Course contentsShow
← System Design
Lesson 1053 of 1,91927. Circuit Breakers and BulkheadsFree lesson

Sliding Window vs Fixed Window

Comparing sliding window and fixed window approaches for tracking failure rates over time.

Sliding Window vs Fixed Window

You'll learn: How sliding and fixed time windows differ when tracking failure rates in circuit breakers — and why the choice matters for accuracy and resource usage.


When a circuit breaker monitors failures to decide whether to open, it needs to count errors over some time period. Two common approaches are fixed windows and sliding windows.

Fixed Window

A fixed window divides time into distinct buckets (e.g., one-minute intervals). All failures within 10:00-10:01 count in one bucket, then at 10:01 the count resets to zero for the new bucket.

Analogy: Imagine counting customers entering a store each hour. At the top of each hour, you reset your counter to zero.

Pros: Simple, memory-efficient (just one counter per window).
Cons: Boundary problem — if 50 failures happen at 10:00:59 and 50 at 10:01:01, the fixed window sees them as separate. The circuit breaker might miss a spike that spans the boundary.

Sliding Window

A sliding window continuously tracks the last N seconds (or N requests) as a moving window. At any moment, you're looking at "the past 60 seconds from right now."

Analogy: A rolling camera recording the last minute of activity, not just snapshots on the hour.

Pros: More accurate failure rate tracking; no boundary artifacts.
Cons: Requires storing timestamps or counters for multiple sub-buckets (e.g., 60 one-second buckets), consuming more memory and CPU.

Which to Choose?

Fixed windows work well for low-memory environments or when approximate failure rates suffice. Sliding windows provide smoother, more accurate detection but at higher computational cost — crucial when milliseconds and precision matter.

Key Takeaway: Fixed windows are simple and efficient but can miss boundary-spanning spikes; sliding windows offer continuous accuracy at the cost of added complexity and resource usage.