1- Number of data sources:
If you are upgrading but would like to increase or decrease the number of data sources you must do so when installing the latest version. For example, if you have a Primavera Data Warehouse setup
with 4 data sources in 15.1 and you are upgrading to 15.2 but only choose 3 data sources during the install you will see issues trying to upgrade. The reason being you are trying to force 4
data sources worth of data into 3. During the installation\configuration there is a box that allows you to enter the number of data sources, by default in 15.2 it was set to 3 so it must be increased
if the data warehouse you are upgrading has more. Even if you don't intend on using those extra data sources from 15.1 they must initially come over or be removed from the 15.1 data warehouse first. It is always safer to set this number of data sources option to a higher number than expected to make sure to cover this scenario and to allow for growth.
2- upgradeSchema vs. upgradeStar
Generally upgradeSchema is a faster option. UpgradeSchema usually takes the existing STAR schema and applies just the database level changes (new tables, new columns, etc.) but does not back up or move any data. UpgradeStar makes a backup of all historical data then rebuilds the STAR schema and reinserts the data and runs an ETL. UpgradeStar will usually take longer than a normal ETL run because it has to backup and restore all this data. In the 15.2 release because there was a major expansion of partitioning upgradeSchema will have a similar runtime to upgradeStar because we need to backup all the data, create new partitioned tables, and then reinsert the data. Moving forward and in previous releases upgradeSchema is generally the faster option and can be run and allow the ETL to run on it's regularly scheduled time.
3- Partitioning vs. Non Partitioning
If you were using a partitioned schema you can not now go to a non partitioned schema as the data is already stored this way. You would need to install and run the ETL processes fresh. With the non partitioning option you would also not have the option for detailed history or slowly changing dimensions so that data that might be in the partitioned schema would not be used anyway.