demo/k8s/helm/logsys/scripts/cardinality-gate.sh

#!/bin/bash
# Cardinality Gate - CI stage validation per Log.md
# Run this script in CI before deploying Vector to ensure labels stay within cardinality limits

set -euo pipefail

LOKI_ENDPOINT="${LOKI_ENDPOINT:-http://loki:3100}"
MAX_CARDINALITY=5000

echo "🔍 Checking label cardinality limits for Vector-Loki deployment..."

# Helper function to get cardinality from Loki
check_label_cardinality() {
  local label=$1
  local max=$2
  
  echo -n "  Checking ${label}... "
  
  # Query Loki for unique values of a label in the last 24h
  # This requires logcli installed: https://github.com/grafana/loki/releases
  COUNT=$(logcli query '{job="vector"}' --since=24h -o raw 2>/dev/null | \
    jq -r ".[] | .${label}" 2>/dev/null | sort | uniq | wc -l) || COUNT=0
  
  if [ "$COUNT" -gt "$max" ]; then
    echo "❌ FAILED (${COUNT} > ${max})"
    return 1
  else
    echo "✅ OK (${COUNT} <= ${max})"
    return 0
  fi
}

# Check critical low-cardinality labels (should be <<5000)
FAILED=false
check_label_cardinality "env" 10 || FAILED=true
check_label_cardinality "app" 50 || FAILED=true
check_label_cardinality "level" 10 || FAILED=true
check_label_cardinality "event_class" 20 || FAILED=true
check_label_cardinality "uri_group" 100 || FAILED=true
check_label_cardinality "status" 10 || FAILED=true

if [ "$FAILED" = true ]; then
  echo ""
  echo "❌ Cardinality check failed! Some labels exceed safe limits."
  echo "   Review the Vector transform rules and reduce uri_group cardinality if needed."
  exit 1
else
  echo ""
  echo "✅ All label cardinality checks passed!"
  echo "   Safe to deploy."
  exit 0
fi