Requests that resulted in permanent errors should not be attempted again about go-diameter

[jaroszan]

Closing since it is not really a bug

[jaroszan]

2 cases should be considered here:

1. CEA contains no app ids and 5010 code
2. CEA contains some app ids and 5010 code

[jaroszan]

This is a different issue than I initially thought and it is located in examples/client/client.go, client attempts to reconnect indefinitely regardless of type of error which happens in the progress.

for {
        c, err := connect()
        if err != nil {
            log.Println(err)
            backoff *= 2
            if backoff > 20 {
                backoff = 20
            }
            time.Sleep(time.Duration(backoff) * time.Second)
            continue
        }
        log.Println("Client connected, handshake ok")
        backoff = 1
        <-c.(diam.CloseNotifier).CloseNotify()
        log.Println("Client disconnected")
    }

One might guard oneself against such situations with something like this in case err != nil:

if err.Error()[len(err.Error())-4:] == "5010" {
                log.Println("Permenent error received, aborting connection attempts")
                break
}

But it does not look reliable (eg. err does not come necessarily from parsing CEA).

Such functionality (connect with backoff) can be potentially located somewhere inside sm/client.go and differentiate between type of errors and how to deal with them.

[fiorix]

That code in the example/client is an example, not something you'd use for real use cases.

One option is to implement the diam.Conn interface with specific client behavior, supporting things like auto reconnection (with back off), circuit breaker, and whatnot.