Dealing with failure
Earlier in this chapter, we discussed the CAP theorem and the concept of having to choose which compromises in our system to make. Alongside this, we must expect that our distributed system will fail due to both factors outside of our control and edge-case failure modes that we accept can happen from time to time, but we accept that risk in favor of delivery speed. Next, we will discuss some patterns we can put in place to mitigate some of these failures.
Two-phase commit (2PC)
As we discussed earlier, consistency is equally (if not more so) important in a distributed system as it is in a monolithic architecture. However, it is near impossible to create distributed transactions and commit atomically. One approach to solve this is to split our work into two phases:
- Preparation phase: We ask each of our sub-systems if it can promise to do the workload we want to complete.
- Completion phase: Tell each sub-system to do the work it just promised to...
