Error codes for timeout different per operating system

[joshbartley]

Describe the bug

I'm setting up Polly with SqlClient and using the error codes to track if a request should be retried or not. I'm testing resiliency with a network failure resulting in a BGP event that takes 2-3 minutes to resolve. I'm using the error codes since it's a number and not a string, but they seem to be different per operating system. My only thought could be that maybe once a connection is in the pool, connecting to a pooled connection that has a TCP timeout results in an error code of 0. Then it should still return some number saying that instead of 0.

Windows 10
If I point to a server that is not accessible, I get
"A network-related or instance-specific error occurred while establishing a connection to SQL Server. The server was not found or was not accessible."
Error Code: 53

Ubuntu 22.04
If I point to a server that is not accessible, I get
"A network-related or instance-specific error occurred while establishing a connection to SQL Server. The server was not found or was not accessible."
Error Code: 0

Exception message:
Stack trace:
A network-related or instance-specific error occurred while establishing a connection to SQL Server. The server was not found or was not accessible. Verify that the instance name is correct and that SQL Server is configured to allow remote connections. (provider: TCP Provider, error: 35 - An internal exception was caught)    at Microsoft.Data.ProviderBase.DbConnectionPool.TryGetConnection(DbConnection owningObject, UInt32 waitForMultipleObjectsTimeout, Boolean allowCreate, Boolean onlyOneCheckConnection, DbConnectionOptions userOptions, DbConnectionInternal& connection)

To reproduce

using (var sqlConn = new SqlConnection(connectionString))
            { 
                sqlConn.Open();    
                SqlCommand cmd = new SqlCommand("SELECT 1;", sqlConn);
                cmd.CommandType = System.Data.CommandType.Text;
                var rdr = cmd.ExecuteReader();
            }

using (var sqlConn = new SqlConnection(connectionString))
{
    await sqlConn.ExecuteAsyncWithRetry("SELECT 1");
}

Code for the dapper/Polly work from https://gist.github.com/hyrmn/ce124e9b1f50dbf9d241390ebc8f6df3 with the added error code 53 added to the case statement.

Expected behavior

I would expect that the Exception.Errors contain an error code of 53 and not 0 like it does on a Linux machine. SqlException.Number should be 53 as well thought I may not know all the details on how that is set.

Further technical details

Microsoft.Data.SqlClient version: 5.0.1
.NET target: 6.0
SQL Server version: SQL Server 2019
Operating system: Windows 10, Ubuntu 22.04

[lcheunglci]

Thanks @joshbartley for bringing this to our attention and providing a repro sample. I feel that it's similar to this issue #1773 except in your case, the exception message is still the same, but the error code differs. We'll take a look soon.

[lcheunglci]

The error 53 comes from the operating system on Windows. On Linux having error code 0 means that we were not able to resolve the error, and the error depends on the .NET API that we use, so it's by design that we cannot capture the same error code between different platforms.

[joshbartley]

This is not great and a sign of a leaky abstraction. Writing a cross platform app I have to check error strings instead of an error code for a can't connect message? How is exposing an error code of 0 helpful at all? This would be like an http client returning a status code 404 on windows but a 0 on linux for not found. This has downstream implications for Entity Framework's own retry logic which relies on error codes as well.