Raspberry Pi

Hardcharger™ runs on a Raspberry Pi 4 or 5 with a 64-bit operating system: the release has a linux-arm64 build and the image has a linux/arm64 variant. A Pi with 8 GB or 16 GB of memory and an SSD makes a capable little analytics server.

Check your Pi

$ uname -m
…
$ getconf GNU_LIBC_VERSION
glibc …

You see	Then
`aarch64` and glibc 2.28 or newer (Raspberry Pi OS “Bookworm” or “Trixie”, Ubuntu 20.04 or newer)	Install natively, below.
`aarch64` and an older glibc	Use Docker, below, or upgrade the OS.
`armv7l` or `armv6l`	That’s a 32-bit OS. Install the 64-bit Raspberry Pi OS: Hardcharger has no 32-bit build.

The installer checks all of this before it downloads anything, and says what to do.

curl https://install.hardcharger.gizmodata.com | sh
HARDCHARGER_ADMIN_PASSWORD='change-me' ~/.hardcharger/bin/hardcharger-server --config pi.yaml

docker run -d --name hardcharger --restart unless-stopped -p 31337:31337 -p 31338:31338 \
  -e HARDCHARGER_ADMIN_PASSWORD='change-me' -v /mnt/ssd/hardcharger:/var/lib/hardcharger \
  ghcr.io/gizmodata/hardcharger:latest --database /var/lib/hardcharger/analytics.duckdb \
  --memory-limit 5GB

Settings for a Pi

# pi.yaml: an 8 GB Pi with its database on an SSD
database: analytics.duckdb        # on the SSD: an SD card is slow and wears out
memory_limit: 5GB                 # leave the OS and page cache some room
session_idle_timeout: 600
max_concurrent_statements: 2      # four cores: queue rather than oversubscribe
init_sql_commands: "SET threads = 4; SET temp_directory = 'tmp';"

Storage. Put the database and DuckDB’s spill directory (temp_directory) on a USB 3 or NVMe SSD.
Memory. memory_limit around 60% of RAM leaves room for the OS. DuckDB spills to disk beyond it rather than failing.
Heat. Sustained queries throttle a Pi without cooling: use a case with a fan or a heatsink.
Network. Bind to your LAN, turn on TLS before anyone else connects, and keep the health port inside your network.