ThrottleTransportAdapter throws OverflowException if system up time is longer than a few days

[Aaronontheweb]

This bug was hilarious - never was able to find it on the build server because the system up time for each build agent is small (they're created on-demand,) but it's come up every time I've tried to run the Akka.Remote test suite on dev on my local machine since #1174 was merged in.

ThrottlerTransportAdapter_must_maintain_average_message_rate would fail, repeatedly, every time.

Here's the issue:

sealed class TokenBucket : ThrottleMode
{
    readonly int _capacity;
    readonly double _tokensPerSecond;
    readonly long _nanoTimeOfLastSend;
    readonly int _availableTokens;

    public TokenBucket(int capacity, double tokensPerSecond, long nanoTimeOfLastSend, int availableTokens)
    {
        _capacity = capacity;
        _tokensPerSecond = tokensPerSecond;
        _nanoTimeOfLastSend = nanoTimeOfLastSend;
        _availableTokens = availableTokens;
    }

    int TokensGenerated(long nanoTimeOfSend)
    {
        // OVERFLOW EXCEPTION!!!
        return Convert.ToInt32(((double)(nanoTimeOfSend - _nanoTimeOfLastSend) / TimeSpan.TicksPerMillisecond) * _tokensPerSecond / 1000);
    }
}

The reason we have an OverflowException occur here is because _nanoTimeOfLastSend is set to 0 on the first calculation, so the result of nanoTimeOfSend - _nanoTimeOfLastSend is equal to the entire system up time on the first calculation, which causes the OverflowException when we attempt to cast down to an int.

Way to fix this is to set the _nanoTimeOfLastSend equal to the current MontonicClock.Ticks value when the TokenBucket is created. That way the delta is going to be something smaller (like, seconds worth of ticks) that can be downcast.