Bypassing a load balancer in a return path of network traffic
Some embodiments provide a method that allows a first data compute node (DCN) to forward outgoing traffic to a second DCN directly in spite of receiving the incoming traffic from the second DCN through a load balancer. That is, the return traffic's network path from the first DCN to the second DCN bypasses the load balancer, even though a request that initiated the return traffic is received through the load balancer. The method receives a first data message from a load balancer to be sent to a DCN. After identifying a particular address embedded in the data message by the load balancer, the method generates a table entry, based on source and destination addresses of the data message and the identified address. This entry is used for modifying a source address of a subsequent data message received from the DCN in response to the data message.
1. A method for forwarding load balanced network traffic, the method comprising:
at a host computer executing a data compute node (DCN),
receiving, from a load balancer, a first data message that has (i) a first address associated with the DCN as a destination address and (ii) a source address identifying a source of the first data message;
identifying a second address embedded in the first data message by the load balancer, the second address originally specified as the destination address in a header of the first data message before being replaced with the first address in a load balancing operation of the load balancer;
based on (i) the source address and the first address of the first data message and (ii) the second address embedded in the first data message, generating a table entry to use subsequently to replace, with the second address, the first address that is specified as a source address of a subsequent data message sent by the DCN to the source of the first data message; and
forwarding the first data message towards the DCN.
2. The method of claim 1 further comprising storing the generated table entry with the second address in a local data storage of the host computer.
3. The method of claim 2 further comprising, prior to generating the table entry, determining that no table entry associated with the first data message exists in the local data storage.
4. The method of claim 1 , wherein
the DCN is one particular DCN in a group of DCNs that all perform a common operation, and
the first address is an internet protocol (IP) address of the DCN and the second address is a virtual IP address of the group of DCNs.
5. The method of claim 1 , wherein the source of the first data message is a client requesting data from a DCN in the group of DCNs.
6. The method of claim 5 further comprising forwarding the second data message towards the client via a path that bypasses the load balancer.
7. The method of claim 1 , wherein the table entry identifies a data flow based on the source and destination addresses of the first data message, by assigning the second address of the first data message as a source address of a reverse flow and assigning the source address of the first data message as a destination address of the reverse flow.
8. The method of claim 7 further comprising generating a second table entry for the data flow of the first data message using the source and destination addresses of the first data message.
9. The method of claim 1 , wherein identifying the second address comprises retrieving a set of values embedded in a particular header field of the first data message and matching the set of retrieved values to an entry in a mapping table, the entry further mapping the set of retrieved values to the second address.
10. The method of claim 1 , wherein the DCN comprises a virtual server of a set of virtual servers that implements a distributed application, wherein the first data message is for requesting data from the distributed application.
11. A non-transitory machine readable medium storing a program which when executed by at a set of processing units of a host computer forwards load balanced network traffic, the processing units of the host computer further executing a data compute node (DCN), the program comprising sets of instructions for:
receiving, from a load balancer, a first data message that has (i) a first address associated with the DCN as a destination address and (ii) a source address identifying a source of the first data message;
identifying a second address embedded in the first data message by the load balancer, the second address originally specified as the destination address in a header of the first data message before being replaced with the first address in a load balancing operation of the load balancer;
based on (i) the source address and the first address of the first data message and (ii) the second address embedded in the first data message, generating a table entry to use subsequently to replace with the second address, the first address that is specified as a source address of a subsequent data message sent by the DCN to the source of the first data message; and
forwarding the first data message towards the DCN.
12. The non-transitory machine readable medium of claim 11 , wherein
the DCN is one particular DCN in a group of DCNs that all perform a common operation, and
the first address is an internet protocol (IP) address of the DCN and the second address is a virtual IP address of the group of DCNs.
13. The non-transitory machine readable medium of claim 11 , wherein the source of the first data message is a client requesting data from a DCN in the group of DCNs.
14. The non-transitory machine readable medium of claim 13 , wherein the program further comprises a set of instructions for forwarding the second data message towards the client via a path that bypasses the load balancer.
15. The non-transitory machine readable medium of claim 11 , wherein the DCN is a particular virtual server in a set of virtual servers that implement a distributed application and the first address is a network address of the particular virtual server.
16. The non-transitory machine readable medium of claim 15 , wherein the second address is a virtual internet protocol (VIP) address associated with the distributed application.
17. The non-transitory machine readable medium of claim 16 , wherein the load balancer selects the particular virtual server over other virtual servers in the set based on a load balancing algorithm.