I recently set up a Retrieval-Augmented Generation (RAG) solution using AWS Bedrock Knowledge Base alongside serverless OpenSearch. We indexed about 800 medium-length web pages, which I thought was a pretty straightforward task. However, I was shocked to see our bill for last month was around $350! There was no new indexing during that period, and very few queries since this is more of an internal side project. Is OpenSearch truly this costly, or is there something I might be overlooking? I'm also curious about how alternatives like Qdrant or ChromaDB might compare in terms of pricing. Is there a more cost-effective way to manage this without breaking the bank?
5 Answers
Avoid AWS Bedrock KBs at all costs! They’re super expensive, don’t scale well, and can be quite slow. We had to switch to Weaviate just two weeks into production and it made a huge difference.
You might want to check out Amazon S3 Vectors; I've heard it can significantly cut costs down to about 90%! You should definitely explore that option.
That's a good suggestion, thanks for sharing!
Yeah, I've set up something similar for a client, and let me tell you, it was running about $800 a month! Looking into alternatives like Pinecone is definitely worth it.
Whoa, that's pricey! Thanks for the heads-up.
Honestly, it’s usually cheaper to just launch a regular OpenSearch Domain with an EC2 instance that suits your traffic needs. Serverless OpenSearch isn’t as "serverless" as it sounds; you still pay by the hour, and the baseline instance size is often overkill for smaller projects.
Great to know, I'll consider switching to a different setup.
Just so you know, OpenSearch serverless has a minimum charge of 2OCUs, which can easily add up to around $350 monthly, especially if you're not utilizing it fully. Looking into self-hosted options might save you a ton.
Thanks for the warning, I’ll keep that in mind.