Reconnection is attempted continuously regardless of ReconnectRetryPolicy

I'm creating my connection like this:

_readConn = ConnectionMultiplexer.Connect(
    new ConfigurationOptions
    {
        EndPoints = { { "localhost", 6380 } },
        Password = dbPass,
        Ssl = false,
        AbortOnConnectFail = true,
        ConnectRetry = 3,
        ReconnectRetryPolicy = new MyPolicy(),
        ConnectTimeout = 2000,
        KeepAlive = 5,
    }
);
_readConn.ConnectionFailed += (object? sender, ConnectionFailedEventArgs args) =>
{
    _logger.LogDebug("ConnectionFailed");
};
_readConn.ConnectionRestored += (object? sender, ConnectionFailedEventArgs args) =>
{
    _logger.LogDebug("ConnectionRestored");
};


public class MyPolicy : IReconnectRetryPolicy
{
    public MyPolicy() { }
    public bool ShouldRetry(long currentRetryCount, int timeElapsedMillisecondsSinceLastRetry)
    {
        return false;
    }
}

I then do two different tests:

• I drop all packets between my .NET microservice and the database using clumsy
• I kill the database service

in both cases the multiplexer continuously attempts to reconnect even though the policy says not to.

I can verify this by:

• putting a breakpoint on PhysicalBridge.TryConnect (it's fired approx every second or so)
• bringing the DB back up / stopping network degradation in clumsy (the connection is immediately restored)

Inspecting the code a bit more, it looks like when the connection is lost, ReadFromPipe will call OnDisconnected, which will then call TryConnect, which, when that connection attempt fails (inside BeginConnectAsync) will once again call OnDisconnected, etc etc.

Shouldn't the policy be applied here and prevent any new connections? Or am I misunderstanding and the reconnect policy is meant for some other case?

Start by tracing the reported path through PhysicalBridge.TryConnect, ReadFromPipe, OnDisconnected, and BeginConnectAsync while reproducing both packet loss and a stopped Redis service. Compare where ReconnectRetryPolicy is consulted with the repeated connection attempts, and use the existing connection-failure behavior as the completion check: returning false should no longer allow continuous retries.