Course contentsShow
← Software Engineering
Lesson 1110 of 1,93333. Error Handling and ResilienceFree lesson

Partial Response Strategies

Returning incomplete but useful results when some data sources fail, with appropriate status indicators.

Partial Response Strategies

What you'll learn: How to return incomplete but still useful data when some parts of your system fail, keeping users informed and productive.

The Core Idea

When building systems that aggregate data from multiple sources, you'll often face situations where some dependencies fail while others succeed. Instead of failing the entire request (all-or-nothing), partial response strategies let you return whatever data you did successfully retrieve, along with clear indicators about what's missing and why.

Think of it like ordering a combo meal at a restaurant. If they're out of fries, would you rather they refuse to serve you anything, or bring your burger and drink while explaining the fries are unavailable? Most customers prefer the partial service.

How It Works

When handling a request that depends on multiple data sources:

  1. Attempt all operations (potentially in parallel)
  2. Collect both successes and failures
  3. Return available data with metadata indicating:
    • Which parts succeeded
    • Which parts failed and why
    • Severity of missing data (critical vs optional)

For example, a user profile page might successfully load basic info and posts, but fail to load recommendations. You return the profile data with a status field indicating recommendations: unavailable.

Benefits

  • Better user experience: Users get something rather than nothing
  • Faster perceived performance: Don't wait for slow/failing services
  • Clearer debugging: Status indicators help identify which dependency failed
  • Graceful degradation: System remains partially functional during outages

Key Takeaway: Partial responses turn total failures into degraded experiences—return what you can, clearly mark what you couldn't, and let clients decide if incomplete data is good enough.