I’m using Gemini for the LLM, I’m noticing a lot of errors in the logs and it appears to be coming from a embeddings backfill task.
Job exception: Net::HTTPBadResponse
/var/www/discourse/plugins/discourse-ai/app/jobs/scheduled/embeddings_backfill.rb:122:in `populate_topic_embeddings’
Based on the architecture diagram above, it should only run on a new topic but it appears that there’s a backfill task which runs on existing topics as well. That in itself it actually a good thing since it helps existing topics as well.
My clarification is related to limits. If Gemini is limited to 60 queries per minute, I suspect that the threshold may be causing the errors (is there a way to confirm?).
When a threshold is hit, does discourse retry to embeddings for the failed topic after a while? If so, how often and for how long? If not, is there a way to manually trigger the backfill task for related topics to go back and handle the unprocessed topics?