Skip to content
Tutorials Center
My learning

Topics

Lesson 32 of 37 ยท Measuring and Monitoring

Synthetic Monitoring and Alerts

Combine scheduled synthetic tests with RUM to catch performance regressions early, set sensible alert thresholds, avoid noisy alarms and tie changes to releases.

A performance fix that nobody watches will quietly rot. A new campaign script, a heavier hero image or a dependency update can undo months of work in a single deploy. Monitoring means measuring continuously and being told when something goes wrong, instead of finding out from a customer or a drop in search rankings.

Two kinds of monitoring complement each other: synthetic (scheduled lab tests) and real-user (RUM, from the previous lesson).

Synthetic monitoring

A synthetic monitor loads your pages on a schedule (for example every 15 minutes) from fixed locations with a fixed device and network profile, using Lighthouse, WebPageTest or a commercial service.

Strengths:

  • Very stable, so small regressions are visible.
  • Works before launch, on staging, and for pages with no traffic yet.
  • Detailed waterfalls and filmstrips for investigation.
  • Can test competitors.
  • Catches outages even at 3 a.m. when nobody is visiting.

Weaknesses:

  • It tests only the pages and flows you scripted.
  • One device and one network are not your whole audience.
  • It cannot see differences in real devices, ads, personalization or logged-in states.

Real-user monitoring

RUM covers everyone, shows the true distribution and exposes segments you never thought to test, but it is noisier and needs traffic. Use both: RUM tells you whether users suffer and where; synthetic tells you why and detects regressions fast.

Synthetic RUM
Data source Scheduled test runs Real visits
Stability High Lower (varied devices)
Coverage Scripted pages only Every page and user
Best at Alerts, debugging, pre-launch Truth, segments, business impact

Choosing what to monitor

  • Key templates: home, category, product or article, search, checkout.
  • Key journeys: "search, open a product, add to cart" using scripted steps.
  • Key metrics: LCP, CLS, INP (RUM) or TBT (synthetic), TTFB, total JavaScript size, number of requests.
  • Segments: mobile and desktop, top countries.

Setting alert thresholds

Alerts that fire constantly are ignored. Good alerts are tied to your budget and robust to noise:

  • Alert when the median of the last N runs crosses the threshold, not a single run.
  • Alert on a relative change, such as "p75 LCP is 20 percent worse than the 7-day baseline".
  • Require the condition to hold for several checks in a row.
  • Separate warning (post in a chat channel) from critical (page someone).
JS A noise-tolerant alert rule

Press Run (or Ctrl + Enter) to see the console output here.

No output.

The single spike does not alert, because the median ignores it, but a sustained change does.

Tie performance to releases

When an alert fires, the first question is "what changed?". Make that easy:

  1. Include the release version or commit hash in RUM beacons and synthetic reports.
  2. Add deploy markers to your charts.
  3. Keep a changelog of third-party tag changes; marketing tags are a frequent culprit.
  4. Run the performance check in the pull request, so most regressions never reach production (see performance in CI).
flowchart LR
  P["Pull request<br/>lab check"] --> S["Staging<br/>synthetic"]
  S --> R["Production<br/>RUM + synthetic"]
  R -->|"alert"| I["Investigate<br/>trace + deploy marker"]

Each stage catches a different class of problem, earlier is cheaper.

What to do when an alert fires

  1. Confirm with a second source (RUM and synthetic).
  2. Check the deploy markers and recent third-party changes.
  3. Reproduce in a lab test with the same throttling.
  4. Roll back if the cause is a recent change and the impact is high; fix forward otherwise.
  5. Write a short note: cause, fix, and what check would have caught it earlier.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. What is the main strength of synthetic monitoring?

    Controlled conditions make changes stand out, even without traffic.
  2. 2. Why should an alert use the median of several runs?

    One outlier should not wake someone up; a sustained shift should.
  3. 3. What does RUM provide that synthetic monitoring cannot?

    RUM measures actual visitors across all pages and conditions.
  4. 4. Why include a release version in performance reports?

    Matching a change in metrics to a release speeds up diagnosis.

Summary

  • Combine synthetic tests (stable, early warning) with RUM (real-user truth).
  • Monitor key templates and journeys, with segments by device and country.
  • Make alerts robust: medians, relative baselines, multiple consecutive checks.
  • Tie metrics to releases and test in pull requests to catch regressions early.

Next: Performance Budgets