AI Search: Speed vs. Relevance Trade-off
AI Search balances retrieval speed and result quality by choosing between dense embeddings, sparse lexical search, and hybrid approaches. Understand which trade-off fits your workload.
Loading…
Topic
Vector
2 posts about Vector. Every summary links to the original.
- 2 posts
- 1 company
- 1 explainer
See the idea
Vector explainers
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Newest first
Posts
Shipping Logs
Shipping Logs describes how Fly.io application stdout moves through the platform and how to centralize and retain it beyond Fly.io’s log lifetime. Apps run in Firecracker VMs, where an injected Rust init process captures stdout and sends it through host-side Go code to Vector, which publishes the logs to an internal NATS cluster; Flaps limits access to an organization’s logs. The Fly Log Shipper runs Vector as a NATS client and forwards organization-wide streams to a configured sink such as Logtail, Loki, Datadog, or Cloudwatch, using secrets for provider credentials. Deployment involves launching the image, removing the generated services section, adding an internal port, and running fly deploy. A SUBJECT environment variable can narrow collection with NATS subjects such as logs.sandwich.>, covering an app across regions and instances.
Fly Behind The Scenes: Fresh Logging
Fly replaced its centralized Graylog logging platform after seven servers could not process peak volumes of 20,000–30,000 logs per second, causing drops. Vector, written in Rust, now runs on every server, parses and transforms logs from journald and other applications, and sends them directly to Elasticsearch, removing Graylog as a middleman. The pipeline adopts the Elasticsearch Common Schema (ECS), with Fly-specific fields for app names, allocation IDs, and regions, while Vector configurations are distributed through the existing configuration management system. The migration took about two weeks alongside other work; logs are now available quickly enough to support failed-allocation queries that previously often raced processing, and future sinks may include Papertrail, Honeycomb, or Kafka.
Related topics
All topics →Weekly digest
The best engineering reads, once a week.