Heroku does not check your app from outside, and it does not tell you when the app is down. Its alerting is limited to two metrics, the 95th percentile response time and the share of requests that return a 5xx, and it exists only on Standard-1X and higher dynos. On Eco and Basic you get nothing, and on every tier a request that never reaches the router never shows up. A production Heroku app needs an external monitor that requests your own domain every 30 seconds and pages a person.
The figures below come from Heroku's DevCenter pages on application metrics, Eco dyno hours, dyno types and error codes, read in November 2026. Where a page does not say something, this article does not claim it.
What Heroku alerts on, by dyno type
Heroku's application metrics documentation is explicit about who gets what. Apps on Eco dynos get no application metrics at all. Basic dynos get metrics, but only 24 hours of data at 10 minute resolution. Threshold alerting is "available to apps running on Professional dynos (standard-1X, standard-2X and performance) and all Fir dynos", and the alertable metrics are the web dyno 95th percentile response time and the percentage of failed requests, meaning 5xx responses. Notifications go to email, PagerDuty and the dashboard.
| Dyno type | List price | Application metrics | Threshold alerts | Sleeps when idle |
|---|---|---|---|---|
| Eco | $5 a month for a shared pool of 1,000 hours | None | No | Yes, after 30 minutes without web traffic |
| Basic | $7 a month, billed at about $0.01 an hour | 24 hours at 10 minute resolution | No | Not stated in the table |
| Standard-1X | $25 a month, about $0.03 an hour | Yes | Response time and failed requests | Not stated in the table |
None of those alert types is "the app does not answer". The router can only measure requests it handled. A DNS change, a lapsed certificate on your custom domain or a Heroku routing problem upstream of your dyno produces no 5xx to count. That is an inference from how the metrics are defined, and it is the reason an outside check exists.
The Eco dyno trap when you add a pinger
Heroku's Eco dyno hours page says an Eco web dyno with no traffic for 30 minutes sleeps, and then wakes after a short delay when a request arrives. It also says that monitoring services such as Pingdom or New Relic monitoring can prevent a web dyno from sleeping. That sounds like a feature until you do the arithmetic.
A 30 day month has 720 hours and a 31 day month has 744. One Eco web dyno that never sleeps because a monitor pings it burns 744 of the 1,000 pool hours. Two such apps need 1,488 hours. When the pool runs out, Heroku forces all Eco dynos on the account to sleep for the rest of the month, so the monitor that was meant to protect two apps ends up taking both offline. Eco is fine for a side project where sleep is acceptable. It is the wrong place for anything a customer pays to use, and uptime monitoring does not fix that.
The failures an outside check sees on Heroku
Heroku's error code reference lists what the router does when your app misbehaves, and every one of its HTTP error pages is a 503. H10 means a crashed web dyno or a boot timeout. H12 is a request that took longer than 30 seconds. H13 is a connection closed without a response. H14 means no web dynos are running, most often after scaling to zero. A monitor that asserts on the status code and a string in the body sees all of them within one interval. It also sees failures the router never records:
- The custom domain stops resolving or the certificate expires.
- The app returns 200 with a framework error page or an empty body.
- A release ships a broken login route while the rest of the site is healthy.
- A third party API your page depends on times out and the page hangs just under the 30 second router limit.
The same logic applies on other hosts. We wrote up the pattern for Fly.io uptime monitoring and for Render, where platform health checks also decide routing rather than notify you. On DigitalOcean the gap is the alert path, see DigitalOcean uptime monitoring.
A worked example: a Standard-1X app with paying customers
Take a SaaS app on one Standard-1X web dyno at $25 a month plus a worker. Heroku alerts are configured at a 5% failed request threshold. At 02:10 a deploy goes out with a bad environment variable and the web dyno crashes on boot. The router serves H10 503s. Those count as failed requests, so the alert fires, but only once there are enough requests in the window to cross the percentage, and it goes to email. At 02:40 the custom domain certificate renewal fails. Browsers show a warning, nobody gets a 5xx, and Heroku's numbers look normal.
An outside check against your domain every 30 seconds catches the first outage at the next interval and the second 30 seconds after the certificate lapses, or days earlier through an expiry warning. Both go to SMS and Slack to whoever is on call, and the failure is confirmed from a second region before it pages, so one dropped route does not wake anyone. If you also want a place to tell customers, pair it with a status page.
What to buy for a production Heroku app
If customers pay for the app, run it on at least a Standard-1X dyno so it does not sleep and so you get threshold alerting, keep those alerts on, and add an external monitor on top. Heroku's alerts tell you the app is slow or erroring while requests flow. The outside check tells you customers cannot reach it at all. AlertPing is flat from $49 a month with 30-second checks from $59, and SMS, email, Slack and webhook alerts are included, so a handful of Heroku apps cost the same as one. The uptime monitoring pricing comparison lays out what per-monitor tools cost by comparison. If you run a hobby app on Eco, you do not need to buy anything, and a plan this size would be the wrong spend.
Other hosted monitors work in the same outside-in way, Uptimehub among them, so judge any of them on interval, region confirmation, assertions and how alerts reach you at 3 a.m. Our guide to getting alerts when your website goes down covers the setup, and the website down alerts page explains how escalation works.
Heroku uptime monitoring questions
Does Heroku have built-in uptime monitoring?
No. Heroku's documentation describes metrics and, on Standard-1X and higher, threshold alerts on response time and failed requests. It does not describe a monitor that requests your app from outside on a schedule and notifies you when it does not answer.
Will an uptime monitor keep my Eco dyno awake?
Yes. Heroku says monitoring services can prevent a web dyno from sleeping. That uses your shared pool of 1,000 hours, about 744 hours in a 31 day month for one dyno, so it works for one app and exhausts the pool with two.
What do H10, H12 and H14 mean on Heroku?
H10 is a crashed web dyno or boot timeout. H12 is a request that ran longer than 30 seconds. H14 means no web dynos are running. All three return a 503 to the visitor, which an external monitor flags as down.