On Fri, Sep 02, 2005 at 02:41:46PM +0100, Tony Finch wrote: > Regardless of queue runner start intervals, the system is
> self-synchronizing.
>
> (1) receiving machine recovers from outage
>
> (2) a queue runner on a sending machine successfully delivers a message,
> then based on the hints db starts pumping messages down the connection
>
> (3) at various intervals other queue runners on the sending machines do
> the same, increasing the receiver's load by stages
>
> (4) when the load is too high, the receiver starts deferring *all* senders
> at the *same* time
>
> (5) thus they put the same retry time in their databases
If the queue runner interval is small enough, yes, that's exactly the
problem.
