#!/usr/bin/env bash
# Comprehensive fix for deployment issues
# Deletes failed/canceled clusters and re-runs Terraform

set -e

SUBSCRIPTION_ID="fc08d829-4f14-413d-ab27-ce024425db0b"
PROJECT_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
TERRAFORM_DIR="$PROJECT_ROOT/terraform/well-architected/cloud-sovereignty"

echo "╔════════════════════════════════════════════════════════════════╗"
echo "║     DEPLOYMENT FIX - COMPREHENSIVE CLEANUP & REDEPLOYMENT      ║"
echo "╚════════════════════════════════════════════════════════════════╝"
echo ""

echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo "Step 1: Delete Failed Clusters (7)"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

FAILED_CLUSTERS=$(az aks list --subscription "$SUBSCRIPTION_ID" \
  --query "[?contains(name, 'az-p-') && provisioningState == 'Failed'].{name:name, rg:resourceGroup}" -o json)

FAILED_COUNT=$(echo "$FAILED_CLUSTERS" | jq '. | length')
echo "Found $FAILED_COUNT failed clusters to delete"
echo ""

if [ "$FAILED_COUNT" -gt 0 ]; then
  echo "$FAILED_CLUSTERS" | jq -r '.[] | "\(.rg)|\(.name)"' | while IFS='|' read -r rg name; do
    echo "Deleting failed cluster: $name (RG: $rg)"
    az aks delete --resource-group "$rg" --name "$name" --subscription "$SUBSCRIPTION_ID" --yes --no-wait 2>&1 | grep -v "^$" || true
    echo "  ✅ Deletion initiated"
    echo ""
  done
  
  echo "Waiting for failed cluster deletions to complete..."
  sleep 30
  
  # Wait for deletions
  echo "$FAILED_CLUSTERS" | jq -r '.[] | "\(.rg)|\(.name)"' | while IFS='|' read -r rg name; do
    echo -n "  Waiting for $name..."
    while az aks show --resource-group "$rg" --name "$name" --subscription "$SUBSCRIPTION_ID" >/dev/null 2>&1; do
      echo -n "."
      sleep 5
    done
    echo " ✅ Deleted"
  done
else
  echo "No failed clusters to delete"
fi

echo ""
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo "Step 2: Delete Canceled Clusters (16)"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

CANCELED_CLUSTERS=$(az aks list --subscription "$SUBSCRIPTION_ID" \
  --query "[?contains(name, 'az-p-') && provisioningState == 'Canceled'].{name:name, rg:resourceGroup}" -o json)

CANCELED_COUNT=$(echo "$CANCELED_CLUSTERS" | jq '. | length')
echo "Found $CANCELED_COUNT canceled clusters to delete"
echo ""

if [ "$CANCELED_COUNT" -gt 0 ]; then
  echo "$CANCELED_CLUSTERS" | jq -r '.[] | "\(.rg)|\(.name)"' | while IFS='|' read -r rg name; do
    echo "Deleting canceled cluster: $name (RG: $rg)"
    az aks delete --resource-group "$rg" --name "$name" --subscription "$SUBSCRIPTION_ID" --yes --no-wait 2>&1 | grep -v "^$" || true
    echo "  ✅ Deletion initiated"
    echo ""
  done
  
  echo "Waiting for canceled cluster deletions to complete..."
  sleep 30
  
  # Wait for deletions (in batches)
  BATCH_SIZE=5
  BATCH_NUM=0
  echo "$CANCELED_CLUSTERS" | jq -r '.[] | "\(.rg)|\(.name)"' | while IFS='|' read -r rg name; do
    echo -n "  Waiting for $name..."
    while az aks show --resource-group "$rg" --name "$name" --subscription "$SUBSCRIPTION_ID" >/dev/null 2>&1; do
      echo -n "."
      sleep 5
    done
    echo " ✅ Deleted"
  done
else
  echo "No canceled clusters to delete"
fi

echo ""
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo "Step 3: Clean Terraform State"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

cd "$TERRAFORM_DIR"

echo "Removing deleted clusters from Terraform state..."
echo ""

# Get list of all cluster resources in state
TERRAFORM_STATE_CLUSTERS=$(terraform state list 2>/dev/null | grep "azurerm_kubernetes_cluster" || true)

if [ -n "$TERRAFORM_STATE_CLUSTERS" ]; then
  echo "Checking state for cluster resources..."
  echo "$TERRAFORM_STATE_CLUSTERS" | while read -r resource; do
    CLUSTER_NAME=$(echo "$resource" | sed 's/.*\.main\[.*\]//' || echo "$resource" | awk -F'.' '{print $NF}')
    echo "  Checking: $resource"
    
    # Try to check if cluster still exists
    if echo "$resource" | grep -q "azurerm_kubernetes_cluster"; then
      echo "    Resource in state: $resource"
    fi
  done
else
  echo "No cluster resources found in Terraform state"
fi

echo ""
echo "Note: Terraform will automatically handle state cleanup during apply"
echo ""

echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo "Step 4: Re-run Terraform Deployment"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

echo "Initializing Terraform..."
terraform init -upgrade >/dev/null 2>&1 || true

echo ""
echo "Re-running Terraform deployment..."
echo "This will recreate all deleted clusters with proper configuration"
echo ""
echo "⚠️  This may take 15-30 minutes depending on region availability"
echo ""

# Run Terraform apply with maximum parallelism
terraform apply -parallelism=128 -auto-approve 2>&1 | tee /tmp/terraform-apply-fixed.log

echo ""
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo "Step 5: Verify Deployment"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

echo "Waiting 30 seconds for clusters to stabilize..."
sleep 30

echo ""
echo "Checking cluster status..."
echo ""

READY_COUNT=$(az aks list --subscription "$SUBSCRIPTION_ID" \
  --query "[?contains(name, 'az-p-') && provisioningState == 'Succeeded'].name" -o tsv | wc -l)

FAILED_COUNT=$(az aks list --subscription "$SUBSCRIPTION_ID" \
  --query "[?contains(name, 'az-p-') && provisioningState == 'Failed'].name" -o tsv | wc -l)

CREATING_COUNT=$(az aks list --subscription "$SUBSCRIPTION_ID" \
  --query "[?contains(name, 'az-p-') && provisioningState == 'Creating'].name" -o tsv | wc -l)

echo "📊 Deployment Status:"
echo "   ✅ Ready (Succeeded): $READY_COUNT"
echo "   ❌ Failed: $FAILED_COUNT"
echo "   ⏳ Creating: $CREATING_COUNT"
echo ""

if [ "$CREATING_COUNT" -gt 0 ]; then
  echo "⚠️  Some clusters are still creating. Monitor with:"
  echo "   az aks list --subscription $SUBSCRIPTION_ID --query \"[?contains(name, 'az-p-')].{name:name, state:provisioningState}\" -o table"
fi

if [ "$FAILED_COUNT" -gt 0 ]; then
  echo "⚠️  Some clusters failed. Check logs:"
  echo "   tail -100 /tmp/terraform-apply-fixed.log"
  echo "   ./scripts/azure/analyze-deployment-failures.sh"
fi

echo ""
echo "✅ Fix process complete!"
echo ""
echo "📝 Logs:"
echo "   • Terraform: /tmp/terraform-apply-fixed.log"
echo "   • This script: Check output above"
echo ""
echo "🎯 Next Steps:"
echo "   1. Monitor cluster creation: az aks list --query \"[?contains(name, 'az-p-')].{name:name, state:provisioningState}\" -o table"
echo "   2. Once ready, run: ./scripts/deployment/wait-and-run-all-next-steps.sh"
echo ""