Observability for Backend Performances
Observability, in a performance context, is the ability to see where a backend spends its time by using metrics, logs, distributed traces, profilers and load tests. It comes first in the roadmap because every later fix depends on evidence about which component is actually slow.
Stage: Measure First · Part of the Backend Performance roadmap · Last reviewed 19 November 2026
What to learn in Observability
- Capture a CPU flame graph from a running service and name the hottest function
- Read a distributed trace and identify the slowest span in a request
- Report p50, p95 and p99 latency and explain why averages mislead
- Write a load test script that ramps virtual users against an endpoint
- Separate a latency problem from a throughput problem using the right metric
Observability interview questions
Why do engineers track p99 latency instead of the average?
Averages hide the slow tail. If one request in a hundred takes several seconds, the mean still looks healthy, yet those users have a bad experience, and in a fan-out system the tail dominates overall response time. Percentiles expose that tail directly.
What does a flame graph show you?
A flame graph stacks sampled call stacks, with width showing the share of CPU time spent in each function. Wide boxes near the top are hot spots worth optimising. It tells you where time goes, not why, so you still read the code.
Where Observability fits in the Backend Performance roadmap
Observability belongs to the Measure First stage (intermediate level), which takes about 14 hours. Find the bottleneck.
Other units in this stage
Back to the full Backend Performance roadmap