Problems associated with long-lasting partitions
Instant Connect’s georedundant clustering mechanism is not intended to operate in a partitioned environment for extended periods of time. In addition to the previously discussed anomalies, the longer two sites are partitioned, the greater the chance for data inconsistencies to emerge, namely:
Kafka messages are discarded after approximately two hours. Messages used to signal a user’s client to take some action (for example, a message to display an intercom channel on the dashboard) will simply never be delivered if sites are partitioned for a long period of time. Users will need to log out and log back in to 'resync' these state inconsistencies.
Cassandra nodes may need to be repaired by invoking the nodetool repair command on them. Instant Connect Enterprise automatically performs this step periodically but system administrators may wish to initiate this immediately following a long partition.
Elasticsearch results (those displayed in people/channel search operations) may produce different results and/or incorrect online/offline status indications.