Pinecone is a vector database that helps users build and deploy generative AI applications at scale. Whether using its serverless architecture or a hosted model, Pinecone allows users to store, search, and retrieve the most meaningful information from their company data with each query, sending only the necessary context to Large Language Models (LLMs). By providing the ability to search and retrieve contextual data, Pinecone enables you to reduce LLM hallucinations and enhance data security.

As organizations seek to migrate and modernize from their traditional tech stacks to AI tech stacks, many are looking to implement vector databases. Applications that use generative AI, LLMs, or semantic search rely on vector embeddings to represent their data in a way that facilitates AI understanding and long-term memory retention. These vector embeddings are stored in vector databases, and enable the AI to complete its tasks effectively. By monitoring your vector database, you’ll gain the ability to optimize performance, control usage at a granular level, and quickly identify any unusual activity. That’s why we are excited to announce our expanded integration with Pinecone, enabling you to monitor the health and performance of your serverless vector databases in real time. Our previous Pinecone integration already allows you to monitor your pod-based indexes, but with the expanded capabilities, Datadog is the first and only observability platform to support the monitoring of your serverless indexes as well.

In this post, we’ll discuss how you can use the Pinecone integration to:

• Gain insight into the performance of your indexes with preconfigured dashboards

• Prevent latency with preconfigured or custom alerts

Gain insight into the performance of your indexes with preconfigured dashboards

Once you’ve created a Pinecone API key and configured the integration, Datadog will begin collecting metrics from your serverless and pod-based indexes. The updated integration includes 25 new metrics, providing comprehensive visibility into the performance of your vector databases.

Datadog provides two out-of-the-box dashboards that help you visualize these metrics, enabling you to analyze the performance of both your pod-based and serverless indexes. The Pinecone Overview dashboards display metrics related to index health and throughput, helping you observe and track usage patterns and easily identify anomalies.

As an example, let’s say you use Pinecone Serverless and want to send large volumes of data to your database using batch processing. As you track the batch process, you notice that the upload is progressing at a slow rate. After you verify that none of your objects are larger than 2MB, you further investigate in the Datadog app. Viewing the Pinecone Overview dashboard, you see that latency for upsert (import) requests is exceedingly high. To decrease latency, you send multiple read and write requests in parallel, helping to increase throughput and finish the batch upload process.

By visualizing metrics, you can identify trends, forecast resource needs, and ensure your Pinecone indexes consistently deliver low-latency, high-accuracy results for your generative AI applications.

Prevent latency with preconfigured or custom alerts

Our integration also includes two recommended monitors, preconfigured to help you detect performance issues before they become full-blown incidents. For your pod-based indexes, our recommended monitor will alert you if storage space is nearing capacity, ensuring that you can continue to store vector data efficiently. For serverless indexes, our monitor alerts you if the number of writes to an index exceeds the configured threshold, preventing the index from becoming overloaded.

For example, let’s say you receive an alert notifying you that your pod-based index is approaching fullness, putting users at risk of experiencing latency. Opening the Datadog app, you pivot from the Monitor Status page to the Pinecone Overview dashboard, seeing that fullness is hovering around 70 percent. To prevent index capacity from becoming exhausted and accommodate more vectors, you decide to vertically scale your pods, enabling you to avoid downtime while doubling your capacity instantly.

If preferred, you can customize our recommended monitors or configure completely custom alerts to monitor your Pinecone vector databases.

Start monitoring your Pinecone vector databases with Datadog today

Monitoring your Pinecone vector databases with our integration allows you to quickly identify and resolve issues, maintain optimal performance, and keep your environment healthy and secure. You can visualize and analyze key metrics with our out-of-the-box dashboards, while our recommended monitors give you the opportunity to intervene before performance issues become full-blown incidents.

You can find a comprehensive list of metrics and configuration instructions in our Pinecone documentation. To learn more about how to monitor your AI stack, read our AI integrations blog post. If you don’t already have a Datadog account, sign up for a 14-day free trial today.