Sampling

Why session-level sampling#

Per-event sampling (sending 50% of errors, 50% of vitals) is broken — you end up with errors whose breadcrumbs were dropped, replays for events that aren't there, and incoherent timelines.

Reliable uses session-level sampling instead. Each session does one dice roll on init: if it loses, the SDK goes fully dark for the rest of the session. No errors, no vitals, no requests, no replay. If it wins, every event is sent. The result: every session you do see is complete, with no missing context.

Configuring the rate#

init({
  publicKey: 'pk_live_rl_...',
  sampleRate: 25,  // 25% of sessions are recorded
});

sampleRate is a percentage from 0 to 100. The default is 100 (every session captured).

When to sample down#

• High-traffic apps — millions of sessions per day means cost. 10–25% is usually plenty for trend signal.
• Anonymous traffic — bots and scrapers don't need full coverage; sample more aggressively.
• Low-value pages — marketing sites, status pages, anything where a missed error doesn't hurt anyone. Sample to taste.

When to keep it at 100#

• Logged-in dashboards — every user matters; you want every error.
• Checkout flows / payment — a failed sample is a lost dollar. Capture every session.
• Low-traffic apps — at < 10k sessions/month, sampling is a false economy.

Dynamic sampling per environment#

const sampleRate = process.env.NODE_ENV === 'production'
  ? Number(process.env.NEXT_PUBLIC_RELIABLE_SAMPLE_RATE ?? 100)
  : 100;

init({
  publicKey: process.env.NEXT_PUBLIC_RELIABLE_KEY!,
  sampleRate,
});

User-tier-based sampling#

For SaaS apps, capture every paid user but sample free-tier traffic. The trick: roll the dice yourself based on user info, then either init or skip:

const isPaidUser = currentUser?.plan && currentUser.plan !== 'free';
const sampleRate = isPaidUser ? 100 : 10;

init({
  publicKey: 'pk_live_rl_...',
  sampleRate,
});