ssbc/go-netwrap

go-netwrap's Issues

There are no timeouts for network operations.

I think you can set a deadline on net.Conn instances like conn.SetDeadline(time.Now().add(Duration{100000000})). The Go devs in their infinite wisdom make you call "SetDuration" over again every time there is a timeout error which is kind of dumb, but without any sort of timeout, I think every network operation that has a failure on the remote end leaves its goroutine hanging forever waiting for that dead connection to work again.

I noticed this trying out "go-ssb." When I left it running for a while, it accumulated a gigabyte of RAM doing basically nothing, and then when I killed the program, tens of thousands of identical error messages flooded the console, along the lines of "fetchFeed(%s:%d) gossip pump failed: %w" in plugins/gossip/handler.go of go-ssb. Then I had to hard kill the program, getting just as many tens of thousands of stack traces from goroutines going belly up in the process. Many of them had been sitting idle for an hour. I think every time "gossip pump" fails, it just hangs the goroutine and leaves all that memory leaking out like a sieve, since there are no timeouts for network operations anywhere I can see.

Needs LICENSE

WithDialer?

In some cases you need to ioctl something on the filedescriptor or simply don't want to use std net.Dial. ConnWrappers only take effect after the fact the connection was opened. ioctl also doesn't work for the Serve() case.

type Dialer interface{
  Dial(net, addr string) (net.Conn, error)
}

func DialWithDialer(d Dialer, addr net.Addr, wrappers ...netwrap.ConnWrapper) (net.Conn, error) {
	conn, err := d.Dial(addr.Network(), addr.String())
	if err != nil {
		return nil, errors.Wrap(err, "error dialing")
	}

	for _, cw := range wrappers {
		conn, err = cw(conn)
		if err != nil {
			return nil, errors.Wrap(err, "error wrapping connection")
		}
	}

	return conn, nil
}

I exported the Dialer func as a type in v0.0.2 as quick remedy but would like a more general solution.