Allow to pass HttpClient Factory and custom TcpClient Factory to GraphiteClient in order to avoid exhausting the connection pool

[MaximKojineMojio]

Assume you have an SF application with 20 microservices running on the same node.
And each of these microservice writes 1000 datapoints every 2 seconds. Windows OS has default setting [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\ Services\Tcpip\ Parameters\ TcpTimedWaitDelay] 240 seconds. Eventually, you will experience TCP port exhaustion.

    private async Task SendInternalAsync(ICollection<Datapoint> datapoints, CancellationToken cancellationToken)
    {
        TcpClient client;
        if (UseDualStack)
        {
            **client = new TcpClient(AddressFamily.InterNetworkV6) {Client = {DualMode = true}};**
        }
        else
        {
            **client = new TcpClient();**
        }
        using (client)
        {
            await client.ConnectAsync(Host, Formatter.Port).ConfigureAwait(false);
            cancellationToken.ThrowIfCancellationRequested();
            using (var stream = client.GetStream())
            {
                await Formatter.WriteAsync(stream, datapoints, cancellationToken).ConfigureAwait(false);
            }
        }
    }

[zivillian]

Have you tried increasing the GraphiteClient.BatchSize? Maybe you should add some kind of aggregator like StatsD .

[MaximKojineMojio]

We did.... And it didn't help. But it might not help because we didn't increase it enough.
We are going to apply a few adjustments

1. Adjust the TcpTimedWaitDelay setting
   https://docs.microsoft.com/en-us/biztalk/technical-guides/settings-that-can-be-modified-to-improve-network-performance
2. Increase BatchSize and decrease Flush Period

[zivillian]

I've implemented a first version of connection pooling 3ab559d
Please let me know, if this would solve your problem.