systemisystemi
Sign in
Tools

App memory sizer

Will your app fit in a 128, 256 or 512 MB container, and what does its database pool cost?

Your app

Pick the runtime and how your app connects to the database.

Measured with: FastAPI + uvicorn.
Each worker is a separate Python process with its own copy of your code and its own DB pool.
On top of a hello-world app. Read it from docker stats of your own app, minus the base above.
For request spikes, caches and garbage collection. Your choice.

App container

2 workers: 87.0 MiB measured idle + 80.0 MiB your code + 25% margin = 209 MiB

128 MB
163% · does not fit
256 MB
82% · tight
512 MB
41% · fits

Database connections

10 connections × 3.36 MiB = 33.6 MiB on the PostgreSQL server. Running PostgreSQL 18 yourself in a container: 53.5 MiB base + connections = 87.1 MiB.

256 MB
34% · fits
512 MB
17% · fits
1 GB
9% · fits

Connections use memory on the database server, not in your app container. Sorts and hash joins add up to work_mem (4 MB by default) per operation while a query runs.

Where the numbers come from

The base for each runtime is a hello-world JSON app measured with docker stats after 50 requests, with no limit and under 256 MB and 128 MB limits: idle memory for PHP, Python, Java, Spring Boot and .NET, measured. Python workers use the measured 1, 2 and 4 uvicorn workers (31.6, 87.0 and 147.6 MiB); other worker counts add 30.3 MiB each, the measured step. The 512 MB bar uses the no-limit numbers, the largest we saw.

Connection memory is PostgreSQL 18 with 50 connections: 1.66 MiB each running SELECT 1, 3.36 MiB after reading 100 tables and 3.80 MiB with 50 prepared statements, on top of 53.5 MiB for the server itself: how much memory a PostgreSQL connection uses, measured.

How to read it

  • Fits means under 80% of the limit, tight 80 to 100%, does not fit over the limit, where the container is killed.
  • A hello-world app is the floor. A real app with an ORM, templates and a few libraries uses more: put that in “your code and libraries”.
  • More workers or threads serve more requests at once, but each Python worker is a full copy of your app and opens its own pool.
  • On systemi, PostgreSQL runs outside your app container, so the connection memory is not counted in your app's memory.

Ready to deploy? See the getting started guide and the databases guide.