Creating an index from a parquet file

[AndyWDev]

Hi Folks
I'm just learning Microsoft Foundry and have created a connection to pick up a parquet file from an Azure Data Lake Gen2 share. The connection brings in the file ok, but when I try and create an index from it the 'crack and chunk' process fails due to unsupported file format (extension not recognized - the file just contains json in the appropriate format).

Searching through documentation online I can find examples of importing the parquet file and different examples of creating an index from say a PDF or CSV file - but I can't find any examples of creating an index from a parquet file - there seems to be a missing conversion step somewhere.

Does anyone have an example of how to handle the parquet file in order to create an index?

[leestott]

Hi @AndyWDev

Microsoft Foundry can store Parquet files as data assets, but it cannot crack and chunk Parquet files for indexing. The indexing pipeline only supports document-style formats (PDF, TXT, HTML, DOCX, CSV, etc.) and fails purely based on the .parquet extension even if the contents are JSON.

Workaround
You must convert Parquet into a supported document format before creating an index.

Recommended patterns:

Parquet → JSONL (1 row = 1 document) → index (best for structured data)
Parquet → CSV → index (simpler, but loses nested structure)
Bypass Foundry ingestion: read Parquet yourself, chunk + embed programmatically, and push directly to Azure AI Search

So guidance is treat Parquet as a staging/analytics format, not an indexable document format, until Foundry adds first‑class support.