The issue has been raised multiple times in the past. Sometimes as a request to implement a simple "ordered" approach to outbound selection, or as a request to add a separate strategy to the balancer - "firstAvailable". #3114 These requests have been declined. However, the desire to have a layered fallback scheme with a set of outbounds of different availability and traffic limits remains.
At the moment, there is a workaround for this issue. One can set up a complicated scheme with balancers, loopbacks, and routing rules to achieve the desired effect. However, this scheme is hard to understand for most users and is prone to mistakes.
I believe the demand for the direct fallback-based selection of outbounds can be met if the balancers are allowed to forward the traffic to another balancer in case none of the outbounds included in it are available. This approach makes it possible to have both the load-balancing within balancers, and strictly ordered fallback scheme across balancers at the same time. And most importantly, it greatly simplifies the configuration required to implement this scheme.
The issue has been raised multiple times in the past. Sometimes as a request to implement a simple "ordered" approach to outbound selection, or as a request to add a separate strategy to the balancer - "firstAvailable". #3114 These requests have been declined. However, the desire to have a layered fallback scheme with a set of outbounds of different availability and traffic limits remains.
At the moment, there is a workaround for this issue. One can set up a complicated scheme with balancers, loopbacks, and routing rules to achieve the desired effect. However, this scheme is hard to understand for most users and is prone to mistakes.
I believe the demand for the direct fallback-based selection of outbounds can be met if the balancers are allowed to forward the traffic to another balancer in case none of the outbounds included in it are available. This approach makes it possible to have both the load-balancing within balancers, and strictly ordered fallback scheme across balancers at the same time. And most importantly, it greatly simplifies the configuration required to implement this scheme.