Hi! Duda Nogueira from Weaviate here.

Looks like this issue is due to the transformer inference container timing out when Weavaite will request a vector for your batch import.

This timeout is by default 50s, and can be changed as documented here:

https://weaviate.io/developers/weaviate/config-refs/env-vars

with the environment variable `MODULES_CLIENT_TIMEOUT`

Note: whenever you start increasing timeouts in Weaviate, is probably a sign that you have resource allocation issues.

I can see from your docker compose that you are not using GPU (ENABLE_CUDA: 0) so my bet is that your transformer is not being able to handle the vectorization load.

What you can do to mitigate this is to have a fine control over you batches. Try sending less objects per batch, in order to allow the transformer to return in less than 50 seconds.

Let me know if that helps!

Thanks!