Browser memory overflow due to large volume of processed documents

Questions
,
1

n8n local Docker
Win 11
Workflow with RAG subflow inserting 2000 PDF documents into PostGres Vector Store local Docker

Basically, I created a workflow that reads the list of documents, divides them into batches (I’ve tried 1, 5, 10, 15, and 20), and sends them to the sub-flow that processes the batch in the Postgres vector store. Completely local. I checked the memory and space, and during the process, it doesn’t take up more than 15%. I increased the number of connections in Postgres and increased its working memory. After approximately 70-120 documents (depending on the size of the documents), n8n simply restarts, completely clearing the execution. It doesn’t even leave the executions. I also disabled the preservation of the successful history of n8n executions. The only point I continue to use, which is recommended to change, is the manual trigger. So, I believe it is a memory control in the browser, not in Postgres or n8n in Docker. Finally, I tried Supabase.

Does anyone have a solution?

My principal workflow

image
image1832×749 49.8 KB

2

It is almost certainly not a browser memory issue, as the n8n execution runs on the server (Docker container), and the browser is only used for the UI/manual trigger initiation
my advice is to shift focus away from Postgres and look critically at the n8n process’s resource limits and how data is being handled in the loop.

and also for a task of this scale (2000 documents), a single synchronous n8n execution is brittle. The try to decouple the processing queue from the execution context.