We have learned so far that in a distributed application, we have many interacting services. When Service A interacts with Service B, it happens through the exchange of data packets. These data packets need to somehow be funneled from Service A to Service B. This process of funneling the data packets from a source to a destination is also called routing. As authors or operators of an application, we do expect the orchestrator to take over this task of routing. As we will see in later chapters, routing can happen on different levels. It is like in real life. Suppose you're working in a big company in one of their office buildings. Now, you have a document that needs to be forwarded to another employee of the company. The internal post service will pick up the document from your outbox and take it to the post office located in the same building. If the target person works in the same building, the document can then be directly forwarded to that person. If, on the other hand, the person works in another building of the same block, the document will be forwarded to the post office in that target building, from where it is then distributed to the receiver through the internal post service. Thirdly, if the document is targeted at an employee working in another branch of the company located in a different city or even country, then the document is forwarded to an external postal service such as UPS, which will transport it to the target location, from where, once again, the internal post service takes over and delivers it to the recipient.
Similar things happen when routing data packets between application services running in containers. The source and target containers can be located on the same cluster node, which corresponds to the situation where both employees work in the same building. The target container can be running on a different cluster node, which corresponds to the situation where the two employees work in different buildings of the same block. Finally, the third situation is when a data packet comes from outside of the cluster and has to be routed to the target container running inside the cluster.
All these situations and more have to be handled by the orchestrator.