query={metric_queries[metric]}&{range_and_interval}")
|
|
|
|
prometheus_response = requests.get(f"{prometheus_range_url}?query={metric_queries[metric]}&{range_and_interval}")
|
|
if prometheus_response.status_code >= 300:
|
|
return (502, None)
|
|
|
|
if len(prometheus_response.json()["data"]["result"]) == 0:
|
|
return (404, None)
|
|
|
|
time_series_data = list(map(
|
|
lambda x: (datetime.fromtimestamp(x[0]), float(x[1])),
|
|
prometheus_response.json()["data"]["result"][0]["values"]
|
|
))
|
|
|
|
plot_bytes = draw_plot_png_bytes(time_series_data, scale=scales[metric], size_x=sizes[size][0], size_y=sizes[size][1])
|
|
|
|
return (200, plot_bytes)
|
|
|
|
|
|
def draw_plot_png_bytes(data, scale, size_x=3, size_y=1):
|
|
|
|
pyplot.style.use("seaborn-dark")
|
|
fig, my_plot = pyplot.subplots(figsize=(size_x, size_y))
|
|
|
|
# x=range(1, 15)
|
|
# y=[1,4,6,8,4,5,3,2,4,1,5,6,8,7]
|
|
|
|
divide_by = 1
|
|
unit = ""
|
|
|
|
if scale > 1024 and scale < 1024*1024*1024:
|
|
divide_by = 1024*1024
|
|
unit = "MB"
|
|
if scale > 1024*1024*1024:
|
|
divide_by = 1024*1024*1024
|
|
unit = "GB"
|
|
|
|
scale /= divide_by
|
|
|
|
if scale > 10:
|
|
my_plot.get_yaxis().set_major_formatter( ticker.FuncFormatter(lambda x, p: "{}{}".format(int(x), unit)) )
|
|
elif scale > 1:
|
|
my_plot.get_yaxis().set_major_formatter( ticker.FuncFormatter(lambda x, p: "{:.1f}{}".format(x, unit)) )
|
|
else:
|
|
my_plot.get_yaxis().set_major_formatter( ticker.FuncFormatter(lambda x, p: "{:.2f}{}".format(x, unit)) )
|
|
|
|
x=list(map(lambda x: x[0], data))
|
|
y=list(map(lambda x: x[1]/divide_by, data))
|
|
|
|
minutes = float((x[len(x)-1] - x[0]).total_seconds())/float(60)
|
|
hours = minutes/float(60)
|
|
days = hours/float(24)
|
|
|
|
week_locator = mdates.WeekdayLocator()
|
|
minute_locator = mdates.MinuteLocator()
|
|
ten_minute_locator = mdates.MinuteLocator(interval=10)
|
|
hour_locator = mdates.HourLocator(interval=6)
|
|
hour_minute_formatter = mdates.DateFormatter('%H:%M')
|
|
day_formatter = mdates.DateFormatter('%b %d')
|
|
|
|
if minutes < 10:
|
|
my_plot.xaxis.set_major_locator(minute_locator)
|
|
my_plot.xaxis.set_major_formatter(hour_minute_formatter)
|
|
elif hours < 2:
|
|
my_plot.xaxis.set_major_locator(ten_minute_locator)
|
|
my_plot.xaxis.set_major_formatter(hour_minute_formatter)
|
|
elif days < 2:
|
|
my_plot.xaxis.set_major_locator(hour_locator)
|
|
my_plot.xaxis.set_major_formatter(hour_minute_formatter)
|
|
else:
|
|
my_plot.xaxis.set_major_locator(week_locator)
|
|
my_plot.xaxis.set_major_formatter(day_formatter)
|
|
|
|
max_value = reduce(lambda a, b: a if a > b else b, y, scale)
|
|
|
|
average=(sum(y)/len(y))/scale
|
|
average=average*1.25+0.1
|
|
|
|
bg_color=color_gradient(average)
|
|
|
|
average -= 0.1
|
|
|
|
fill_color=color_gradient(average)
|
|
highlight_color=lerp_rgb_tuples(fill_color, (1,1,1), 0.5)
|
|
|
|
my_plot.fill_between( x, max_value, color=bg_color, alpha=0.13)
|
|
my_plot.fill_between( x, y, color=highlight_color, alpha=0.3)
|
|
my_plot.plot(x, y, 'r-', color=highlight_color)
|
|
|
|
if size_y < 4:
|
|
my_plot.set_yticks([0, scale])
|
|
my_plot.set_ylim(0, scale)
|
|
|
|
my_plot.xaxis.label.set_color(highlight_color)
|
|
my_plot.tick_params(axis='x', colors=highlight_color)
|
|
my_plot.yaxis.label.set_color(highlight_color)
|
|
my_plot.tick_params(axis='y', colors=highlight_color)
|
|
|
|
if size_x < 4:
|
|
my_plot.set_xticklabels([])
|
|
if size_y < 1:
|
|
my_plot.set_yticklabels([])
|
|
|
|
image_binary = BytesIO()
|
|
fig.savefig(image_binary, transparent=True, bbox_inches="tight", pad_inches=0.05)
|
|
return image_binary.getvalue()
|
|
|
|
|
|
def lerp_rgb_tuples(a, b, lerp):
|
|
if lerp < 0:
|
|
lerp = 0
|
|
if lerp > 1:
|
|
lerp = 1
|
|
return (
|
|
a[0]*(1.0-lerp)+b[0]*lerp,
|
|
a[1]*(1.0-lerp)+b[1]*lerp,
|
|
a[2]*(1.0-lerp)+b[2]*lerp
|
|
)
|
|
|
|
def color_gradient(value):
|
|
if value < 0:
|
|
value = 0
|
|
if value > 1:
|
|
value = 1
|
|
if value < 0.5:
|
|
return lerp_rgb_tuples(green, blue, value*2)
|
|
else:
|
|
return lerp_rgb_tuples(blue, red, (value-0.5)*2) |