Add "pgx_inference" to the "shared_preload_libraries" parameter in the spec.fepChildCrVal.customPgParams section of the FEPCluster custom resource. This will launch the load launcher.

Additionally, add the number of databases for which this feature is enabled plus one to the "max_worker_processes" parameter. This determines the number of load launchers started by this feature.

We provide a Dockerfile for building the inference server as a sample file. The Dockerfile is stored in the operator's Pod. Use the command below to retrieve the Dockerfile.

# Use the following command to check the OperatorPod name.
$ kubectl get pods --selector=name=fep-ansible-operator -o name pod/fep-ansible-operator-5985969857-ldj2x
# For Linux clients, copy the plugin using the following command.
$ kubectl cp fep-ansible-operator-5985969857-ldj2x:/opt/fepopr-dockerfile/inference-triton.docker  <destination-path>/inference-triton.docker

Please build the inference server from the acquired Docker file. For build details, refer to the Fujitsu Enterprise Postgres Knowledge Data Management Feature User's Guide. After building, store the image in the container repository. Use the following command to push the inference server:

$ podman push my-triton-image:latest repository.io/triton-inference-server