[BUG] Can't pass file to Unstructured file loader for upsertion

[kennyakers]

Describe the bug
When trying to upsert a document via the Unstructured File Loader via the Flowise API, the file attached in the POST call is ignored, and the file specified in the node in the UI is uploaded instead. This was previously fixed for the PDF loader (thank you!), but is still present for Unstructured File Loader.

In this call, a PDF of a short story (several pages long) is attached. As you can see in the response, the file that was upserted was the file "blank.pdf", which is a single page with the words "ERROR" on it that was uploaded directly into the Unstructured File Loader node:

Expected Behavior
The attached file (above: Girl Jamaica Kincaid.pdf) should replace the file in the node (above: blank.pdf) and be upserted.

Flow & Config
Below is an example flow and override configuration to recreate this behavior:

Setup

• Installation: RepoCloud instance
• Flowise v2.2.2

[HenryHengZJ]

I was able to override the files. But you need to enable the override configuration:

1.) I have a chatflow with doc1.txt uploaded:

2.) Then, from the API, I override the files with another file:

[kennyakers]

@HenryHengZJ Thanks for taking a look at this. I also was able to replicate your results with .txt files - I believe I've narrowed down the issue: PDFs specifically cause this problem. For example:

1. Using the same flow, which worked with the .txt files as in your example

2. ePub file successfully upserts

3. PDF file still upserts the "doc1.txt" file uploaded in the flow

[HenryHengZJ]

thanks for testing it!

found the fix, will be added to this PR (b2ce4fa)

[saatchi-david]

@HenryHengZJ , I'm testing this now and finding the same error occurs when uploading a file via the ui/embed. The file set in the doc loader is the only file that gets passed, regardless of the filetype. Would you please have a look? Let me know if helpful to open a new issue.