I’m trying to build a chatbot using Google Cloud’s Vertex AI Agent Builder and running into a frustrating issue with schema configuration.
Here’s what happened: I created a new Data Store and uploaded my CSV file first. The file has two main columns - one for questions and one for answers. Everything seemed to upload fine.
But then I realized I needed to set the question column as my primary key field. When I went back to modify the schema settings, I got hit with this error message:
“Schema update doesn’t support adding the key property annotation for schema with active documents. Key property mapping mismatch for field ‘Question’.”
It seems like once you have data in your store, you can’t go back and change which field serves as the primary key. Has anyone found a workaround for this? Do I need to delete everything and start over with the schema configured properly first?
Unfortunately, you’ll need to start fresh. Vertex AI doesn’t allow changes to primary keys once documents have been uploaded to the datastore. I encountered this issue six months ago while creating a knowledge base. Here’s what I did: I deleted the existing datastore, created a new one, and configured the schema with the primary key first before re-uploading the CSV. While it’s frustrating, it only takes about 15 minutes unless dealing with a large dataset. It’s essential to set up your schema before any uploads; I learned that the hard way. Ensure that your question column is formatted as text and double-check the field mapping prior to uploading.
Yeah, this gotcha moment hits everyone at least once with Vertex AI. Hit this wall 8 months ago rushing a demo for stakeholders.
Vertex locks primary key changes the second you index any documents. No exceptions or workarounds that actually work.
My process now: Create datastore, set complete schema including primary key, then upload. I use a checklist because realizing you missed something halfway through sucks.
Tip that saves time - export your current data before deleting everything if you can. Upload formatting sometimes changes slightly, so having the processed version prevents weird edge cases on re-upload.
Rebuilds go faster the second time since you know what you want. Just make sure your question column has unique values or you’ll hit another error.
Hit this exact problem three weeks ago during a client migration. The primary key restriction is hardcoded in Vertex AI’s backend - once you index documents, it treats any schema changes with keys as breaking operations. I spent hours trying partial exports and incremental updates, but nothing worked. You’ve got to rebuild. One thing that saved me time: validate your CSV structure locally with pandas first. Check for duplicate keys and null values in your question column before uploading. This prevents failures after you recreate the datastore. Also, smaller CSV files upload way faster the second time around, so split large datasets if you’re going to be rebuilding regularly.
Been there - super annoying but there’s no workaround. Had to nuke my entire datastore last week for the same reason. Just bite the bullet and rebuild, but set up your primary key BEFORE uploading anything. Saves you this headache later.
This exact thing happened to me last month on a client project. Google’s docs aren’t clear about it, but once you’ve got documents in your datastore, you can’t change the primary key. I tried partial deletions, schema exports - nothing worked. You’ll need to rebuild the datastore from scratch. Export your schema config first if you’ve got other fields mapped correctly. You can use it as reference when setting up the new one. Also, check your CSV for duplicate values in the primary key column and clean those up, or you’ll hit another error during upload. Rebuilding sucks but it’s straightforward once you know the steps.