From Crisis to 8.5M Users: The News Platform Story Nobody Tells You

There’s a moment in every platform’s life when the dashboard starts whispering things the boardroom can’t hear yet.

For Free Malaysia Today (FMT), that whisper started in Feb–Mar 2023: traffic getting sticky in the wrong places, pages slow to breathe, a codebase that had more plot twists than a soap opera. A search for a fixer had been running for a year.

October 2023: they found me. November 2023: I joined. March 6, 2024: I shipped V3.

And then—eight months later—I tore it down and rebuilt it again.

Not because people were complaining. Because they weren’t.

Because “works” isn’t the same as “right.”

The Inheritance (V2): When “Pages Router” Means “Pages Roulette”

Let’s name it: V2. Technically “Pages Router,” practically “mystery novel.” Architecture without a map, patterns without pattern. It served content, yes—but fixing it would’ve taken longer than rebuilding.

So I did what any choosy engineer with a ticking clock does: kept the vendor on life support to keep V2 standing while I rebuilt V3 from scratch.

Why not just tidy V2? Because sometimes the fastest path forward is past the rubble, not through it.

V3 (Mar 6, 2024): The App Router Bet

I went all-in on the buzz: Next.js App Router. It promised modern patterns and pretty performance graphs. For a while, we got them.

The site picked up.
Editors were happier.
Users felt the speed.

Eight to nine months in, the whisper came back.

The Problems You Only See If You Stare at Logs

App Router’s SPA-style navigation is fantastic for apps. For news? It quietly rearranges your furniture.

Ad targeting drifted. User jumps from Business → Lifestyle. The page shell stays; the ad params don’t refresh. Wrong ads, real money.
SEO metadata went stale mid-session. Title, description, JSON-LD, keywords—not reliably refreshed on client transitions. Crawlers get yesterday’s label on today’s jar.
Analytics didn’t tell the truth. Pageviews/time-on-page depend on real navigations. SPA transitions blur counting. Editorial decisions drift.
CDN friction. The App Router + Cloudflare “Cache Everything” dance kept stepping on toes. Invalidation got fussy.

None of this sets off alarms on Day 1. On Day 240, it becomes your future.

So I did the unpopular thing: I rebuilt again—this time to what actually fits a news platform.

V4 (April 2025): Pages Router + ISR, Done Like You Mean It

Tools don’t win. Fit does.

I moved FMT to Next.js Pages Router + ISR—not because “Pages is better,” but because news needs fully refreshed pages:

True navigations → ad params reset correctly
Server-rendered metadata → search engines index what readers see
Reliable analytics → editors get reality, not vibes
Clean CDN edges → Cloudflare behaves like a performance team, not a referee

And then I solved the unglamorous parts—the ones that actually move graphs.

1) Invalidate Cache Like a Surgeon, Not a Meteor

Most teams still reach for cache.clear(). That’s not invalidation; it’s arson.

We built dependency-aware invalidation with an LRU cache—no Redis, no ceremony:

// When caching a category page, embed its dependencies
cache.set(`category/business`, {
  data,
  deps: ["post:12345", "post:12346", "post:12347"],
});

// On content change, invalidate only dependents
function onContentChange(id) {
  cache.invalidateWhere((entry) => entry.deps.includes(`post:${id}`));
}

Result: updates land fast without stampeding origin or flushing the universe.

2) ISR That Respects Reality

Static where it’s safe, fresh where it matters:

// pages/article/[slug].js
export async function getStaticProps({ params }) {
  const article = await fetchArticle(params.slug);
  return {
    props: { article },
    revalidate: 60, // refresh on demand after a minute
  };
}

Server HTML for crawlers and ad stacks
Predictable freshness for editors
No SPA illusions for analytics

3) Real-Time, But Prioritized (Because Not All Updates Are Noble)

WebSub (PubSubHubbub) pipes content events into priority queues:

const PRIORITY = {
  CRITICAL: { concurrency: 10 }, // Homepage
  HIGH: { concurrency: 5 }, // Sections
  MEDIUM: { concurrency: 20 }, // Articles
  LOW: { concurrency: 3 }, // Archives
};

The homepage shouldn’t wait politely behind a 2019 archive tweak.

4) Cloudflare: Cache HTML, Purge Precisely

We went all-edge:

Cache HTML (not just assets)
Purge by URL/tag, not “purge everything”
Let the edge do the heavy lifting; let origin breathe

This is where the cost line starts behaving.

5) Database: Boring Moves That Overperform

Prune ancient revisions (keep a few first + last; drop the noise)
Add indexes for what you actually query
Move 80% reads to replicas; keep writes single and safe

You don’t need a new data religion. You need good housekeeping.

What Changed (V2 → V4)

Let’s talk outcomes, not adjectives.

Monthly users: 3M → 8.5M (+184%)
Avg. load time: 5.2s → 1.3s (−75%)
Organic share: 40% → 68%
Recirculation: 1% → 23% (users stopped bouncing and started browsing)
Cloud spend: $45–60K → $12–15K (v4 is where the drop happens; v3 was still pricey)
Uptime across migrations: 99.9%
Content at scale: 200+ articles/day; updates in <2 minutes; viral bursts to 14k+ concurrent without drama

The headline isn’t “we used X.” It’s “we matched the job to the tool and trimmed everything that didn’t help.”

“But Wasn’t V2 Also Pages Router?”

Yes. V2 used the same tool I chose for V4. The difference is the gap between “has a piano” and “plays it.”

V2: Pages Router without architecture
V4: Pages Router with ISR, dependency-aware caching, real-time priorities, and a CDN strategy that behaves

Same label, different craft.

The Steal-This Part: What Actually Works for News Sites

If you only copy five things, copy these:

Choose the framework for the behavior you need.
- News/content: Pages Router + ISR (server HTML, true navigations)
- App-like dashboards: App Router (SPAs shine there)
Never nuke caches. Track dependencies; invalidate surgically. Your origin will thank you.
Prioritize freshness by audience impact. Homepage/sections first, then the long tail.
Cache HTML at the edge. Purge URLs/tags, not the universe. Let Cloudflare earn its keep.
Do the boring database work early. Prune revisions, add the obvious indexes, read-replicate. It’s 80/20 that feels like magic.

Two Rebuilds, Zero Drama

Could I have coasted on V3? Absolutely. Everyone was happy.

But future problems rarely arrive with a siren. They arrive as small mismatches—ad params that don’t refresh, meta tags that lie to crawlers, analytics that quietly drift. If you ship for a living, you hear them early.

So I rebuilt again—not loudly, not to make a point—just to make the platform right.

The Minimal Code That Paid Rent

A few tiny patterns did a lot of work.

Dependency-aware invalidation

function dependsOn(entry, id) {
  return entry.deps?.includes(id);
}

export function invalidateBy(ids = []) {
  for (const key of cache.keys()) {
    const entry = cache.get(key);
    if (ids.some((id) => dependsOn(entry, id))) cache.delete(key);
  }
}

ISR with clear expectations

export async function getStaticProps({ params }) {
  const article = await fetchArticle(params.slug);
  return { props: { article }, revalidate: 60 };
}

Priority queues (shape, not ceremony)

queue.add({ kind: "homepage", id }, { priority: "CRITICAL" });
queue.add({ kind: "section", id }, { priority: "HIGH" });
queue.add({ kind: "post", id }, { priority: "MEDIUM" });

Small pieces. Big leverage.

What This Proves (Quietly)

Fit beats fashion. App Router is brilliant—for apps. We’re a news site.
Operational taste matters. Knowing when “fast enough” isn’t honest.
Craft compounds. CDN, ISR, caching, indexes—each is modest; together they’re a new curve.

If you want fanfare, ship a redesign. If you want results, ship foundations.

Footnotes for the Curious

Pages Router deployment: April 2025 (V4)
App Router run: ~8–9 months (V3)
Ad/SEO/analytics drift: all traced to SPA-style transitions
Cache: custom LRU (no Redis)
Edge: Cloudflare “Cache Everything,” purge by tag/URL
Content scale: hundreds of thousands of items; 100K+ migrated through the new path with zero data loss

If you’re wrestling with a high-traffic news or content platform and the graphs won’t listen, I’m happy to compare notes on what’s signal and what’s noise. The playbook above is open book—and battle-tested.

View My Other Work · Let’s Talk

All metrics and statistics in this article are from real production systems serving 8.5M+ monthly users. Because if you can't measure it, you can't improve it.