CBDDC/src/ZB.MOM.WW.CBDDC.Core/Resilience/RetryPolicy.cs

using System;
using System.IO;
using System.Net.Sockets;
using System.Threading;
using System.Threading.Tasks;
using Microsoft.Extensions.Logging;
using Microsoft.Extensions.Logging.Abstractions;
using ZB.MOM.WW.CBDDC.Core.Exceptions;
using ZB.MOM.WW.CBDDC.Core.Network;
using TimeoutException = ZB.MOM.WW.CBDDC.Core.Exceptions.TimeoutException;

namespace ZB.MOM.WW.CBDDC.Core.Resilience;

/// <summary>
///     Provides retry logic for transient failures.
/// </summary>
public class RetryPolicy : IRetryPolicy
{
    private readonly ILogger<RetryPolicy> _logger;
    private readonly IPeerNodeConfigurationProvider _peerNodeConfigurationProvider;

    /// <summary>
    ///     Initializes a new instance of the <see cref="RetryPolicy" /> class.
    /// </summary>
    /// <param name="peerNodeConfigurationProvider">The provider for retry configuration values.</param>
    /// <param name="logger">The logger instance.</param>
    public RetryPolicy(IPeerNodeConfigurationProvider peerNodeConfigurationProvider,
        ILogger<RetryPolicy>? logger = null)
    {
        _logger = logger ?? NullLogger<RetryPolicy>.Instance;
        _peerNodeConfigurationProvider = peerNodeConfigurationProvider
                                         ?? throw new ArgumentNullException(nameof(peerNodeConfigurationProvider));
    }

    /// <summary>
    ///     Executes an operation with retry logic.
    /// </summary>
    /// <typeparam name="T">The result type returned by the operation.</typeparam>
    /// <param name="operation">The asynchronous operation to execute.</param>
    /// <param name="operationName">The operation name used for logging.</param>
    /// <param name="cancellationToken">A token used to cancel retry delays.</param>
    public async Task<T> ExecuteAsync<T>(
        Func<Task<T>> operation,
        string operationName,
        CancellationToken cancellationToken = default)
    {
        var config = await _peerNodeConfigurationProvider.GetConfiguration();
        Exception? lastException = null;

        for (var attempt = 1; attempt <= config.RetryAttempts; attempt++)
            try
            {
                _logger.LogDebug("Executing {Operation} (attempt {Attempt}/{Max})",
                    operationName, attempt, config.RetryAttempts);

                return await operation();
            }
            catch (Exception ex) when (attempt < config.RetryAttempts && IsTransient(ex))
            {
                lastException = ex;
                int delay = config.RetryDelayMs * attempt; // Exponential backoff

                _logger.LogWarning(ex,
                    "Operation {Operation} failed (attempt {Attempt}/{Max}). Retrying in {Delay}ms...",
                    operationName, attempt, config.RetryAttempts, delay);

                await Task.Delay(delay, cancellationToken);
            }

        if (lastException != null)
            _logger.LogError(lastException,
                "Operation {Operation} failed after {Attempts} attempts",
                operationName, config.RetryAttempts);
        else
            _logger.LogError(
                "Operation {Operation} failed after {Attempts} attempts",
                operationName, config.RetryAttempts);

        throw new CBDDCException("RETRY_EXHAUSTED",
            $"Operation '{operationName}' failed after {config.RetryAttempts} attempts",
            lastException!);
    }

    /// <summary>
    ///     Executes an operation with retry logic (void return).
    /// </summary>
    /// <param name="operation">The asynchronous operation to execute.</param>
    /// <param name="operationName">The operation name used for logging.</param>
    /// <param name="cancellationToken">A token used to cancel retry delays.</param>
    public async Task ExecuteAsync(
        Func<Task> operation,
        string operationName,
        CancellationToken cancellationToken = default)
    {
        await ExecuteAsync(async () =>
        {
            await operation();
            return true;
        }, operationName, cancellationToken);
    }

    private static bool IsTransient(Exception ex)
    {
        // Network errors are typically transient
        if (ex is NetworkException or SocketException or IOException)
            return true;

        // Timeout errors are transient
        if (ex is TimeoutException or OperationCanceledException)
            return true;

        return false;
    }
}