raoufchebri

Here is an article about PITR in Neon https://neon.tech/blog/point-in-time-recovery-in-postgres

Here is a step by step guide on how to deploy Mistral Large to Azure: https://neon.tech/blog/deploy-mistral-large-to-azure-and-cha...

Excited about the Microsoft partnership

Last November, we had several outages and have since directed much of our engineering efforts towards improving stability. During this phase, cold start latency regressed, but the root cause and magnitude were not immediately apparent to us. We made the call to continue working on stability and revisit the latency issue later.

Today, we want to share the story of tracking down and fixing that regression, which turned out to be in the Pageserver’s walredo mechanism.

Really cool. Do you see this being integrated into PostgreSQL in the future with the rise of open-source LLMs?

Neon is cloud-native Postgres with a storage system written in Rust and a persistence layer stored in AWS S3. Because the cloud storage layer is decoupled from the rest of the architecture, you can bring your own S3.

We’re excited to announce the release of logical replication on Neon to stream your data to external data stores, allowing for change data capture and real-time analytics.

How do you handle conflict with pgvector's hnsw if you want to install both extensions ?

CREATE INDEX semantic_image ON image_table USING hnsw (v dist_cos_ops) WITH (M=5, ef=30, ef_construction=30, dims=512);

Awesome idea! like the UI

The index doesn't need to fit in the shared buffers to be built, but you can speed the queries if you load it to memory

It's under Apache license

We used 1M vectors of 1536 dimensions (OpenAI). The index size was ~6GB.

We tested pgvector’s implementation of HNSW for you and compared it to pg_embbeding’s. This covers why this milestone is important for Postgres and why Neon is committed to supporting pgvector.

I think you can find Qdrant here https://ann-benchmarks.com/glove-100-angular_10_angular.html

This is different from the dataset we used. We used GIST-960. It could interesting to test different vector store indeed.

Adding https://ann-benchmarks.com/

@jonathan-adly do you use ivfflat with pgvector?

Hey @lysecret, what's your use-case?

Blog post author here. Let me know if you have any questions regarding the benchmark and pg_embedding

Neon separates storage and compute. The compute part is serverless and can scale up and down on demand. The storage has multiple layers: Safekeepers for durability and Pageservers to reconstruct pages from WAL.

The data is also offloaded to S3. Here is a diagram that describes the architecture: https://neon.tech/_next/image?url=https%3A%2F%2Fneon-hwp.dre...

Thanks, I didn't know about serverless-pg. I'll have a look.

Connection caching is achieved on proxy side. Edge <-> Proxy<->Compute(Postgres)

Thank you. We fixed it now.

Hey @PaulWaldman, can you tell us more about your use case with gRPC and Postgres?

> I'd love to know what fraction of the time is spent in processing, transit, querying the db, etc.

We experimented with SELECT 1, so most of the latency is transit

Yeah! these are just normal websockets. We're definitely interested in supporting and testing more protocols and ways of accessing postgres and neon as we can. We're also super excited to see the development and adoption of new protocols!

We use a proxy that redirects the request to the compute node. The proxy uses WebSockets in this case. https://neon.tech/blog/serverless-driver-for-postgres

We have only compared on Vercel Edge and Serverless environments. I think it makes a lot of sense to compare our driver to pg on CF with connect.

Blog post author here. Let me know if you have any questions!

The blog post is about how the use of Rust Persistent Data Structures (rpds) outperformed other methods in Neon's WAL indexing.

pg_tiktoken solves the problem of tokenizing text data within a Postgres database. The tiktoken_encode function allows you to tokenize text inputs and returns a tokenized output, making it easier to analyze and process text data for various applications. The tiktoken_count function enables users to return the number of tokens in a text, which is useful for checking text length limits, like those imposed by OpenAI's language models.