java 50 lines · 9 steps

Computing latency percentiles in Java

An immutable stats class sorts latency samples once, then answers percentile queries with linear interpolation.

Explained by highlit
Intermediate percentiles interpolation immutability streams statistics
1import java.util.Arrays;
2import java.util.List;
3 
4public final class LatencyStats {
5 
6 private final long[] sorted;
7 
8 public LatencyStats(List<Long> responseTimesMicros) {
9 this.sorted = responseTimesMicros.stream()
10 .mapToLong(Long::longValue)
11 .sorted()
12 .toArray();
13 }
14 
15 public long percentile(double percentile) {
16 if (percentile < 0.0 || percentile > 100.0) {
17 throw new IllegalArgumentException("percentile must be in [0, 100]");
18 }
19 if (sorted.length == 0) {
20 throw new IllegalStateException("no samples recorded");
21 }
22 if (sorted.length == 1) {
23 return sorted[0];
24 }
25 
26 double rank = percentile / 100.0 * (sorted.length - 1);
27 int lower = (int) Math.floor(rank);
28 int upper = (int) Math.ceil(rank);
29 
30 if (lower == upper) {
31 return sorted[lower];
32 }
33 
34 double weight = rank - lower;
35 return Math.round(sorted[lower] + weight * (sorted[upper] - sorted[lower]));
36 }
37 
38 public long p50() { return percentile(50.0); }
39 
40 public long p95() { return percentile(95.0); }
41 
42 public long p99() { return percentile(99.0); }
43 
44 public long max() { return sorted[sorted.length - 1]; }
45 
46 @Override
47 public String toString() {
48 return "LatencyStats" + Arrays.toString(sorted);
49 }
50}
01 / 01
STEP 01

Walkthrough

Space play ←→ step click any line
Three takeaways
  1. 1Sorting the samples once in the constructor makes every later percentile query cheap.
  2. 2Linear interpolation between neighbouring ranks gives smoother estimates than picking a single element.
  3. 3Validating inputs and edge cases up front keeps the core math simple and safe.

Related explainers

java 
@Component
@Converter
public class EncryptedStringConverter implements AttributeConverter<String, String> {
 

Transparent column encryption in Spring & JPA

encryption aes-gcm jpa-converter
Advanced 10 steps
rust 
use axum::{
    extract::{Path, State},
    response::sse::{Event, KeepAlive, Sse},
};

Streaming import progress with SSE in Axum

server-sent-events streams watch-channel
Advanced 7 steps
java 
package com.acme.billing.config;
 
import org.springframework.boot.autoconfigure.condition.ConditionalOnProperty;
import org.springframework.boot.context.properties.ConfigurationProperties;

Feature-flagged beans with Spring @ConditionalOnProperty

feature-flags conditional-beans strategy-pattern
Intermediate 5 steps
java 
public static Map<String, String> parseCookieHeader(String header) {
    Map<String, String> cookies = new LinkedHashMap<>();
    if (header == null || header.isBlank()) {
        return cookies;

Parsing an HTTP Cookie header in Java

string-parsing http url-decoding
Intermediate 6 steps
go 
package logging
 
import (
	"context"

Deduplicating log attributes in Go's slog

decorator-pattern structured-logging immutability
Intermediate 8 steps
java 
public class TimedSocketReader {
 
    private static final int READ_TIMEOUT_MS = 5_000;
    private static final int CONNECT_TIMEOUT_MS = 3_000;

Reading a socket with connect and read timeouts

sockets timeouts io
Intermediate 8 steps

Share this explainer

Here's the card — post it anywhere.

Computing latency percentiles in Java — share card
Made with highlit — turn any snippet into a walkthrough like this in about a minute.
Explain your code