August 18, 2023

Podman Install

If you don't like docker compose, don't want a bare-metal installation, but still want to leverage the benefits from the isolation and modularity of containers...

Anon

Podman Installation Guide

If you don’t like docker compose, don’t want a bare-metal installation, but still want to leverage the benefits from the isolation and modularity of containers - this is the guide you should use.

Likewise, If you are actively developing RumiTalk, aren’t using the service productively (i.e production environments), you should avoid this guide and look to something easier to work with such as docker compose.

Important: docker and podman commands are for the most part, interoperable and interchangeable. The code instructions below will use (and heavily favor) podman.

Creating the base image

Since RumiTalk is very active in development, it’s recommended for now to build the image locally for the container you plan on using. Thankfully this is easy enough to do.

In your target directory, run the following:

git clone https://github.com/danny-avila/RumiTalk

This will add a directory, RumiTalk into your local environment.

Without entering the RumiTalk directory, add a script ./image.sh with the following:

If you don’t want to run this as a script, you can run the container command rather images

podman build --tag "librechat:local" --file ./RumiTalk/Dockerfile

Note: the downside of running a base container that has a live root is that image revisions need to be done manually. The easiest way is to remove and recreate the image when the container is no longer. If that’s not possible for you, manually updating the image to increment versions can be done manually. Simply amend $image with the version you’re building.

We’ll document how to go about the update process more effectively further on. You wont need to remove your existing containers, or lose any data when updating.

Setting up the env file

Execute the following command to create a env file solely for RumiTalk containers:

cp ./RumiTalk/.env.example .env

This will add the env file to the top level directory that we will create the containers, allowing us to pass it easily as via the --env-file command argument.

Follow this guide to populate the containers with the correct env values for various apis. There are other env values of interest that might be worth changing, documented within the env itself. Afterwords, edit the following lines in the .env file.

HOST=0.0.0.0
MONGO_URI=mongodb://librechat-mongodb:27017/RumiTalk
MEILI_HOST=http://librechat-meilisearch:7700
MEILI_NO_ANALYTICS=true

These values will be uses by some of our containers to correctly use container DNS, using the RumiTalk network.

Creating a network for RumiTalk

If you’re going about this the manual way, it’s likely safe to assume you’re running more than a few different containers and services on your machine. One of the nice features offered by most container engines is that you don’t need to have every single container exposed on the host network. This has the added benefit of not exposing your data and dependant services to other containers on your host.

podman network create librechat

We will be using this network when creating our containers.

Creating dependant containers

RumiTalk currently uses mongoDB and meilisearch, so we’ll also be creating those containers.

Mongodb

Install and boot the mongodb container with the following command:

podman run \
  --name="librechat-mongodb" \
  --network=librechat \
  -v "librechat-mongodb-data:/data/db" \
  --detach \
  docker.io/mongo \
  mongod --noauth

Meilisearch

Install and boot the melisearch container with the following command:

podman run \
  --name="librechat-meilisearch" \
  --network=librechat \
  --env-file="./.env" \
  -v "librechat-meilisearch-data:/meili_data" \
  --detach \
  docker.io/getmeili/meilisearch:v1.0;

Starting RumiTalk

podman run \
  --name="librechat" \
  --network=librechat \
  --env-file="./.env" \
  -p 3080:3080 \
  --detach \
  librechat:local;

If you’re using RumiTalk behind another load balancer, you can omit the -p declaration, you can also attach the container to the same network by adding an additional network argument:

--network=librechat \

--network=mybalancernetwork \

As described by the original -p command argument, it would be possible to access librechat as librechat:3080, mybalancernetwork would be replaced with whatever network your balancer exists.

Auto-starting containers on boot (podman + Linux only)

Podman has a declarative way to ensure that pod starts up automatically on system boot using systemd.

To use this method you need to run the following commands:

First, let’s stop any running containers related to RumiTalk:

podman stop librechat

podman stop librechat-mongodb

podman stop librechat-meilisearch

Next, we’ll update our user’s systemd configuration to enable lingering. In systemd-based systems, when a user logs in and out, user-based services typically terminate themselves to save CPU, but since we’re using rootless containers (which is podman’s preferred way of running), we need to indicate that our user has permission to have user-locked services running after their session ends.

loginctl enable-linger $(whoami)

Next, we’ll create a script somewhere in our home directory using a text editor. Let’s call the script ./install.sh

#!/bin/bash
# Install podman container as systemd container
set -e
name="$1";
podman generate systemd --name "$name" > ~/.config/systemd/user/container-$name.service
systemctl --user enable --now container-$name;

After saving, we’ll update the script to be executable:

chmod +x ./install.sh

Assuming we aren’t running those RumiTalk containers from before, we can enable on-boot services for each of them using the following:

./install.sh librechat-mongodb 

./install.sh librechat-meilisearch 

./install.sh librechat 

The containers (assuming everything was done to par), will be now running using the systemd layer instead of the podman layer. This means services will load on boot, but also means managing these containers is a little more manual and requires interacting with systemd instead of podman directly.

For instance, instead of podman stop {name}, you would instead do systemctl --user stop container-{name} to perform maintenance (such as updates or backups). Likewise, if you need to start the service again you simply can run systemctl --user start container-{name}. If wanting to use auto-boot functionality, interacting with managed containers using podman can cause issues with systemd’s fault tolerance as it can’t correctly indicate the state of a container when interfered with.

Backing up volume containers (podman only)

The podman containers above are using named volumes for persistent data, which means we can’t simply copy files from one place to another. This has benefits though. In podman, we can simply backup the volume into a tape archive format (tarball). To do this, run the following commands:

It’s recommended you stop the containers before running these commands.

podman volume export librechat-meilisearch-data --output "librechat-meilisearch-backup-$(date +"%d-%m-%Y").tar"

podman volume export librechat-mongodb-data --output "librechat-mongodb-backup-$(date +"%d-%m-%Y").tar"

These will leave archive files that you can do what you wish with, including reverting volumes to a previous state if needed. Refer to the official podman documentation for how to do this.

Updating RumiTalk

RumiTalk is still under development, so depending on published images isn’t a huge viability at the moment. Instead, it’s easier to update using git. Data persistence in librechat is managed outside of the main container, so it’s rather simple to do an in-place update.

In the parent directory containing the RumiTalk repo:

(cd RumiTalk && git pull)

systemctl --user stop container-librechat

podman rm -f librechat

podman rmi -f librechat:local

podman build --tag "librechat:local" --file ./RumiTalk/Dockerfile

podman run --name="librechat" --network=librechat --env-file="./.env" -p 3080:3080 --detach librechat:local

podman stop librechat && systemctl --user start container-librechat

Integrating the Configuration File in Podman Setup

When using Podman for setting up RumiTalk, you can also integrate the rumitalk.yaml configuration file.

This file allows you to define specific settings and AI endpoints, such as Mistral AI, tailoring the application to your needs.

After creating your .env file as detailed in the previous steps, follow these instructions to integrate the rumitalk.yaml configuration:

1. Place your rumitalk.yaml file in your project’s root directory.
2. Modify the Podman run command for the RumiTalk container to include a volume argument that maps the rumitalk.yaml file inside the container. This can be done by adding the following line to your Podman run command:

-v "./rumitalk.yaml:/app/rumitalk.yaml"

For example, the modified Podman run command for starting RumiTalk will look like this:

It seems like there was a misunderstanding regarding your request as I provided the split commands earlier. To make it into one command, I would combine all the elements into a single command block:

docker run --name="librechat" --network=librechat --env-file="./.env" -p 3080:3080 --detach librechat:local

By mapping the rumitalk.yaml file into the container, Podman ensures that your custom configurations are applied to RumiTalk, enabling a tailored AI experience.

Ensure that the rumitalk.yaml file is correctly formatted and contains valid settings.

Any errors in this file might affect the functionality of RumiTalk. For more information on configuring rumitalk.yaml, refer to the Custom Endpoints & Configuration.