c-relay/tests/event_config_tests.sh

#!/bin/bash

# Comprehensive Error Handling and Recovery Testing for Event-Based Configuration System
# Tests various failure scenarios and recovery mechanisms

set -e

# Configuration
RELAY_BINARY="./build/c_relay_x86"
TEST_DB_PREFIX="test_relay"
LOG_FILE="test_results.log"

# Colors for output
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

# Test results tracking
TESTS_PASSED=0
TESTS_FAILED=0
TESTS_TOTAL=0

# Function to print colored output
print_test_header() {
    echo -e "${BLUE}[TEST]${NC} $1"
    ((TESTS_TOTAL++))
}

print_success() {
    echo -e "${GREEN}[PASS]${NC} $1"
    ((TESTS_PASSED++))
}

print_failure() {
    echo -e "${RED}[FAIL]${NC} $1"
    ((TESTS_FAILED++))
}

print_info() {
    echo -e "${YELLOW}[INFO]${NC} $1"
}

# Clean up function
cleanup_test_files() {
    print_info "Cleaning up test files..."
    pkill -f "c_relay_" 2>/dev/null || true
    rm -f ${TEST_DB_PREFIX}*.nrdb* 2>/dev/null || true
    rm -f test_*.log 2>/dev/null || true
    sleep 1
}

# Function to start relay and capture output
start_relay_test() {
    local test_name="$1"
    local timeout="${2:-10}"
    
    print_info "Starting relay for test: $test_name"
    timeout $timeout $RELAY_BINARY > "test_${test_name}.log" 2>&1 &
    local relay_pid=$!
    sleep 2
    
    if kill -0 $relay_pid 2>/dev/null; then
        echo $relay_pid
    else
        echo "0"
    fi
}

# Function to stop relay
stop_relay_test() {
    local relay_pid="$1"
    if [ "$relay_pid" != "0" ]; then
        kill $relay_pid 2>/dev/null || true
        wait $relay_pid 2>/dev/null || true
    fi
}

# Function to check if relay started successfully
check_relay_startup() {
    local log_file="$1"
    if grep -q "First-time startup sequence completed\|Existing relay startup" "$log_file" 2>/dev/null; then
        return 0
    else
        return 1
    fi
}

# Function to check if relay has admin keys
check_admin_keys() {
    local log_file="$1"
    if grep -q "Admin Private Key:" "$log_file" 2>/dev/null; then
        return 0
    else
        return 1
    fi
}

# Function to check database file creation
check_database_creation() {
    if ls *.nrdb 2>/dev/null | head -1; then
        return 0
    else
        return 1
    fi
}

# Function to check configuration event in database
check_config_event_stored() {
    local db_file="$1"
    if [ -f "$db_file" ]; then
        local count=$(sqlite3 "$db_file" "SELECT COUNT(*) FROM events WHERE kind = 33334;" 2>/dev/null || echo "0")
        if [ "$count" -gt 0 ]; then
            return 0
        fi
    fi
    return 1
}

echo "========================================"
echo "Event-Based Configuration System Tests"
echo "========================================"
echo

# Ensure binary exists
if [ ! -f "$RELAY_BINARY" ]; then
    print_failure "Relay binary not found. Please build first: make"
    exit 1
fi

print_info "Starting comprehensive error handling and recovery tests..."
echo

# TEST 1: Normal First-Time Startup
print_test_header "Test 1: Normal First-Time Startup"
cleanup_test_files

relay_pid=$(start_relay_test "first_startup" 15)
sleep 5
stop_relay_test $relay_pid

if check_relay_startup "test_first_startup.log"; then
    if check_admin_keys "test_first_startup.log"; then
        if db_file=$(check_database_creation); then
            if check_config_event_stored "$db_file"; then
                print_success "First-time startup completed successfully"
            else
                print_failure "Configuration event not stored in database"
            fi
        else
            print_failure "Database file not created"
        fi
    else
        print_failure "Admin keys not generated"
    fi
else
    print_failure "Relay failed to complete startup"
fi

# TEST 2: Existing Relay Startup
print_test_header "Test 2: Existing Relay Startup (using existing database)"

relay_pid=$(start_relay_test "existing_startup" 10)
sleep 3
stop_relay_test $relay_pid

if check_relay_startup "test_existing_startup.log"; then
    if ! check_admin_keys "test_existing_startup.log"; then
        print_success "Existing relay startup (no new keys generated)"
    else
        print_failure "New admin keys generated for existing relay"
    fi
else
    print_failure "Existing relay failed to start"
fi

# TEST 3: Corrupted Database Recovery
print_test_header "Test 3: Corrupted Database Recovery"

if db_file=$(check_database_creation); then
    # Corrupt the database by truncating it
    truncate -s 100 "$db_file"
    print_info "Database corrupted for recovery test"
    
    relay_pid=$(start_relay_test "corrupted_db" 10)
    sleep 3
    stop_relay_test $relay_pid
    
    if grep -q "ERROR.*database\|Failed.*database\|disk I/O error" "test_corrupted_db.log"; then
        print_success "Corrupted database properly detected and handled"
    else
        print_failure "Corrupted database not properly handled"
    fi
fi

# TEST 4: Missing Database File Recovery
print_test_header "Test 4: Missing Database File Recovery"
cleanup_test_files

# Create a database then remove it to simulate loss
relay_pid=$(start_relay_test "create_db" 10)
sleep 3
stop_relay_test $relay_pid

if db_file=$(check_database_creation); then
    rm -f "$db_file"*
    print_info "Database files removed to test recovery"
    
    relay_pid=$(start_relay_test "missing_db" 15)
    sleep 5
    stop_relay_test $relay_pid
    
    if check_relay_startup "test_missing_db.log"; then
        if check_admin_keys "test_missing_db.log"; then
            print_success "Missing database recovery successful (new keys generated)"
        else
            print_failure "New admin keys not generated after database loss"
        fi
    else
        print_failure "Failed to recover from missing database"
    fi
fi

# TEST 5: Invalid Configuration Event Handling
print_test_header "Test 5: Configuration Event Structure Validation"

# This test would require injecting an invalid configuration event
# For now, we check that the validation functions are properly integrated
if grep -q "nostr_validate_event_structure\|nostr_verify_event_signature" src/config.c; then
    print_success "Configuration event validation functions integrated"
else
    print_failure "Configuration event validation functions not found"
fi

# TEST 6: Database Schema Version Check
print_test_header "Test 6: Database Schema Consistency"

if db_file=$(check_database_creation); then
    # Check that the database has the correct schema version
    schema_version=$(sqlite3 "$db_file" "SELECT value FROM schema_info WHERE key = 'version';" 2>/dev/null || echo "")
    if [ "$schema_version" = "4" ]; then
        print_success "Database schema version is correct (v4)"
    else
        print_failure "Database schema version incorrect: $schema_version (expected: 4)"
    fi
    
    # Check that legacy tables don't exist
    if ! sqlite3 "$db_file" ".tables" 2>/dev/null | grep -q "config_file_cache\|active_config"; then
        print_success "Legacy configuration tables properly removed"
    else
        print_failure "Legacy configuration tables still present"
    fi
fi

# TEST 7: Memory and Resource Management
print_test_header "Test 7: Resource Cleanup and Memory Management"

relay_pid=$(start_relay_test "resource_test" 15)
sleep 5

# Check for memory leaks or resource issues (basic check)
if kill -0 $relay_pid 2>/dev/null; then
    # Send termination signal and check cleanup
    kill -TERM $relay_pid 2>/dev/null || true
    sleep 2
    
    if ! kill -0 $relay_pid 2>/dev/null; then
        if grep -q "Configuration system cleaned up" "test_resource_test.log"; then
            print_success "Resource cleanup completed successfully"
        else
            print_failure "Resource cleanup not logged properly"
        fi
    else
        kill -KILL $relay_pid 2>/dev/null || true
        print_failure "Relay did not shut down cleanly"
    fi
else
    print_failure "Relay process not running for resource test"
fi

# TEST 8: Configuration Cache Consistency
print_test_header "Test 8: Configuration Cache Consistency"

if db_file=$(check_database_creation); then
    # Check that configuration is properly cached and accessible
    config_count=$(sqlite3 "$db_file" "SELECT COUNT(*) FROM events WHERE kind = 33334;" 2>/dev/null || echo "0")
    if [ "$config_count" -eq 1 ]; then
        print_success "Single configuration event stored (replaceable event working)"
    else
        print_failure "Multiple or no configuration events found: $config_count"
    fi
fi

# TEST 9: Network Port Binding
print_test_header "Test 9: Network Port Availability and Binding"

relay_pid=$(start_relay_test "network_test" 10)
sleep 3

if kill -0 $relay_pid 2>/dev/null; then
    # Check if port 8888 is being used
    if netstat -tln 2>/dev/null | grep -q ":8888"; then
        print_success "Relay successfully bound to network port 8888"
    else
        print_failure "Relay not bound to expected port 8888"
    fi
    stop_relay_test $relay_pid
else
    print_failure "Relay failed to start for network test"
fi

# TEST 10: Multiple Startup Attempts (Port Conflict)
print_test_header "Test 10: Port Conflict Handling"

relay_pid1=$(start_relay_test "port_conflict_1" 10)
sleep 2

if kill -0 $relay_pid1 2>/dev/null; then
    # Try to start a second relay (should fail due to port conflict)
    relay_pid2=$(start_relay_test "port_conflict_2" 5)
    sleep 1
    
    if [ "$relay_pid2" = "0" ] || ! kill -0 $relay_pid2 2>/dev/null; then
        print_success "Port conflict properly handled (second instance failed to start)"
    else
        print_failure "Multiple relay instances started (port conflict not handled)"
        stop_relay_test $relay_pid2
    fi
    
    stop_relay_test $relay_pid1
else
    print_failure "First relay instance failed to start"
fi

# Final cleanup
cleanup_test_files

# Test Results Summary
echo
echo "========================================"
echo "Test Results Summary"
echo "========================================"
echo "Tests Passed: $TESTS_PASSED"
echo "Tests Failed: $TESTS_FAILED"
echo "Total Tests:  $TESTS_TOTAL"
echo

if [ $TESTS_FAILED -eq 0 ]; then
    print_success "ALL TESTS PASSED! Event-based configuration system is robust."
    exit 0
else
    print_failure "$TESTS_FAILED tests failed. Review the results above."
    echo
    print_info "Check individual test log files (test_*.log) for detailed error information."
    exit 1
fi