使用Polly正确处理CosmosDb Mongo 429错误的方法

3

我在Azure Web应用程序上部署了一个应用程序,它使用 .netCore 3 中的mongoDb驱动程序与CosmosDB数据库交互。

根据此文档,当RU/s不可用时,我必须设置重试策略以处理429错误代码。考虑到我没有看到429错误发生时显示的一种错误,在Polly中找到适当的处理策略并不容易。

唯一可行的方法是使用以下策略:

_retryPolicy = Policy
.Handle<MongoCommandException>(r => r.Message.Contains("Request rate is large"))
.WaitAndRetry(3, i => TimeSpan.FromSeconds(1));

以下是Polly策略的用法:

   public long CountProjetByProjectNumber(string projectNumber)
    {
        long result = 0;
        _retryPolicy.Execute(() =>
        {
            result = _mongoCollection.CountDocuments(x => x.ProjectNumber == projectNumber);
        });
        return result;
    }

有人能提供在使用Mongodb驱动程序时,当CosmosDb出现429异常时的正确报错显示,或者能否向我展示如何正确处理它的方法。

1个回答

9
实际上,为了正确处理速率限制和超时问题,您需要处理一些其他异常 - 尤其是在使用较新的MongoDB V3.6端点(而不是较旧的V3.2端点)时。
  • 对于V3.2端点:您关心的两个异常是MongoCommandExceptionMongoExecutionTimeoutExceptionMongoCommandException在其Result字段中包含一个BsonDocument属性。此文档包含一个StatusCode,您可以使用它来检测429。 也就是说,根据我的测试,我还发现我必须处理Http服务不可用(1)和操作超出时间限制(50)的状态码。
  • 对于V3.6端点:您可能还想处理MongoWriteExceptionMongoBulkWriteException。这些异常在异常消息中包括RetryAfterMs = (不总是)。不幸的是,这个值似乎并没有直接通过类属性公开 - 最可能是因为这是CosmosDB特定功能,因此不适用于MongoDB驱动程序定义的异常。
下面的代码实现了.NET Standard 2.0,并应该为您提供了一个很好的起点。根据您的情况和测试,您肯定需要调整一些常量。
    public static class Policies
    {
        public const int HttpThrottleErrorCode = 429;
        public const int HttpServiceIsUnavailable = 1;
        public const int HttpOperationExceededTimeLimit = 50;
        public const int RateLimitCode = 16500;
        public const string RetryAfterToken = "RetryAfterMs=";
        public const int MaxRetries = 10;
        public static readonly int RetryAfterTokenLength = RetryAfterToken.Length;

        private static readonly Random JitterSeed = new Random();

        public static readonly IAsyncPolicy NoPolicy = Policy.NoOpAsync();

        public static Func<int, TimeSpan> SleepDurationProviderWithJitter(double exponentialBackoffInSeconds, int maxBackoffTimeInSeconds) => retryAttempt
            => TimeSpan.FromSeconds(Math.Min(Math.Pow(exponentialBackoffInSeconds, retryAttempt), maxBackoffTimeInSeconds)) // exponential back-off: 2, 4, 8 etc
               + TimeSpan.FromMilliseconds(JitterSeed.Next(0, 1000)); // plus some jitter: up to 1 second

        public static readonly Func<int, TimeSpan> DefaultSleepDurationProviderWithJitter =
            SleepDurationProviderWithJitter(1.5, 23);


        public static readonly IAsyncPolicy MongoCommandExceptionPolicy = Policy
            .Handle<MongoCommandException>(e =>
            {
                if (e.Code != RateLimitCode || !(e.Result is BsonDocument bsonDocument))
                {
                    return false;
                }

                if (bsonDocument.TryGetValue("StatusCode", out var statusCode) && statusCode.IsInt32)
                {
                    switch (statusCode.AsInt32)
                    {
                        case HttpThrottleErrorCode:
                        case HttpServiceIsUnavailable:
                        case HttpOperationExceededTimeLimit:
                            return true;
                        default:
                            return false;
                    }
                }

                if (bsonDocument.TryGetValue("IsValid", out var isValid) && isValid.IsBoolean)
                {
                    return isValid.AsBoolean;
                }

                return true;
            })
            .WaitAndRetryAsync(
                retryCount: MaxRetries,
                DefaultSleepDurationProviderWithJitter
            );

        public static readonly IAsyncPolicy ExecutionTimeoutPolicy = Policy
            .Handle<MongoExecutionTimeoutException>(e =>
                e.Code == RateLimitCode || e.Code == HttpOperationExceededTimeLimit
            )
            .WaitAndRetryAsync(
                retryCount: MaxRetries,
                DefaultSleepDurationProviderWithJitter
            );

        public static readonly IAsyncPolicy MongoWriteExceptionPolicy = Policy
            .Handle<MongoWriteException>(e =>
            {
                return e.WriteError?.Code == RateLimitCode
                       || (e.InnerException is MongoBulkWriteException bulkException &&
                           bulkException.WriteErrors.Any(error => error.Code == RateLimitCode));
            })
            .WaitAndRetryAsync(
                retryCount: MaxRetries,
                sleepDurationProvider: (retryAttempt, e, ctx) =>
                {
                    var timeToWaitInMs = ExtractTimeToWait(e.Message);
                    if (!timeToWaitInMs.HasValue && e.InnerException != null)
                    {
                        timeToWaitInMs = ExtractTimeToWait(e.InnerException.Message);
                    }
                    return timeToWaitInMs ?? DefaultSleepDurationProviderWithJitter(retryAttempt);
                },
                onRetryAsync: (e, ts, i, ctx) => Task.CompletedTask
            );

        public static readonly IAsyncPolicy MongoBulkWriteExceptionPolicy = Policy
            .Handle<MongoBulkWriteException>(e =>
            {
                return e.WriteErrors.Any(error => error.Code == RateLimitCode);
            })
            .WaitAndRetryAsync(
                retryCount: MaxRetries,
                sleepDurationProvider: (retryAttempt, e, ctx) =>
                {
                    var timeToWaitInMs = ExtractTimeToWait(e.Message);
                    return timeToWaitInMs ?? DefaultSleepDurationProviderWithJitter(retryAttempt);
                },
                onRetryAsync: (e, ts, i, ctx) => Task.CompletedTask
            );

        /// <summary>
        /// It doesn't seem like RetryAfterMs is a property value - so unfortunately, we have to extract it from a string... (crazy??!)
        /// </summary>
        private static TimeSpan? ExtractTimeToWait(string messageToParse)
        {
            var retryPos = messageToParse.IndexOf(RetryAfterToken, StringComparison.OrdinalIgnoreCase);
            if (retryPos >= 0)
            {
                retryPos += RetryAfterTokenLength;
                var endPos = messageToParse.IndexOf(',', retryPos);
                if (endPos > 0)
                {
                    var timeToWaitInMsString = messageToParse.Substring(retryPos, endPos - retryPos);
                    if (Int32.TryParse(timeToWaitInMsString, out int timeToWaitInMs))
                    {
                        return TimeSpan.FromMilliseconds(timeToWaitInMs)
                               + TimeSpan.FromMilliseconds(JitterSeed.Next(100, 1000));
                    }
                }
            }
            return default;
        }

        /// <summary>
        /// Use this policy if your CosmosDB MongoDB endpoint is V3.2
        /// </summary>
        public static readonly IAsyncPolicy DefaultPolicyForMongo3_2 = Policy.WrapAsync(MongoCommandExceptionPolicy, ExecutionTimeoutPolicy);

        /// <summary>
        /// Use this policy if your CosmosDB MongoDB endpoint is V3.6 or V3.2
        /// </summary>
        public static readonly IAsyncPolicy DefaultPolicyForMongo3_6 = Policy.WrapAsync(MongoCommandExceptionPolicy, ExecutionTimeoutPolicy, MongoWriteExceptionPolicy, MongoBulkWriteExceptionPolicy);
    }

    public static IAsyncPolicy DefaultPolicy { get; set; } = Policies.DefaultPolicyForMongo3_6;

网页内容由stack overflow 提供, 点击上面的
可以查看英文原文,
原文链接