Custodian requires MonsterDB and Java 1.8 as a prerequisites, these are available from the links on the installation pages. But it is essential to validate the versions being used are correct. The major version number of the monsterDB and Custodian HAVE to be the same, in most cases the monster and custodian code has been aligned specifically to ensure that new features required by custodian are built into the same version numbers. To ensure this look for the version number of the installation you are using and upgrade if necessary. Databases do not need to be upgraded between versions and will always be upwardly compatible.
Custodian does not need to be run on multiple machines, although it can be, it will always use the metadata and data from one database and you should plan your installation that metadata is not changed in a cluster without the metadata cache being cleared from the other Custodian machines, please see the advanced section on how to clear the metadata cache in the event of changes like this.
In some cases it is important to store data on machines that are geographically separated or maybe for performance reasons you have too much data for one physical machine and wish to split the data and processing across a group of them. MonsterDB will handle the sharding of this data at the database level and it has nothing to do with the Custodian server. However you should be aware that once you have started to load data into a custodian server running on a monsterDB database that isn’t sharded and then decide it should have been, although monster will happily allow for this situation, there will be times when processing will take considerably longer as it tries to relocate documents from the original machine to its new home one-by-one!
So plan your database installation in advance during the design phase of your project, it will save time later on!