Controlling flow processing by an edge cluster spanning multiple datacenter locations of a public cloud

A non-transitory machine readable medium storing a program for execution by at least one processing unit for controlling data message flow processing by an edge cluster that comprises (i) a first set of edge machines operating in a first set of locations of a particular public cloud and (ii) a second set of edge machines operating in a second set of locations of the particular public cloud, the program comprising sets of instructions for:

• at a set of one or more controllers that controls the edge cluster:
• configuring first and second sets of managed forwarding elements (MFEs) operating in the first and second location sets respectively with first and second sets of forwarding rules to respectively forward first and second sets of data message flows to the first and second sets of edge machines for performing a set of services on the first and second sets of data message flows, the first set of forwarding rules specifying a first set of network addresses associated with the first set of edge machines and the second set of forwarding rules specifying a second set of network addresses associated with the second set of edge machines;
• configuring a first set of Top-of-Rack (ToR) switches operating in the first location set to forward a third set of data message flows to the first set of edge machines to perform the set of services on the third set data message flows;
• configuring a second set of ToR switches operating in the second location set to forward a fourth set of data message flows to the second set of edge machines to perform the set of services on the fourth set data message flows;
• monitoring each edge machine in the first and second sets of edge machines to determine whether the edge machine is available to perform the set of services by periodically sending heartbeat data messages to the edge machine;
• and
• after determining that each edge machine in the first set of edge machines is not available, reassigning the first set of network addresses from the first set of edge machines to the second set of edge machines such that the first MFE set forwards the first set of data message flows to the second set of edge machines based on the first set of forwarding rules.