How to Find Slow Endpoints Using HAR Files

A practical guide for frontend developers · 2025

The Problem

Your app feels slow but you don't know where. Users complain about load times, Google Pagespeed scores are mediocre, and you've stared at the waterfall panel for hours. HAR files contain everything your browser downloaded — request/response timing, sizes, caching headers, errors.

Step 1: Export a HAR File

Open DevTools (F12) → Network tab → Check "Preserve log" → Refresh → Right-click → "Save all as HAR with requests".

⚠️ Make sure the network request count isn't filtered — sort by Time column first so you capture the slowest requests too.

Step 2: Look at the Waterfall

The waterfall view shows every resource stacked vertically. Look for horizontal gaps between when a request starts and ends. Long bars = slow responses.

Type	What to look for	Typical fix
Server wait time	TTFB > 500ms	Optimize backend queries, add caching
Download speed	Slow transfer for large resources	Compress, lazy-load, use CDN
Queue/delay	Many requests waiting sequentially	Prioritize critical path, parallelize

Step 3: Focus on What Matters

Not all endpoints matter equally. A third-party analytics script taking 800ms may not hurt UX. But an API call that blocks rendering?

Categorize by:

CSS/JS bundles — block rendering, should be small & cached
Images — large downloads, usually non-critical path
API/XHR calls — directly impact perceived performance
Third-party scripts — ads, analytics, chat widgets
Fonts — FOUT/FOIP causes layout shifts

Step 4: Extract the Numbers You Need

import json

with open('export.har') as f:
    har = json.load(f)

for entry in har['log']['entries']:
    print(f"{entry['request']['url']}:")
    print(f"  TTFB: {entry['time']*1000:.0f}ms")

Why Structure Your Analysis?

HAR files vary wildly in size. Clean exports have ~20 entries. Production exports can have hundreds. Manually reviewing is impractical.

This is why I built HAR Timing Report — a tool that parses HAR files entirely in your browser, extracts only allowlisted timing metrics, and outputs clean CSV reports grouped by category.

Try HAR Timing Report →