I’m trying to set up a data store using Google Cloud Vertex AI Agent Builder for my chatbot project. I uploaded a CSV file with question and answer pairs first, then tried to change the schema settings afterward.
When I attempted to designate the “Question” column as the primary key field, I got 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 the system won’t let me modify the key property after documents are already imported. Has anyone run into this issue before? Is there a way to update the schema without having to delete and re-upload all my data?
Changing the schema in Google Cloud can be tricky after you’ve imported documents. I learned that the hard way.
Best way to tackle this? Start over with a fresh data store and set the “Question” as your primary key first before uploading data.
If you’re set on using the existing data store, you’ll have to back everything up, wipe it clean, update the schema to adjust the key property, and then re-import your CSV.
Quick tip: always define your schema completely before uploading. I usually test with a few rows first to avoid issues.
Yes, the design of Google Cloud Vertex AI Agent Builder restricts schema changes once active documents exist in your data store, particularly with key properties. This limitation prevents inconsistencies in document relationships and indexing. I faced a similar challenge recently. The best approach is to establish a new data store initially with the appropriate schema. Before you upload your CSV, ensure that the Question column is designated as the key property. I found that exporting the current data via the console first, then creating the new data store correctly, helped avoid complications in the long run.
had the same prob too! it’s super frustrating. like, why can’t we just change the schema? but yeah, backup your stuff, delete the current data, update the schema, and reupload. totally a hassle but it’s what worked for me.
Unfortunately, Google Cloud Vertex AI Agent Builder locks this down once you’ve ingested documents - it’s designed to prevent indexing conflicts and protect data integrity. Once docs are in, you can’t modify key properties without corrupting existing relationships. I’ve dealt with similar data stores before, and honestly, your best bet is creating a fresh data store. Set up your schema right this time - make the Question column your primary key before uploading anything. Then migrate your CSV data over. Seems like extra work, but it’s actually faster than trying hacky workarounds that usually don’t work anyway. The system just wasn’t built for post-ingestion schema changes on key properties. Definitely a ‘measure twice, cut once’ situation - always lock down your schema before ingesting data.