Blue-green, canary va shadow: xavfsiz chiqarish

import pandas as pd
show(pd.DataFrame([
    {"Usul": "To'g'ridan chiqarish", "Xavf": "To'liq", "Yangiga trafik": "100%",
     "Xatoni qachon bilasiz": "Foydalanuvchi ko'rgandan keyin"},
    {"Usul": "Blue-green", "Xavf": "To'liq, qaytish tez", "Yangiga trafik": "0% -> 100%",
     "Xatoni qachon bilasiz": "Almashtirgandan keyin"},
    {"Usul": "Canary", "Xavf": "Qisman", "Yangiga trafik": "5% -> 100%",
     "Xatoni qachon bilasiz": "Kichik guruh ko'rganda"},
    {"Usul": "Shadow", "Xavf": "**Nol**", "Yangiga trafik": "100% (javob tashlanadi)",
     "Xatoni qachon bilasiz": "Hech kim ko'rmasdan"},
]))

python3 -c "
import sys, pathlib; sys.path.insert(0, 'src')
if not pathlib.Path('model.onnx').exists():
    import pandas as pd
    from sklearn.ensemble import RandomForestClassifier
    from sklearn.model_selection import train_test_split
    from quvur.xususiyat import XUSUSIYATLAR, tayyorla
    from quvur.format import onnx_ga
    d = pd.read_csv('../datasets/ijara.csv')
    x, y = tayyorla(d), d['broker']
    Xtr, _, ytr, _ = train_test_split(x, y, test_size=0.25, stratify=y, random_state=42)
    m = RandomForestClassifier(n_estimators=150, min_samples_leaf=3,
                               random_state=42, n_jobs=1).fit(Xtr, ytr)
    onnx_ga(m, len(XUSUSIYATLAR), 'model.onnx')
print('model tayyor')
"
docker images mlops-servis:v1 -q | grep -q . || docker build -q -f Dockerfile.servis -t mlops-servis:v1 . > /dev/null 2>&1

docker rm -f mlops-blue mlops-green > /dev/null 2>&1
docker run -d --name mlops-blue  -p 18140:8000 -e BROKER_CHEGARASI=0.5 mlops-servis:v1 > /dev/null 2>&1
docker run -d --name mlops-green -p 18141:8000 -e BROKER_CHEGARASI=0.35 mlops-servis:v1 > /dev/null 2>&1

for P in 18140 18141; do
  for i in $(seq 1 40); do
    [ "$(curl -s -o /dev/null -w %{http_code} http://127.0.0.1:${P}/ready 2>/dev/null)" = "200" ] && break
    sleep 0.5
  done
done
echo -n "BLUE  (chegara 0.5) : "; curl -s http://127.0.0.1:18140/ready
echo ""
echo -n "GREEN (chegara 0.35): "; curl -s http://127.0.0.1:18141/ready
echo ""

events {}

http {
  # 95% blue, 5% green
  split_clients "${remote_addr}${request_id}" $manzil {
      5%     green;
      *      blue;
  }

  upstream blue  { server host.docker.internal:18140; }
  upstream green { server host.docker.internal:18141; }

  server {
    listen 80;
    location / {
      proxy_pass http://$manzil;
      proxy_set_header Host $host;
      add_header X-Versiya $manzil always;
    }
  }
}

docker rm -f mlops-canary > /dev/null 2>&1
docker run -d --name mlops-canary -p 18142:80 \
  --add-host host.docker.internal:host-gateway \
  -v "$(pwd)/nginx-canary.conf:/etc/nginx/nginx.conf:ro" \
  nginx:1.27-alpine > /dev/null 2>&1
for i in $(seq 1 40); do
  [ "$(curl -s -o /dev/null -w %{http_code} http://127.0.0.1:18142/healthz 2>/dev/null)" = "200" ] && break
  sleep 0.5
done

python3 -c "
import urllib.request, collections
n = collections.Counter()
for i in range(200):
    r = urllib.request.Request('http://127.0.0.1:18142/healthz')
    try:
        with urllib.request.urlopen(r, timeout=5) as j:
            n[j.headers.get('X-Versiya', '?')] += 1
    except Exception:
        n['xato'] += 1
print('200 sorov taqsimoti:')
for k, v in sorted(n.items()):
    print(f'  {k:8s}: {v:>4}  ({v/2:.1f}%)')
"

python3 -c "
import json, urllib.request, collections
import pandas as pd

d = pd.read_csv('../datasets/ijara.csv').head(300)
XUS = ['xonalar','maydon','qavat','jami_qavat','narx_mln','tavsif_uzunligi',
       'rasmlar_soni','telefon_takrorlanishi','kunlik_elonlar','egasi_tasdiqlangan']

def sora(port, elon):
    r = urllib.request.Request(f'http://127.0.0.1:{port}/bashorat',
                               data=json.dumps(elon).encode(),
                               headers={'Content-Type': 'application/json'})
    with urllib.request.urlopen(r, timeout=10) as j:
        return json.loads(j.read())

farq = 0
blue_broker = green_broker = 0
ehtimollar = []
for e in d[XUS].to_dict('records'):
    b = sora(18140, e)      # BLUE - foydalanuvchi shuni oladi
    g = sora(18141, e)      # GREEN - soya, javob tashlanadi
    blue_broker += b['broker']
    green_broker += g['broker']
    if b['broker'] != g['broker']:
        farq += 1
    ehtimollar.append(abs(b['broker_ehtimoli'] - g['broker_ehtimoli']))

n = len(d)
print(f'{n} ta elon, SOYA solishtiruvi:')
print(f'  BLUE  broker dedi : {blue_broker:>4}  ({blue_broker/n*100:.1f}%)')
print(f'  GREEN broker dedi : {green_broker:>4}  ({green_broker/n*100:.1f}%)')
print(f'  QAROR FARQI       : {farq:>4}  ({farq/n*100:.1f}%)')
print(f'  ehtimol farqi     : {max(ehtimollar):.4f} (eng katta)')
print()
print('Foydalanuvchi FAQAT blue javobini oldi. Green hech kimga korinmadi.')
"

import pandas as pd
show(pd.DataFrame([
    {"Holat": "Kichik kod tuzatishi", "Tavsiya": "To'g'ridan + qaytarish (32-bob)"},
    {"Holat": "Katta o'zgarish, tez qaytish kerak", "Tavsiya": "Blue-green"},
    {"Holat": "Yangi xususiyat, foydalanuvchi xulqi noma'lum", "Tavsiya": "Canary"},
    {"Holat": "**Yangi model**", "Tavsiya": "**Shadow**, keyin canary"},
    {"Holat": "Ma'lumotlar bazasi sxemasi", "Tavsiya": "Hech biri — alohida, bosqichma-bosqich"},
]))

docker rm -f mlops-blue mlops-green mlops-canary > /dev/null 2>&1 && echo "tozalandi"
docker ps --format '{{.Names}}' | grep -v k3d