Deploying Tesseracts

Tesseracts built via tesseract build are standard Docker images, so they can be shared, pushed, pulled, and deployed like any other container.

Using Docker tools

Built Tesseracts are Docker images named after the Tesseract. Standard Docker commands work directly:

# Invoke a Tesseract via `docker run`
$ docker run vectoradd apply --help
Usage: tesseract-runtime apply [OPTIONS] JSON_PAYLOAD

  Apply the Tesseract to the input data.

  Multiplies a vector `a` by `s`, and sums the result to `b`.
  ...

# Push a Tesseract to a container registry
$ docker push vectoradd
...

# Pull a Tesseract from a container registry
$ docker pull mytesseract
...

# Save a pre-built Tesseract image to a tar file
$ docker image save vectoradd -o vectoradd.tar
...

# Spawn a Tesseract server
$ docker run vectoradd serve
...

This gives you fine-grained control over Tesseract images and lets you use any container-aware tooling.

Tip

Since podman has a docker-compatible CLI, all commands above work with podman as well.

Example: Deploying on Azure Virtual Machines

Note

This example assumes you have an Azure account and are familiar with cloud infrastructure. Using Azure VMs is just one of many deployment options. Cloud resources may incur costs.

The general process:

  1. Push the Tesseract image to Azure Container Registry.

  2. Create a virtual machine.

  3. Install Docker on the VM.

  4. (Optional) Install NVIDIA drivers and the CUDA toolkit.

  5. Pull the Tesseract image on the VM.

  6. Start the container via docker run serve, listening on port 8000.

A reference script is available here: create-vm-azure.sh.

Warning

This script contains placeholders and will not work out of the box. It assumes you have a resource group, VNet, Subnet, and Azure Container Registry already configured.

Populate the variables at the top of the script and run it:

$ bash create-vm-azure.sh vectoradd:latest

The script assumes Docker is installed locally and authenticated to your Azure Container Registry:

$ az login
$ az acr login --name <registry-name>

Deploying without a container engine

If no container engine is available, you can deploy using tesseract-runtime serve. This provides the same functionality as tesseract serve and docker run myimage serve, and can be queried the same way (e.g., with curl or via Tesseract.from_url in the Python SDK).

This approach requires a Python environment with the dependencies from tesseract_requirements.txt installed, as documented in Running Tesseracts without containers. For a complete HPC example, see the HPC deployment tutorial.

Example:

$ pip install -r tesseract_requirements.txt
$ TESSERACT_API_PATH=/path/to/tesseract_api.py tesseract-runtime serve