Why Search Breaks in Production

This article explores the factors that can cause search systems to behave differently in production compared to their development environment, even when they have similar features.

đź’ˇ

Why it matters

Understanding the complex factors that can cause search systems to break in production is crucial for building reliable and scalable search experiences.

Key Points

  • 1Search systems are more than just retrieval - they need to work with filters, counts, sorting, paging, reranking, freshness, access control, and business rules
  • 2Query shape, document shape, and retrieval scope can significantly impact search system behavior
  • 3Execution shape, including approximation, candidate generation, and filter timing, affects relevance, latency, and system alignment
  • 4Hardware, update, and traffic pressures can expose hidden costs and instability in search systems
  • 5The product contract, such as low latency and reliable counts, defines what
  • 6 means for a search system

Details

The article discusses how seemingly similar search systems can behave very differently in production. It explains that the key differences come from the interaction of six factors: query shape, document shape, retrieval scope, execution shape, operating pressure, and product contract. Query shape refers to the complexity and purpose of user queries, which can range from simple lookups to complex constraint-based retrieval. Document shape, such as short product records versus long contracts, affects how content is structured and matched against queries. Retrieval scope, or the live slice of data being searched, also plays a major role. Execution shape covers lower-level choices like approximation, candidate generation, and filter timing, which impact relevance, latency, and system alignment. Hardware, update, and traffic pressures can expose hidden costs and instability in search systems. Finally, the product contract, including requirements like low latency and reliable counts, defines what

Like
Save
Read original
Cached
Comments
?

No comments yet

Be the first to comment

AI Curator - Daily AI News Curation

AI Curator

Your AI news assistant

Ask me anything about AI

I can help you understand AI news, trends, and technologies