Recode Hive
How do you decide whether a performance issue is worth optimizing? #994
-
|
When profiling a system, it’s common to find multiple performance bottlenecks. In practice, how do you decide which performance issues are actually worth fixing versus those that should be documented and left alone? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
I usually decide based on impact, not technical curiosity. The main criteria I look at: User-visible impact: latency, error rates, or throughput affecting real users. Frequency: a slow path hit millions of times is often more important than a very slow edge case. Cost amplification: performance issues that directly increase infrastructure or operational costs. If an optimization doesn’t materially improve any of those dimensions, I document it instead of fixing it. Over time, this approach has helped avoid complex optimizations that add maintenance burden without delivering meaningful value. |
Beta Was this translation helpful? Give feedback.
I usually decide based on impact, not technical curiosity.
The main criteria I look at:
User-visible impact: latency, error rates, or throughput affecting real users.
Frequency: a slow path hit millions of times is often more important than a very slow edge case.
Cost amplification: performance issues that directly increase infrastructure or operational costs.
If an optimization doesn’t materially improve any of those dimensions, I document it instead of fixing it.
Over time, this approach has helped avoid complex optimizations that add maintenance burden without delivering meaningful value.