Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision		 Previous revision
docs:guide-user:virtualization:docker_host [2023/08/16 15:50] – Podman stokitodocs:guide-user:virtualization:docker_host [2024/10/08 16:59] (current) – [Docker Community Edition] luci-app-dockerman now depends on dockerd stokito
Line 1: Line 1:
 ======= OpenWrt as Docker container host ======= ======= OpenWrt as Docker container host =======
  
-OpenWrt can be a [[wp>Docker_(software)|Docker]] host on x86-64, Aarch64, and other supported architectures.+[[wp>Docker_(software)|Docker]] uses OS-level virtualization to deliver software in packages called containers. This is used to automate the deployment of applications in lightweight containers so that applications can work efficiently in different environments in isolation.
  
-There are two ways to use Docker as a host, install Docker Community Edition, or use native OpenWrt tools that support Docker container specification.+To run containers, users may install Docker Community Edition, use native OpenWrt tools, or use Podman. While Docker CE is perhaps the most typical method, this guide covers several options.
  
-You will probably need to [[docs:guide-user:storage:usb-drives|set up storage first]] as a place to store the containers and data.+===== Prerequisites =====
  
-Also in most cases you will be running the container as a specific user and will give it access to some folder outside the container, where it can store its configuration and the data. So you will probably need to [[docs:guide-user:additional-software:create-new-users|create new users and groups for applications or system services]], create the folders for the configuration and data, and then change the owner of these folders to the user you will run the container as.+For devices with small flash partitions you may need to [[docs:guide-user:storage:usb-drives|set up storage]] to store the containers and data.
  
-===== Install Docker Community Edition ===== +Also in many cases you will be running the container as a specific user that will need access to some folder outside the container for its configuration and data. So you will probably need to [[docs:guide-user:additional-software:create-new-users|create new users and groups]] for applications, create folders, and then change the owner of these folders to the user who will run the container.
-  * Install **docker-ce** package for command line tools\\ +
-  * Install **luci-app-dockerman** package to get a control panel for containers in Luci+
  
-The default folder for docker in the dockerman luci interface is **/opt/docker** so you want to mount your storage at **/opt** or change the folder in **Docker** >  **Overview** > **Docker Root Dir** and then restart the dockerd service.+===== Docker Community Edition ===== 
 +First you need to install the ''dockerd'' (''opkg install dockerd'' ~16 Mb) daemon that will run containers. 
 +It provides an [[https://docs.docker.com/engine/api/|Docker Engine API]] available by a local UNIX socket and HTTP REST that can be called from outside if allowed.
  
-==== Add an image ==== +Then you need a //client// e.g. ''docker'' (''opkg install docker'' ~12 Mb) to connect for the daemon API and start containers. 
-To add an image, search it on [[https://hub.docker.com/|Docker Hub]], then copy the image name from the **docker pull** text box. For example, if the text is **docker pull linuxserver/transmission**, then copy **linuxserver/transmission**.+To save space you can use the ''docker'' from another computer if the Docker Engine is opened. 
 + 
 +As a GUI client install ''luci-app-dockerman'' (''opkg install luci-app-dockerman'') to get a LuCI page to manage containers. It can work with ''dockerd'' on both remote and local hosts. Since the ''luci-app-dockerman'' version v0.6 it will also install the ''dockerd'' and even ''docker-compose'' as a dependency. 
 + 
 +The default folder for docker in the dockerman Luci interface is **/opt/docker** so mount your storage at **/opt** or change the folder in **Docker** >  **Overview** > **Docker Root Dir** then restart the dockerd service. 
 + 
 + 
 +==== Adding images ==== 
 +Search for an image on [[https://hub.docker.com/|Docker Hub]], then copy the image name from the **Docker Pull Command** text box. For example, if the text is **docker pull linuxserver/transmission**, then copy **linuxserver/transmission**.
  
 In Luci go to **Docker** > **Images** and paste that text in the **Pull Image** box, then click **Pull**. The page will show the download progress. In Luci go to **Docker** > **Images** and paste that text in the **Pull Image** box, then click **Pull**. The page will show the download progress.
  
-For longer container pulls Luci might timeout so you will need to use the command line. For example, unifi-controller images include java runtime environment and approach 500MB, so you could SSH in and enter: docker pull linuxserver/unifi-controller.+Note for larger container pulls LuCI might timeout, so you will need to use the command line. For example, Unifi-network-application includes java runtime environment and approaches 500MB. For this you can instead use SSH and enter: ''docker pull lscr.io/linuxserver/unifi-network-application:latest''.
  
-Then in Luci go to **Docker** > **Containers** > **Add**. In the new container page, select the docker image from the **Docker Image** menu, and then set all other parameters (usually the available/useful parameters are described in the description of the container on Docker Hub), then press Submit to create the container.+Once you have your images, in Luci go to **Docker** > **Containers** > **Add**. In the new container page select the docker image from the **Docker Image** menu, then set all other parameters (usually the available/useful parameters are described in the description of the container on Docker Hub), then press Submit to create the container.
  
-==== Configure The Docker CE Engine Daemon ====+==== Configure the Docker daemon ====
 Config is located in ''/etc/config/dockerd''. Config is located in ''/etc/config/dockerd''.
  
Line 46: Line 54:
  
  
- +===== Native OpenWrt tools ===== 
-===== Use native OpenWrt tools ===== +Instead of running Docker CE users may want to use the procd init system which supports Open Container Initiative Runtime Specification set by [[https://opencontainers.org/|Opencontainers.org]]. This extends its slim containers ('ujail') capability. The uxc command line tool handles the basic operations on containers as defined by the spec. This allows to use it as a drop-in replacement for Docker's 'runc' (or 'crun') on OpenWrt hosts with a reduced footprint.
-Procd init system now supports Open Container Initiative Runtime Specification, extending its slim containers ('ujail') capability.\\ +
-The uxc command line tool handles the basic operations on containers as defined by the spec.\\ +
-This allows to use it as a drop-in replacement for Docker's 'runc' (or 'crun') on OpenWrt hosts with a significantly reduced footprint.+
  
 Detailed but possibly outdated info available on https://gitlab.com/prpl-foundation/prplos/prplos/-/wikis/uxc Detailed but possibly outdated info available on https://gitlab.com/prpl-foundation/prplos/prplos/-/wikis/uxc
-==== install packages ==== 
-For 20.0x install the following: 
-<code> 
-opkg install kmod-veth uxc ujail-console 
-</code> 
- 
-For newer snapshots: 
  
 +==== Install packages ====
 +Install the following:
 <code> <code>
 opkg install kmod-veth uxc procd-ujail procd-ujail-console opkg install kmod-veth uxc procd-ujail procd-ujail-console
 </code> </code>
-==== create veth pair for container ====+ 
 +==== Create veth pair for container ====
 <code> <code>
 uci batch <<EOF uci batch <<EOF
Line 83: Line 84:
 </code> </code>
  
-====creating an OCI run-time bundle====+==== Creating an OCI run-time bundle ====
 To create an OCI run-time bundle, which is needed for uxc, follow these steps. To create an OCI run-time bundle, which is needed for uxc, follow these steps.
  
Line 101: Line 102:
 This is quite cumbersome. If someone knows a better way, please do update this page. This is quite cumbersome. If someone knows a better way, please do update this page.
  
-====import a OCI runtime container====+==== Import a OCI runtime container ====
 (assuming OCI run-time bundle with config.json in /mnt/sda3/debian) (assuming OCI run-time bundle with config.json in /mnt/sda3/debian)
 <code> <code>
Line 117: Line 118:
  
 ===== Podman ===== ===== Podman =====
-https://podman.io/ is another alternative to Docker and it is compatible with docker client commands. +https://podman.io/ is alternative to Docker and it is compatible with Docker client commands. 
-It has a smaller size than the Docker so you may consider to use it instead. +Here is example setup using podman to create web server container with proxy.
-There is also [[https://github.com/oskarirauta/packages2/tree/main/luci/luci-app-podman|luci-app-podman]] in active development.+
  
 +Install necessary packages:
 +<code bash>
 +opkg install conmon crun catatonit netavark podman external-protocol
 +</code>
 +
 +If you want to use rootless containers, you need additional packages, such as slirp4netns.
 +There's also authoritative dns server for netavark available as package aardvark-dns.
 +This guide excludes their setup currently.
 +
 +**Network**
 +
 +Let's start by reviewing our container network's settings.
 +/etc/containers/networks/podman.json:
 +<code>
 +{
 +     "name": "podman",
 +     "id": "5ef894788befd4d42498314b6e66282ca730aa2e1e82f9b9597bf4d1725ca074",
 +     "driver": "bridge",
 +     "network_interface": "podman0",
 +     "created": "2023-02-20T08:56:34.652030952Z",
 +     "subnets": [
 +          {
 +               "subnet": "10.129.0.0/24",
 +               "gateway": "10.129.0.1"
 +          }
 +     ],
 +     "ipv6_enabled": false,
 +     "internal": false,
 +     "dns_enabled": true,
 +     "ipam_options": {
 +          "driver": "host-local"
 +     }
 +}
 +</code>
 +
 +I have a rather large network (10.0.0.0/9) so that's why the //odd// subnet - you can choose your own, but this is a example setup based on my settings.
 +In this file you define your network named podman. You can have multiple networks.
 +
 +**Firewall/Zone**
 +
 +Next, we make sure internal port forwarding/firewalling of podman is disabled, check /etc/config/netavark:
 +<code>
 +config firewall
 + option driver 'none'
 +</code>
 +
 +We do this to rather use openwrt's own firewall, as when using podman's/netavark's, rules are lost every time that firewall is re-loaded, including
 +when you start your first container network, when podman0 interface comes up.
 +
 +Next is time to setup network and firewall on the openwrt's side, add this to /etc/config/network:
 +<code>
 +config interface 'podman'
 + option proto 'external'
 + option device 'podman0'
 +</code>
 +
 +And we also want to use firewall, in this setup we allow access from lan to podman, but not the other way around, we also
 +grant access from wan, and to wan. /etc/config/firewall:
 +<code>
 +config zone
 + option name 'podman'
 + option input 'REJECT'
 + option output 'ACCEPT'
 + option forward 'ACCEPT'
 + option mtu_fix '1'
 + list network 'podman'
 +
 +config forwarding
 + option src 'lan'
 + option dest 'podman'
 +
 +config forwarding
 + option src 'podman'
 + option dest 'wan'
 +
 +config forwarding
 + option src 'wan'
 + option dest 'podman'
 +</code>
 +
 +Now that we have blocked access to LAN, our containers are missing access to DNS, unless we configure them to use something else,
 +such as 8.8.8.8 - so we make a exception, containers can connect to lan, but only on port 53 (DNS):
 +<code>
 +config rule
 + option name 'Allow-Podman-DNS'
 + option src 'podman'
 + option dest_port '53'
 + option target 'ACCEPT'
 +</code>
 +
 +Now initial network setup is complete.
 +
 +**Service**
 +
 +Next we make sure that podman service is started on boot. This is optional, but if you want to follow this guide, it comes handy later.
 +Podman service does not create/start/etc any containers, it only starts background service and creates unix socket used for communication.
 +This socket is located at /var/run/podman/podman.sock - this socket accepts similar/compatible communication as docker. A so called
 +podman-docker-compatibility package usually only contains a link to this socket to standard docker's socket path, and a docker wrapper
 +script that forwards it's command-line to podman command. Service actually starts when you use podman, but we want to make sure it's listed
 +as openwrt instance, and for further more advanced setup, this is helpful.
 +
 +<code bash>
 +/etc/init.d/podman enable
 +</code>
 +
 +After reboot, you can start using your podman setup. Guide continues, we make a web server and caddy proxy as example projects
 +and handle forwarding of traffic to them. We will store our container data in /srv. 
 +
 +**Container storage (optional) **
 +
 +I have a lot of disk space available, so I set my graphroot to hard drive, instead of
 +storing container data in RAM, such as /tmp or /var which it defaults. First make a proper path:
 +<code bash>
 +mkdir -p /srv/.podman/storage
 +</code>
 +
 +And edit your /etc/containers/storage.conf:
 +<code>
 +graphroot = "/srv/.podman/storage"
 +</code>
 +default was: "///var/lib/containers/storage//"
 +
 +Without setting graphroot, your setup works also, it's just that I happen to have a lot of disk
 +space in my setups, I rather store them on hard drive instead of memory, such as /tmp or /var.
 +
 +**Local image storage (optional)**
 +
 +My setup builds my containers on every boot and I want to speed up that process, so I want images for Caddy proxy and nginx web server to
 +be stored in permanent storage. But beware, every time you want to make changes to these images, you need to restart this process completely
 +from beginning, first by resetting //additionalimagestores// setting to it's default value, remove physical image files and ofcourse before
 +all this, you must remove containers using these images, along the images, from Podman's own system, and finally pull new images, and
 +restore //additionalimagestores// setting to your image path. Also as Podman's service lacks stop functionality, some reboots are
 +necessary as well. Complicated? Yes it is, but in the end it might pay off, as images don't need to be pulled after every reboot.
 +To begin with, reboot your computer and make sure any containers are not created, started or even exist, this will help to avoid problems.
 +
 +Create directory /srv/.podman/images:
 +<code>
 +mkdir -p /srv/.podman/images
 +</code>
 +
 +Then check your /etc/containers/storage.conf:
 +<code>
 +additionalimagestores = []
 +</code>
 +
 +If you have to change this line, you must reboot to podman service restart so this takes effect.
 +Following line should be found, it is default setting. With this line, we do not have images locally stored in permanent store.
 +Next issue following commands:
 +<code>
 +podman --root /srv/.podman/images images
 +podman --root /srv/.podman/images pull docker.io/me/my_caddy_image:latest
 +podman --root /srv/.podman/images pull docker.io/me/my_nginx_image:latest
 +</code>
 +replace image urls with your own, first command sets up path images as local image store,
 +and next commands pull your images to local image store. You can also pull your pause
 +image, for example: //k8s.gcr.io/pause:3.5//
 +
 +After this, make changes to your /etc/containers/storage.conf:
 +<code>
 +additionalimagestores = [
 +        "/srv/.podman/images"
 +]
 +</code>
 +and reboot. Now when creating containers that use locally stored images, they do not need
 +to be pulled from internet, they are instantly available. If you use not locally stored
 +images, they work fine; they just are pulled from internet.
 +
 +This is a one time operation; changes, removals, adds of locally stored images cannot
 +be modified very easily. To restart process; remove containers using locally stored images,
 +remove then these images from podman (podman image rm <imageid>), change
 +your //additionalimagestores// back to [], disable podman service, and make sure, podman
 +service doesn't start by creation/start of containers during boot. Then you remove all files,
 +from /srv/.podman/images:
 +<code>
 +rm -rf /srv/.podman/images
 +</code>
 +
 +reboot again, and begin this again from start of this section of guide.
 +
 +**Pod**
 +We will start by creating a pod. Pod can hold multiple containers, they share some attributes,
 +such as ip address. As we are trying to build a web server setup, we want IP address to be
 +always same for this pod. I have created a script /srv/create.sh to construct this pod:
 +<code bash>
 +#!/bin/sh
 +
 +podman pod create \
 + --replace \
 + --name servers \
 + --hostname srv \
 + --ip 10.129.0.2
 +
 +podman pod start servers
 +</code>
 +This creates, or replaces if one exists, pod named servers, gives it a hostname srv (not important) and a static
 +IP address 10.129.0.2.
 +
 +**Containers**
 +All configurations and statically exported data is also in /srv. In /srv/caddy I have all needed to build my
 +caddy container, such as configurations and what ever caddy container of your choice needs. I also have a
 +build script there, /srv/caddy/create.sh:
 +<code bash>
 +#!/bin/sh
 +
 +podman create \
 + --name caddy \
 + --pod servers \
 + --replace \
 + --systemd false \
 + --label app=caddy \
 + --volume /srv/caddy/conf/:/etc/caddy/:Z,rw \
 + --volume /srv/caddy/htdocs/:/var/htdocs/:z,rw \
 + --volume /srv/caddy/logs/:/var/log/:z,rw \
 + --volume /dev/log:/dev/log:Z,rw \
 + --mount="type=bind,src=/etc/acme/domain.tld_ecc/domain.tld.cer,dst=/etc/caddy/ssl/server.pem,ro=true,idmap=uids=0-82-1;gids=0-82-1" \
 + --mount="type=bind,src=/etc/acme/domain.tld_ecc/domain.tld.key,dst=/etc/caddy/ssl/server.key,ro=true,idmap=uids=0-82-1;gids=0-82-1" \
 + docker.io/me/my_caddy_image:latest
 +
 +podman start caddy
 +</code>
 +
 +In this guide I do not review configuration of Caddy, look it up from caddy's docs. My caddy is set to run as user www:www-data which in that
 +setup are uid 82 and gid 82, acme is used to fetch certificates, but user www(82) cannot read root owned files, so we use idmapping to map
 +those 2 files for user www:www-data. There are multiple ways to do this, this is just one approach. You could also setup a system that chmod's
 +those files to be available for reading to everyone, or at least for user and/or group 82. Or copy them locally and chown them in that
 +location statically.
 +
 +And I have a similar script for nginx:
 +<code bash>
 +#!/bin/sh
 +
 +podman create \
 + --name nginx \
 + --pod servers \
 + --replace \
 + --systemd false \
 + --label app=nginx \
 + --volume /srv/nginx/conf/:/etc/nginx/:Z,rw \
 + --volume /srv/nginx/logs/:/var/log/nginx/:Z,rw \
 + --volume /srv/nginx/htdocs/:/var/htdocs/:z,rw \
 + --volume /dev/log:/dev/log:Z,rw \
 + docker.io/me/my_nginx_image:latest
 +
 +podman start nginx
 +</code>
 +
 +Now after you have configured properly your caddy and nginx, we should have a server properly running.
 +We need to setup redirections from wan.
 +
 +**Expose to wan**
 +
 +Now that we have caddy serving at 10.129.0.2, ports 80 and 443, we edit /etc/config/firewall again:
 +<code>
 +config redirect
 + option name 'Allow-HTTP'
 + option src 'wan'
 + option dest 'podman'
 + option src_dport '80'
 + option dest_ip '10.129.0.2'
 + option dest_port '80'
 + option proto 'tcp'
 + option reflection '0'
 + option target 'DNAT'
 + option enabled '1'
 +
 +config redirect
 + option name 'Allow-HTTPS'
 + option src 'wan'
 + option dest 'podman'
 + option src_dport '443'
 + option dest_ip '10.129.0.2'
 + option dest_port '443'
 + option proto 'tcp'
 + option reflection '0'
 + option target 'DNAT'
 + option enabled '1'
 +</code>
 +
 +**Automation**
 +
 +Finally, we want our pod and containers to build and start during boot, we also have
 +acme handling our certificates, so we want to restart caddy when certificates are renewed.
 +
 +I added /srv/scripts directory, and added there file restart_caddy.sh:
 +<code bash>
 +#!/bin/sh
 +
 +/etc/init.d/podman enabled || exit
 +logger -t acme -p daemon.info "SSL certificates renewed, restarting container servers:caddy"
 +
 +podman stop caddy
 +sleep 1
 +podman start caddy
 +</code>
 +
 +And then rest is handled by /etc/rc.local:
 +<code bash>
 +# Put your custom commands here that should be executed once
 +# the system init finished. By default this file does nothing.
 +
 +add_podman_trigger() {
 +
 + local counter=10
 + local running=0
 +
 + [ -x "/etc/init.d/acme" ] || exit
 + /etc/init.d/acme enabled || exit
 +
 + while [ "$counter" -gt 0 ]; do
 + [ "$(service podman status)" = "running" ] && {
 + running=1
 + counter=0
 + } || {
 + sleep 1
 + counter=$(($counter-1))
 + }
 + done
 +
 + [ "$running" -eq 1 ] && {
 + ubus call service set '{ "name": "podman", "triggers": [[ "acme.renew", [[ "run_script", "/srv/scripts/restart_caddy.sh" ]], 2000 ]], "data": {}}'
 + logger -t podman -p daemon.info "podman: added service trigger for acme.renew event to restart servers:caddy"
 + }
 +}
 +
 +start_podman_services() {
 +
 + /etc/init.d/podman enabled && {
 + [ -f /tmp/.podman_created ] || {
 +
 + touch /tmp/.podman_created
 +
 + sleep 1
 + /srv/create.sh
 + sleep 2
 + /srv/caddy/create.sh
 + sleep 2
 + /srv/nginx/create.sh
 +
 + add_podman_trigger &
 + }
 + }
 +}
 +</code>
  
 +This is why starting podman service with /etc/init.d/podman comes handy, we can ignore
 +all container related during boot, by just simply disabling service as nothing podman
 +related is started if service is disabled. This builds our pod and both containers and
 +then adds a trigger for podman service to restart caddy when SSL certificates are renewed.
 +There's a routine that checks if podman service has started, because trigger must be
 +added AFTER podman service has started.
  • Last modified: 2023/08/16 15:50
  • by stokito