Overlay Transport Virtualization (OTV) will allow you to stretch L2 between the data centers while maintaining separate failure domains and giving exiting traffic in each VLAN its own, local gateway to leave the data center. You could control the return path of the traffic using OSPF and putting a higher cost on one path than the other with the "ip ospf cost" command.
In the solution above, I'm assuming one Data Center is active and the other is a standby, in case of a failure. In that scenario, all services have to run in one data center or the other. If you need both to be simultaneously active, then you need something like Locator/ID Separation Protocol (LISP) to solve that problem. You could still stretch the DCs with OTV or VXLAN, but wouldn't use OSPF the same way as above.
