Read this chapter to learn more about what to do if two (out of three) nodes are offline.

In case of planned shutdown, the remaining node still accepts update commands.

In case of unplanned shutdown, the cluster goes into a "split brain" situation and the remaining primary node accepts only read commands until at least one of the offline nodes rejoins the cluster again.

For more information, please see section Updating as well as chapter Restart Cluster Sequence.