So here's my case. I'm incluiding a topology diagram in the end for better understanding.

The network we're talking about is a ISP with two BGP Upstreams (660k routes each, full tables), so multi-homed and also a GGC/Netflix OCA Upstream.(about 50 routes)

However ISP1 and ISP2 are 300 miles away from each other. GGC/OCA is in the middle of the backbone, so they arent linked directly to each other, theres another routers, Wireless links, Fiberchannels, etc.

As described in the image, I need a way so I can forward the upload traffic of my customer with it's prefix announced on ISP2, to ISP2, and at the same time, maintaining the upload traffic of my customer announced on ISP1 to ISP1 aswell.

Pretty much having 2 default routes on every network "EDGE RT", as i described on the diagram.

However... having 2 default routes means ROUTING MARKS.

When I mark my customer thru a routing mark, the upload goes straight to the marked ISP, ignoring all the other local routes... that means my GGC in the middle of the backbone doesnt take upstream traffic anymore, or any of my peerings that are also in the middle of the backbone.

Is there a way that I can do something like "Use main table for local upstream and downstream traffic, if destination not found on local table, send through the routing mark to a ISP Upstream, cause it's obvioulsy internet traffic if the route is not found locally"

Anyone?

Not sure to understand your picture, but maybe you can solve the problem with command "/ip route rule" with some action like lookup on table main. Anyway, working with routing marks always difficults path traceability. Good luck.

Just updating on this,

A workaround Ive used was adding all prefixes received via BGP to a address-list.

Then I made a Mangle Rule with a accept action before the mark routing to other ISP based on the dst-address on that address list.

That way I could ignore the routing mark for local addresses received via bgp, however still it's not a good solution for the long term...

I've never done this on Mikrotik, but you could have the two ISPs in separate VRFs and the local routes and backbone peers in the main table.By using route target imports, you can leak the global table into the VRFs but keep the ISP learned prefixes separate from each other. No more routing marks. You'd need to set up BGP into multiprotocol and potentially also enable MPLS for any P routers to be able to reach the edges.

One question though, are the two IP ranges PA/PI space that you originate, or are they owned and originated by the ISPs? If they are your own prefixes, the easier option would be to advertise both prefixes to both ISPs, gain some fault tolerance and not worry about which upstream ISP is carrying any particular users traffic.