Quorum turns “I cannot see the other machines” into a safe decision. Instead of every surviving node trying to continue, the cluster counts votes. The side with enough votes may keep service or promote a replacement. The side without enough votes should stop, become read-only, or wait. This is how many systems avoid split-brain during network partitions.
For example, a three-node cluster can usually continue when one node disappears because two voters remain. A two-node cluster is harder: if the link breaks, each side sees one vote and neither can safely know it is the majority. That is why two-node designs often add a quorum witness, qdevice, cloud witness, or another tiebreaking mechanism. Without it, operators may be tempted to force service back and accidentally create two primaries.
Quorum is not a magic number. It must match the failure domains, network shape, and operational process of the service. The rule should be documented together with what the minority side is expected to do.