N8n memory error

Questions
,
1

Hi everyone,

I’m running into memory issues on n8n Cloud and hoping to get some clarity.

Use case

  • Lead enrichment workflows

  • 5k–10k leads per run

  • ~20 data points per lead

  • Includes AI node for lead personalization + Code node

  • Takes several hours to complete (3-6 hours)

  • Often running multiple workflows in parallel (onboarding multiple clients per day)

Problem
We regularly hit memory errors, especially:

  • with large batches

  • when running 2+ workflows at the same time

What I’ve already tried

  • Using Split in Batches

  • Moving parts into sub-workflows

  • Reducing payload size where possible

These help a bit, but don’t fully solve it - especially under parallel running workflows. One active execution works, but we need to be able to support multiple at the same time.

Questions

  1. n8n Cloud memory limits
    I’ve seen mentions of:

    • ~320MB (Starter)

    • ~640MB (Pro)
      Are these still accurate? And are these hard limits per instance/execution?

  2. Self-hosting
    If we move to a VPS (e.g. 4GB or 8GB RAM):

    • Do we effectively get access to that memory for workflows?

    • Or are there internal limits/overhead we should consider?

    • Will this realistically solve issues like ours?

We need to support multiple high-volume workflows daily that typically take 3-6 hours to complete.

Thanks a lot!

4

Hi @Sam_Bastiaens consider getting yourself a self hosted enterprise level license n8n instance, with the maxed out hardware config, as that is what will straightaway solve your problem. And yes those measures are accurate, n8n very impressively handles the memory management but still you could do at least you could toggle some settings in the n8n env which you cannot do on cloud. So get yourself a VPS and everything would really fall into place.

5

@Sam_Bastiaens the real fix here is chunking via executeWorkflow so each batch runs in its own execution and memory gets freed between chunks, here’s a pattern that does exactly that

put your actual enrichment + AI logic in a separate sub-workflow, set ENRICH_WORKFLOW_ID in your variables, batch size 200 keeps each execution well under the 320mb ceiling even on cloud.