Back-pressure refers to the resistance or force opposing the desired flow of fluid through pipes.
We can tweak this definition to refer to the flow of data within software. In this context, back-pressure refers to the resistance or force opposing the desired flow of data through software.
For further clarity on back-pressure, consider the example of a processing node that takes some input, processes the input, and returns the output.
Suppose that the incoming flow rate of the data is greater than the outgoing flow rate. There can be several reasons for this. For example, perhaps the processing speed of the node is slow. As the incoming rate is greater than the outgoing rate, pressure will be exerted on the node. This is called back-pressure.
Below, we discuss techniques that we can use to solve the problem of back-pressure in distributed systems.
If the callee is facing back-pressure, it can send a message to the caller to notify them about the back-pressure at the callee. This is the most efficient solution for back-pressure.
Another technique to exert back-pressure is to block the caller. If there is back-pressure, the callee can block callers. In this case, the callers will get an error message. An advantage of this technique is that the callee will handle the error itself. The callee will not have to send a message, as it is always stuck in back-pressure.
The requests from the caller are kept in a queue and the caller takes requests from this queue to process them. The caller can drop requests from this queue in case of back-pressure. The caller can drop the newest or the oldest requests, depending on the strategy chosen by the developer.
Another technique to solve back-pressure is to scale up the callee, which means to increase the processing power of the callee, either by its replacement with a high-speed callee or by the addition of another callee. Scaling will allow the callee to respond to the requests faster and thus prevent back-pressure.
View all Courses