One of my recent ODI projects is a relatively complex transformation job. I am effectively building up a master/detail set of records from a single table. The single table isn’t really a single table in the source, it’s multiple tables. Within ODI I make several passes on it, dialing in the fields with interfaces and procedures. I opted to use the in-memory engine (MEMORY_ENGINE) because I thought the architecture would be a little cleaner, and the amounts of data being pushed through are not huge.
Everything was fine, until I hit a legitimate ODI bug. I actually found a relevant case in Oracle support for it: ODI-1228 “statement is not in batch mode”. There was even a patch! Unfortunately, the patch required a version of ODI higher than what I had available. So on a tight deadline my choices were to push through an ODI upgrade or to find some workaround.
I decided to see if I had a low-cost option of switching from using the memory engine to just using an Oracle schema as a stage (note that the package and interfaces themselves are all just moving data between various Oracle servers, nothing Hyperion related even). So I went into my model for the staging table that I was using, and just switched it from the In-memory Engine to Oracle (using the drop down). No complaints from ODI there.
Next I went into one of the interfaces that was previously setup to have source/staging/target in terms of its whole process. I went straight to the Flow tab but had some issues and fun little NullPointerException errors which is always a fun time. The thing is, I changed a technology on a model being used in various interfaces but it’s not like any part of ODI went into those interfaces to say “Hey, this changed…” – in fact, when you change the technology of a model, ODI helpfully and plainly just says “Hey, this is likely to break stuff. Proceed at your own risk… THAR BE DRAGONS.” Or something like that.
Anyway, I found that I could sort of ‘kick’ the interface when I opened it, by checking the “Staging Area Other Than Target” option, then turning it off again (it was off in most of my interfaces). This forced the interface to sort of recalculate and reset the flow, which took into account the updated technology of the model. There might be a better way to do this rather than this “fuzzy” method, but it worked and I didn’t have to redo the plethora of interfaces in this package.
Hopefully this helps someone else out someday!