If all gateways are down, 1) replication does not proceed, and 2) an alarm fires to warn the administrator. Since replication cannot proceed, the bucket files cannot be deleted (because they contain the unreplicated updates). Therefore, there isn’t much impact other than storage consumption on the source cluster. However, once the gateways are restored, the backlog of unreplicated data must be pushed. This is costly. How costly depends on how much data has not been replicated.
In addition, if replication does not proceed for some time, pending deletes might be lost. This might result in data, which should have been deleted in a replica, not being deleted. By default, pending deletes are kept for 24 hours. If all gateways are down (and they might be down for some time), you should increase the table edit deletettl value enough to ensure that all deletes are replicated eventually. The deletettl value is set in the MapR CLI. See table edit in the MapR user documentation.
The table edit deletettl command takes effect immediately, and it applies to deletes already in the backlog so you don’t have to fret about losing previous deletes; that is, as long as you change the deletettl before the current value (24 hours by default) has passed.
Retrieving data ...