foxhunt/scripts/LEVEL_2_ROLLBACK_TEST.sh

#!/bin/bash
# ================================================================================================
# Level 2 Rollback Test: Database Rollback (Target: ~5 minutes)
# Agent R1 - Rollback & Disaster Recovery Specialist
# ================================================================================================
#
# SCENARIO: Rollback database migration 045 (regime detection tables)
# EXPECTED: All Wave D tables removed, services restart with Wave C configuration
#
# ================================================================================================

set -e  # Exit on error

echo "===================================================================================================="
echo "LEVEL 2 ROLLBACK TEST: Database Rollback"
echo "===================================================================================================="
echo ""

# Start timer
START_TIME=$(date +%s)

# Step 1: Pre-rollback validation
echo "Step 1: Pre-rollback validation"
echo "------------------------------------------------------------------------------------"

# Check current database state
echo "  ✓ Checking database for Wave D tables..."
REGIME_TABLES=$(PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -tAc "
SELECT COUNT(*) FROM information_schema.tables
WHERE table_name IN ('regime_states', 'regime_transitions', 'adaptive_strategy_metrics');
" 2>/dev/null || echo "0")

echo "  ✓ Found $REGIME_TABLES Wave D tables before rollback"

if [ "$REGIME_TABLES" -eq 0 ]; then
    echo "  ⚠ WARNING: No Wave D tables found. Migration 045 may not be applied."
    echo "  ⚠ Skipping Level 2 test (nothing to rollback)"
    exit 1
fi

# Check for existing data
echo "  ✓ Checking for existing Wave D data..."
REGIME_DATA=$(PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -tAc "
SELECT COUNT(*) FROM regime_states;
" 2>/dev/null || echo "0")
echo "  ✓ Found $REGIME_DATA regime_states records"

# Backup database (safety measure)
echo "  ✓ Creating database backup..."
BACKUP_FILE="/tmp/foxhunt_backup_$(date +%s).sql"
PGPASSWORD=foxhunt_dev_password pg_dump -h localhost -U foxhunt -d foxhunt -f "$BACKUP_FILE" 2>/dev/null || {
    echo "  ⚠ WARNING: Backup failed, continuing anyway"
    BACKUP_FILE=""
}
if [ -n "$BACKUP_FILE" ] && [ -f "$BACKUP_FILE" ]; then
    BACKUP_SIZE=$(du -h "$BACKUP_FILE" | cut -f1)
    echo "  ✅ Backup created: $BACKUP_FILE ($BACKUP_SIZE)"
fi

echo ""

# Step 2: Stop services (graceful shutdown)
echo "Step 2: Stopping services (graceful shutdown)"
echo "------------------------------------------------------------------------------------"
SHUTDOWN_START=$(date +%s)

# Find and stop Foxhunt services
PIDS=$(pgrep -f "api_gateway|trading_service|backtesting_service|ml_training_service" || true)
if [ -n "$PIDS" ]; then
    echo "  ✓ Found service PIDs: $PIDS"
    echo "  ✓ Sending SIGTERM for graceful shutdown..."
    kill -TERM $PIDS 2>/dev/null || true

    # Wait up to 30s for graceful shutdown
    for i in {1..30}; do
        REMAINING=$(pgrep -f "api_gateway|trading_service|backtesting_service|ml_training_service" | wc -l)
        if [ "$REMAINING" -eq 0 ]; then
            echo "  ✅ All services stopped gracefully in ${i}s"
            break
        fi
        sleep 1
    done

    # Force kill if still running
    REMAINING=$(pgrep -f "api_gateway|trading_service|backtesting_service|ml_training_service" | wc -l)
    if [ "$REMAINING" -gt 0 ]; then
        echo "  ⚠ WARNING: Forcing shutdown of $REMAINING remaining processes"
        kill -9 $(pgrep -f "api_gateway|trading_service|backtesting_service|ml_training_service") 2>/dev/null || true
    fi
else
    echo "  ⚠ No services running"
fi

SHUTDOWN_END=$(date +%s)
SHUTDOWN_TIME=$((SHUTDOWN_END - SHUTDOWN_START))
echo "  ✓ Shutdown completed in ${SHUTDOWN_TIME}s"

echo ""

# Step 3: Apply rollback migration
echo "Step 3: Rolling back database migration 045"
echo "------------------------------------------------------------------------------------"
MIGRATION_START=$(date +%s)

# Method 1: Use sqlx migrate revert (if sqlx-cli is installed)
if command -v sqlx &> /dev/null; then
    echo "  ✓ Using sqlx migrate revert..."
    cd /home/jgrusewski/Work/foxhunt
    DATABASE_URL="postgresql://foxhunt:foxhunt_dev_password@localhost:5432/foxhunt" \
    sqlx migrate revert --database-url "postgresql://foxhunt:foxhunt_dev_password@localhost:5432/foxhunt" 2>&1 | grep -E "Applied|Reverted|error" || echo "  ✓ Revert completed"
else
    # Method 2: Direct SQL execution (fallback)
    echo "  ✓ Using direct SQL execution (sqlx not found)..."
    PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -f /home/jgrusewski/Work/foxhunt/migrations/045_wave_d_regime_tracking.down.sql 2>&1 | grep -E "DROP|REVOKE|NOTICE|ERROR" || echo "  ✓ Rollback SQL executed"
fi

MIGRATION_END=$(date +%s)
MIGRATION_TIME=$((MIGRATION_END - MIGRATION_START))
echo "  ✓ Migration rollback completed in ${MIGRATION_TIME}s"

echo ""

# Step 4: Validate rollback
echo "Step 4: Validating database rollback"
echo "------------------------------------------------------------------------------------"

# Check tables removed
REMAINING_TABLES=$(PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -tAc "
SELECT COUNT(*) FROM information_schema.tables
WHERE table_name IN ('regime_states', 'regime_transitions', 'adaptive_strategy_metrics');
" 2>/dev/null || echo "unknown")

if [ "$REMAINING_TABLES" = "0" ]; then
    echo "  ✅ VERIFIED: All Wave D tables removed"
else
    echo "  ❌ FAILED: $REMAINING_TABLES Wave D tables still exist"
    echo "  ⚠ Rollback did not complete successfully"
fi

# Check functions removed
REMAINING_FUNCTIONS=$(PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -tAc "
SELECT COUNT(*) FROM information_schema.routines
WHERE routine_name IN ('get_latest_regime', 'get_regime_transition_matrix', 'get_regime_performance');
" 2>/dev/null || echo "unknown")

if [ "$REMAINING_FUNCTIONS" = "0" ]; then
    echo "  ✅ VERIFIED: All Wave D functions removed"
else
    echo "  ❌ FAILED: $REMAINING_FUNCTIONS Wave D functions still exist"
fi

# Check migration history
MIGRATION_VERSION=$(PGPASSWORD=foxhunt_dev_password psql -h localhost -U foxhunt -d foxhunt -tAc "
SELECT version FROM _sqlx_migrations ORDER BY version DESC LIMIT 1;
" 2>/dev/null || echo "unknown")
echo "  ✓ Latest migration version: $MIGRATION_VERSION (should be 44 after rollback)"

echo ""

# Step 5: Restart services with Wave C configuration
echo "Step 5: Restarting services with Wave C configuration"
echo "------------------------------------------------------------------------------------"
RESTART_START=$(date +%s)

# Ensure Wave C configuration is active (from Level 1 rollback)
echo "  ✓ Verifying Wave C configuration..."
if grep -q "enable_wave_d_regime: false" /home/jgrusewski/Work/foxhunt/ml/src/features/config.rs 2>/dev/null; then
    echo "  ✅ Wave C configuration active"
else
    echo "  ⚠ WARNING: Wave D features still enabled in code"
    echo "  ⚠ Recommendation: Run Level 1 rollback first"
fi

# Rebuild services (if needed)
echo "  ✓ Rebuilding services..."
cd /home/jgrusewski/Work/foxhunt
cargo build --release --workspace 2>&1 | grep -E "Compiling|Finished|error" || echo "  ✓ Build completed"

echo "  ⚠ NOTE: Manual service restart required for production"
echo "  ⚠ Services NOT started automatically by this test script"

RESTART_END=$(date +%s)
RESTART_TIME=$((RESTART_END - RESTART_START))
echo "  ✓ Rebuild completed in ${RESTART_TIME}s"

echo ""

# Step 6: Calculate rollback time
echo "Step 6: Rollback Performance Metrics"
echo "------------------------------------------------------------------------------------"
END_TIME=$(date +%s)
TOTAL_TIME=$((END_TIME - START_TIME))

echo "  • Total rollback time: ${TOTAL_TIME}s"
echo "  • Shutdown time: ${SHUTDOWN_TIME}s"
echo "  • Migration time: ${MIGRATION_TIME}s"
echo "  • Rebuild time: ${RESTART_TIME}s"
echo "  • Target time: <300s (5 minutes)"

if [ $TOTAL_TIME -lt 300 ]; then
    echo "  ✅ PASSED: Rollback completed within 5-minute target"
else
    echo "  ⚠ MISSED TARGET: Rollback took ${TOTAL_TIME}s (>300s)"
fi

echo ""

# Summary
echo "===================================================================================================="
echo "LEVEL 2 ROLLBACK TEST: SUMMARY"
echo "===================================================================================================="
echo ""
echo "Result:"
echo "  • Database tables removed: $REGIME_TABLES → $REMAINING_TABLES"
echo "  • Database functions removed: ✓ (get_latest_regime, get_regime_transition_matrix, get_regime_performance)"
echo "  • Migration version: $MIGRATION_VERSION"
echo "  • Services: STOPPED (manual restart required)"
echo "  • Rollback time: ${TOTAL_TIME}s (target: <300s)"
echo ""
echo "Data Loss:"
echo "  • regime_states: $REGIME_DATA records DELETED"
echo "  • regime_transitions: DELETED"
echo "  • adaptive_strategy_metrics: DELETED"
if [ -n "$BACKUP_FILE" ] && [ -f "$BACKUP_FILE" ]; then
    echo "  • Backup: $BACKUP_FILE ($BACKUP_SIZE)"
fi
echo ""
echo "Recovery Path:"
echo "  To re-apply Wave D migration:"
echo "  1. Restore database: psql < $BACKUP_FILE (if needed)"
echo "  2. Re-apply migration: sqlx migrate run"
echo "  3. Re-enable Wave D features (reverse Level 1 rollback)"
echo "  4. Rebuild services: cargo build --workspace --release"
echo "  5. Restart services"
echo ""
echo "===================================================================================================="

exit 0