The primary focus here was the multi-orchestrator setup. In such cases, the possible orchestrators compete for a lock in redis. The orchestrator which manages to claim that lock becomes active, with others being kept around as hot spares, doing nothing. If connectivity to redis is lost and restored then either:
a) originally active orchestrator remains active, others remain passive
b) if originally active orchestrator goes away completely, one of the passive orchestrators should step up and become active

Currently in situation b), the passives will go into a half-broken state, where the process stays alive, but will not try to claim the lock anymore.