Skip to content
Tutorials Center
My learning

Topics

Lesson 6 of 37 ยท Apps, Dynos and Buildpacks

Dynos and Dyno Types

Understand Heroku dynos: what they are, process types, the Eco, Basic, Standard and Performance families, sleeping, memory limits and how dyno hours are counted.

A dyno is the unit of compute on Heroku: a lightweight, isolated Linux container that runs exactly one process of your app from a release's slug. Everything your app does happens inside dynos. When people say "scale the app," they mean "run more dynos" or "run bigger dynos."

This lesson explains what dynos are made of, the families you can choose from, and the rules (sleeping, memory, hours) that influence cost and behavior. Exact prices, memory numbers and hour allowances change over time, so use the figures here to understand the shape of the choices and confirm them on the Dev Center and pricing pages before you commit.

What a dyno is and is not

  • It runs one command defined by a process type in your Procfile (see Procfile and process types).
  • It starts from your release's slug, with the config vars of that release in its environment.
  • It has its own CPU share, memory and a temporary filesystem.
  • It is not a virtual machine you log in to and customize. If you need to look around, heroku run bash starts a separate one-off dyno.
  • Several dynos of the same type share nothing with each other except the services you attach (database, cache). That is why a stateless design is essential (see dyno lifecycle).

The set of process types and how many dynos each one has is called the app's formation. The default formation after a first deploy is one web dyno.

Dyno families

Think of dynos in tiers. Names and limits below describe the general design; check the current table for exact values.

Choosing a dyno family

Eco Cheapest. Sleeps when idle.
Basic Always on. Small, one per process type.
Standard Production. Scale out, metrics.
Performance Dedicated resources, big memory.

Check the current pricing page for exact limits.

1 / 2

In more detail:

  • Eco dynos draw from a shared monthly pool of dyno hours for the whole account, and a web dyno sleeps after about 30 minutes without incoming traffic. The next request wakes it, which adds a delay of several seconds. Eco is meant for learning, prototypes and side projects.
  • Basic dynos are always on and billed by usage up to a monthly cap. They are small and limited to a single dyno per process type, so they are a step up for light production use.
  • Standard dynos (Standard-1X and Standard-2X) are the normal choice for real apps. You can run many of them, they support horizontal scaling and metrics, and they run in shared infrastructure.
  • Performance dynos (Performance-M, Performance-L and larger) get dedicated compute, much more memory and more predictable performance. They suit heavy workloads and apps where latency consistency matters.
  • Private and Shield spaces put dynos in isolated networks, adding network control and, for Shield, compliance-oriented features. They cost more and matter mostly to larger organizations.

The type is a property of a process type, so your web dynos can be Standard-2X while your worker dynos are Standard-1X. You pick it with a scale command: heroku ps:type web=standard-2x changes the size, and heroku ps:scale web=3 changes the count.

Memory limits

Each dyno size has a memory ceiling, from a few hundred megabytes on small dynos up to many gigabytes on large Performance dynos. Two thresholds matter:

  • Above the limit the platform logs an R14 Memory quota exceeded warning and begins swapping to disk, which is very slow.
  • Well above it (roughly twice the limit) the platform logs R15 Memory quota vastly exceeded and kills the dyno.

A growing memory line in the metrics usually means a leak or too many worker threads in a process. Fix the cause before you size up, because a bigger dyno only delays the crash.

Sleeping and idle behavior

Only web dynos on Eco sleep. A sleeping dyno still counts as part of your formation but uses no hours. The first request after sleep waits while the dyno boots, so the app feels slow once per idle period. Dynos on paid families never sleep, which is one of the main reasons teams leave Eco for production.

Dyno hours

Dyno usage is metered per second. For a set of dynos running for a month, the arithmetic is simple: dynos x hours running. A single always-on dyno uses about 720 hours in a 30-day month; two use about 1,440. If an Eco pool is, say, 1,000 hours, one always-on dyno is already beyond it, which is why sleeping exists. The stand-in below models that arithmetic. The pool size is a practice value, not a promise about the current plans.

Python Stand-in: dyno hours against a shared pool

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

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The last scenario is the classic surprise: hours are counted across all your apps on the account, so several small apps add up.

Stand-in: simulate a sleeping dyno

This model shows why the first request after an idle period is slow on a sleeping plan. Times are minutes since the program started, and the 30-minute idle rule mirrors the real behavior of Eco web dynos.

Python Stand-in: idle sleep and wake-up delay

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

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

You can avoid the wake-up delay by using an always-on dyno type. Pinging an Eco dyno from outside to keep it awake works against the point of the plan and burns your hours.

Choosing sizes in practice

  1. Start with the smallest type that is appropriate (Eco for learning, Standard-1X for a first real deployment).
  2. Watch memory and response times in the metrics tab.
  3. Scale out (more dynos) for request throughput and up (bigger dyno) for memory-hungry work. See the scaling chapter of this course.
  4. Re-check costs after every change; billing follows the formation, not your intentions.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. How many processes does one dyno run?

    A dyno runs one command from your Procfile. To run more processes, you run more dynos.
  2. 2. What happens to an Eco web dyno after about 30 minutes without traffic?

    Eco web dynos sleep when idle. The first request afterward waits while the dyno boots.
  3. 3. Which situation usually means you should fix the code rather than buy a bigger dyno?

    Steadily growing memory is typically a leak or an unbounded cache. A larger dyno only delays the crash.
  4. 4. What does the R15 error indicate?

    R14 is a warning for exceeding the quota, while R15 means it is far over and the platform terminates the dyno.

Summary

  • A dyno is a small, isolated container running one process; the set of dyno types and counts is the formation.
  • Families run from Eco (cheap, sleeps) through Basic and Standard to Performance (dedicated, large).
  • Memory over the limit causes R14 warnings; far over causes R15 and a kill.
  • Dyno time is metered per second and pooled across the account on Eco.

Next: Procfile and process types