If the master cannot connect to a segment instance, it marks that segment as down in the Greenplum Database system catalog. The segment instance remains offline until an administrator takes steps to bring the segment back online. The process for recovering a failed segment instance or host depends on the failure cause and whether or not mirroring is enabled. A segment instance can be unavailable for many reasons:
postgresdatabase listener process.
Figure 1 shows the high-level steps for each of the preceding failure scenarios.
Parent topic: Enabling High Availability and Data Consistency Features