Why Every DevOps Engineer Should Learn NumPy in 2026

# Use a virtual environment so system Python stays clean
python3 -m venv ~/venvs/ops
source ~/venvs/ops/bin/activate
pip install numpy requests

# Check the install
python -c "import numpy as np; print(np.__version__)"

import numpy as np

# Response times (ms) copied from a load test
rt = np.array([120, 95, 101, 2300, 110, 99, 130, 105, 98, 115])

print("mean  :", rt.mean())          # 327.3  <- pulled up by one slow request
print("median:", np.median(rt))      # 107.5
print("max   :", rt.max())           # 2300
print("slow  :", (rt > 1000).sum())  # 1 request over 1 second

import numpy as np

# One response time per line, in milliseconds
lat = np.loadtxt("latency_ms.txt")

p50, p95, p99 = np.percentile(lat, [50, 95, 99])
print(f"requests: {lat.size}")
print(f"p50={p50:.1f}  p95={p95:.1f}  p99={p99:.1f} ms")
print(f"mean={lat.mean():.1f} ms (for comparison only)")

import numpy as np

lat = np.loadtxt("latency_ms.txt")
TARGET_MS = 300
SLO = 0.95                     # 95% of requests under 300 ms

good = (lat <= TARGET_MS).mean()
allowed_bad = 1 - SLO
actual_bad = 1 - good
budget_used = actual_bad / allowed_bad

print(f"within target : {good:.2%}")
print(f"error budget used: {budget_used:.0%}")
if good < SLO:
    print("SLO MISSED - review before next release")

import numpy as np
from numpy.lib.stride_tricks import sliding_window_view

# cpu.csv columns: ts,cpu (one row per minute)
ts, cpu = np.loadtxt("cpu.csv", delimiter=",", skiprows=1, unpack=True)

w = 30                                   # compare with previous 30 minutes
win = sliding_window_view(cpu[:-1], w)   # windows that end before each point
mu, sd = win.mean(axis=1), win.std(axis=1)
z = (cpu[w:] - mu) / np.where(sd == 0, 1, sd)

for i in np.flatnonzero(np.abs(z) > 3):
    print(int(ts[i + w]), round(cpu[i + w], 1), "z =", round(z[i], 1))

# Robust alternative for spiky data
med = np.median(cpu)
mad = np.median(np.abs(cpu - med))
robust = 0.6745 * (cpu - med) / mad
print("MAD outliers at rows:", np.flatnonzero(np.abs(robust) > 3.5))

import numpy as np

# Daily used space in GB for the last 10 days
used = np.array([410, 414, 417, 422, 425, 431, 434, 438, 441, 446.0])
CAPACITY_GB = 500

days = np.arange(used.size)
slope, intercept = np.polyfit(days, used, 1)

if slope <= 0:
    print("usage is flat or shrinking")
else:
    days_left = (CAPACITY_GB - used[-1]) / slope
    print(f"growth: {slope:.1f} GB/day")
    print(f"about {days_left:.0f} days until full - plan resize early")

import time
import numpy as np
import requests

PROM = "http://localhost:9090"   # your Prometheus URL
query = 'sum(rate(node_cpu_seconds_total{mode!="idle"}[5m]))'
end = time.time()
start = end - 24 * 3600            # last 24 hours

r = requests.get(f"{PROM}/api/v1/query_range",
                 params={"query": query, "start": start,
                         "end": end, "step": "60"}, timeout=30)
r.raise_for_status()
series = r.json()["data"]["result"][0]["values"]  # [[ts, "value"], ...]

data = np.array(series, dtype=float)
ts, val = data[:, 0], data[:, 1]
print("points:", val.size)
print("p50/p95/max:", np.percentile(val, [50, 95]).round(2), val.max().round(2))

import re
import numpy as np

# Nginx log format must end with request time, e.g. ... "$http_user_agent" $request_time
pat = re.compile(r'" (\d{3}) .* (\d+\.\d+)$')
status, rt = [], []
with open("access.log") as f:
    for line in f:
        m = pat.search(line.rstrip())
        if m:
            status.append(int(m.group(1)))
            rt.append(float(m.group(2)))

status, rt = np.array(status), np.array(rt)
codes, counts = np.unique(status, return_counts=True)
for c, n in zip(codes, counts):
    print(c, n)
print(f"5xx rate: {(status >= 500).mean():.2%}")
print("p95 request time (s):", np.percentile(rt, 95).round(3))

WEEKLY RELIABILITY REPORT (generated with NumPy)

Service: ____________   Week: ____________

[ ] Requests analysed           : ________
[ ] p50 / p95 / p99 latency     : ____ / ____ / ____ ms
[ ] Share within SLO target     : ____ %   (target ____ %)
[ ] Error budget used           : ____ %
[ ] 5xx error rate              : ____ %
[ ] Anomalies flagged (z > 3)   : ____  -> linked to incident? Y/N
[ ] Disk growth (GB/day)        : ____  -> days to full: ____
[ ] Alert rule changes proposed : ____________________

Scripts and data stored in: repo/ops-analysis/