Why does “Continue (using error output)” on OpenAI v2 not pass the full error body to the next node?

[mohamed3nan]

Hi @Jelle

Yes indeed, I thought this would work until I tested it myself.

So as far as I know, there are two possible solutions I’m thinking of right now:

1. Use the HTTP Request node to call the OpenAI API, enable the “Never Error” option, then parse the response as described here:

How to get full HTTP request error message?

I had the exact same issue but with various other nodes. It seems that it is not specifically linked to the Xero API, but a limitation of how the nodes based on HTTP requests are handling the error messages when the error is caused by a HTTP response code (4xx, 5xx,…). Somehow, it does not return the full response but perform some sort of parsing and only provide some parts of the HTTP response body. This typically occur when the JSON returned by the query contain more than one message key or w…

2. Continue using the OpenAI node but use it in a sub-workflow, and set its On Error setting to “Stop Workflow”:

Then, in the main workflow, configure the sub-workflow’s setting On Error > Continue (using error output), then get the execution ID of the sub-workflow, use it to get execution information, and extract the full error object so that you can parse it as desired.

here is an example:

I know it seems like a complicated workaround, but this is what came to mind until there is a simpler built-in feature for this..