Workflow interrupted with connection lost/offline during large Loop Over Items run

Questions
1

Hi everyone,

I am on n8n Cloud Starter, version 2.17.5.
A workflow processing ~120–400 items through Loop Over Items and HTTP Requests gets “connection lost / offline” and the execution is interrupted.
The execution data is unreliable or not saved properly.
This happens even after reducing payload and using Data Tables.
Can you confirm if this is a Cloud resource/timeout issue and whether my plan can support this workload?

2

Hi @kourG, welcome!
I would say this number can be a limitation to your cloud starter plan if your workflow is not optimised, although 120-400 is still a lot for starter plan compute access:

Here is some guidance you can consider for optimising that flow:

Upgrading your plan can also fix that, but it depends on your use case; if it grows, you might need to go even higher.

3

On Cloud Starter, 120 to 400 items through Loop Over Items plus HTTP Requests hitting “connection lost” is usually the execution exceeding the plan’s memory or time limit rather than your internet, especially if each item carries a sizable payload, because n8n holds execution data in memory through the run.

A few things that help. Reduce what each item carries before the loop, drop fields you do not need so the in-memory payload stays small, the loop holds all of it. Process in smaller sub-batches and write results out as you go (to a Data Table or external store) instead of accumulating everything in one execution, so a failure midway does not lose the whole run. And consider splitting the work: one workflow that queues the items, a second triggered per batch, so no single execution has to hold 400 items of state.

The “execution data unreliable or not saved” part is the real risk for you, because a run that dies partway through a loop has often already done half the HTTP writes, and you cannot tell which from the broken execution. Make the writes idempotent and track which items completed in an external table, so a re-run skips the ones already done rather than duplicating them. Is it failing at a consistent item count, or randomly? Consistent points at the memory ceiling.

4

Welcome to the n8n community @kourG

Could you please update to the stable version 2.23.4 and let us know if the behavior persists?

5

Thank you for the replies.

I have already tried reducing the payload size and storing intermediate data in Data Tables. However, when the “connection lost / offline” issue occurs, no data seems to remain in the table, as if the execution never completed or the intermediate results were never saved.

Because of this, I am trying to understand whether the problem is related to a Cloud Starter resource limit or execution timeout.

Regarding the suggestion to use batches, if I understand correctly, you are recommending that I split the 120–400 items into smaller groups and process them incrementally instead of in a single large execution. I am also considering whether it would be possible to load a new batch after the current batch has finished processing in the loop, so the workflow can continue in smaller chunks. I will experiment with this approach and see whether it improves stability.

Also, I am using n8n Cloud Starter and do not have direct control over upgrading the n8n version, since the version is managed by n8n Cloud. Could you please confirm whether version 2.23.4 is already available on Cloud, or whether any action is required from the n8n team?

6

Yes, but make each batch its own execution. A Loop Over Items that loads the next batch inside the same run still shares one timeout/memory envelope, so when that run is interrupted the checkpoint can disappear with it.

Use one run to claim a small batch, write each item result/status before moving to the next item, then let a Schedule or Execute Workflow start the next batch from rows still marked pending. For the Data Table symptom, post where the write node sits: before the HTTP Request, after each item, or only after the whole loop finishes?