I am using the variable KETTLE_EMPTY_STRING_DIFFERS_FROM_NULL=Y in kettle.properties and when I run a job directly from the spoon interface it works perfectly.

When I run the exact job using kitchen I get "cannot insert null" errors from the database because of the non null varchar columns that are being converted to empty string are not working as intended.

In the job I am also using a set variables step and in the properties file section I wrote the full path to the properties file. Using "Valid in the Java Virtual Machine scope".

Any idea what I am missing?

Using PDI, JRE 1.8.0_131 and Win10.