Containers

• Containers
  • Exercises
    • Running Containers
    • Images
    • Misc
  • Questions
    • Containers 101
    • Commands Commands
    • Images
      • Registry
      • Tags
      • Containerfile
    • Storage
    • Architecture
    • Docker Architecture
    • Docker Compose
    • Networking
    • Docker Networking
    • Security
    • Docker in Production
    • OCI
    • Scenarios

Exercises

<a name="exercises-running-containers"></a>

Running Containers

<a name="exercises-containers-images"></a>

Images

<a name="exercises-containers-misc"></a>

Misc

Questions

<a name="questions-containers-101"></a>

Containers 101

This can be tricky to answer since there are many ways to create a containers:

• Docker
• systemd-nspawn
• LXC

If to focus on OCI (Open Container Initiative) based containers, it offers the following definition: "An environment for executing processes with configurable isolation and resource limitations. For example, namespaces, resource limits, and mounts are all part of the container environment." </b></details>

OCI provides a good explanation: "Define a unit of software delivery called a Standard Container. The goal of a Standard Container is to encapsulate a software component and all its dependencies in a format that is self-describing and portable, so that any compliant runtime can run it without extra dependencies, regardless of the underlying machine and the contents of the container." </b></details>

• An image of a container contains the application, its dependencies and the operating system where the application is executed.

• It's a collection of read-only layers. These layers are loosely coupled

  • Each layer is assembled out of one or more files </b></details>

The primary difference between containers and VMs is that containers allow you to virtualize multiple workloads on a single operating system while in the case of VMs, the hardware is being virtualized to run multiple machines each with its own guest OS. You can also think about it as containers are for OS-level virtualization while VMs are for hardware virtualization.

• Containers don't require an entire guest operating system as VMs. Containers share the system's kernel as opposed to VMs. They isolate themselves via the use of kernel's features such as namespaces and cgroups
• It usually takes a few seconds to set up a container as opposed to VMs which can take minutes or at least more time than containers as there is an entire OS to boot and initialize as opposed to containers which has share of the underlying OS
• Virtual machines considered to be more secured than containers
• VMs portability considered to be limited when compared to containers </b></details>

You should choose VMs when:

• You need run an application which requires all the resources and functionalities of an OS
• You need full isolation and security

You should choose containers when:

• You need a lightweight solution
• Running multiple versions or instances of a single application </b></details>

1. Write a Containerfile/Dockerfile that includes your app (including the commands to run it) and its dependencies
2. Build the image using the Containerfile/Dockefile you wrote
3. You might want to push the image to a registry
4. Run the container using the image you've built </b></details>

• Reusable: container can be used by multiple different users for different usages - production vs. staging, development, testing, etc.
• Lightweight: containers are fairly lightweight which means deployments can be done quickly since you don't need to install a full OS (as in VMs for example)
• Isolation: Containers are isolated environments, usually changes made to the OS won't affect the containers and vice-versa </b></details>

<a name="questions-common-commands"></a>

Commands Commands

Note: I've used Podman in the answers, but other containers engines can be used as well (e.g. Docker)

podman run ubuntu </b></details>

Because the container immediately exits after running the ubuntu image. This is completely normal and expected as containers designed to run a service or a app and exit when they are done running it. To see the container you can run podman ps -a

If you want the container to keep running, you can run a command like sleep 100 which will run for 100 seconds or you can attach to terminal of the container with a command similar: podman container run -it ubuntu /bin/bash </b></details>

podman container ls </b></details>

podman container exec -it [container id/name] bash

This can be done in advance while running the container: podman container run -it [image:tag] /bin/bash </b></details>

False. You have to stop the container before removing it. </b></details>

podman container stop <container id/name> && podman container rm <container id/name> </b></details>

1. Docker client posts the command to the API server running as part of the Docker daemon
2. Docker daemon checks if a local image exists
3. If it exists, it will use it
4. If doesn't exists, it will go to the remote registry (Docker Hub by default) and pull the image locally
5. containerd and runc are instructed (by the daemon) to create and start the container </b></details>

With the -d flag. It will run in the background and will not attach it to the terminal.

docker container run -d httpd or podman container run -d httpd </b></details>

False. Running that command will override the entry point so the httpd service won't run and instead podman will run the ls command. </b></details>

False. podman restart creates an entirely new container with the same ID while reusing the filesystem and state of the original container. </b></details>

podman run -d --name apache1 -p 8080:8080 registry.redhat.io/rhel8/httpd-24
curl 127.0.0.1:8080

</b></details>

podman ps -a will shows also the details of a stopped container. </b></details>

podman search --list-tags IMAGE_NAME </b></details>

<a name="questions-images"></a>

Images

• Most of the images don't contain Kernel. They share and access the one used by the host on which they are running
• Containers intended to run specific application in most cases. This means they hold only what the application needs in order to run </b></details>

podman search snake-game. Surprisingly, there are a couple of matches :)

INDEX       NAME                                                                DESCRIPTION                                      STARS
docker.io   docker.io/dyego/snake-game                                                                                           0
docker.io   docker.io/ainizetap/snake-game                                                                                       0
docker.io   docker.io/islamifauzi/snake-games                                                                                    0
docker.io   docker.io/harish1551/snake-game                                                                                      0
docker.io   docker.io/spkane/snake-game                                         A console based snake game in a container        0
docker.io   docker.io/rahulgadre/snake-game                                     This repository contains all the files to ru...  0

</b></details>

CONTAINER_BINARY=podman
$CONTAINER_BINARY images

Note: you can also use $CONTAINER_RUNTIME image ls </b></details>

CONTAINER_BINARY=podman
$CONTAINER_BINARY pull rhel

</b></details>

True. You should stop and remove the container before trying to remove the image it uses. </b></details>

True </b></details>

False. While this might be true in some cases, it's not guaranteed that you'll pull the latest published image when using the 'latest' tag.

For example, in some images, 'edge' tag is used for the most recently published images. </b></details>

Depends on the container technology being used. For example, in case of Docker, images are stored in /var/lib/docker/ </b></details>

• The layers of an image is where all the content is stored - code, files, etc.
• Each layer is independent
• Each layer has an ID that is an hash based on its content
• The layers (as the image) are immutable which means a change to one of the layers can be easily identified </b></details>

True. These hashes are content based and since images (and their layers) are immutable, any change will cause the hashes to change. </b></details>

In case of Docker, you can use docker image inspect <name> </b></details>

False. They share and access the one used by the host on which they are running. </b></details>

True. When listing images, you might be able to see two images with the same ID but different tags. </b></details>

It's an image without tags attached to it. One way to reach this situation is by building an image with exact same name and tag as another already existing image. It can be still referenced by using its full SHA. </b></details>

In the case of Docker, you could use docker history <name> </b></details>

Creates a new image from a running container. Users can apply extra changes to be saved in the new image version.

Most of the time the user case for using podman commit would be to apply changes allowing to better debug the container. Not so much for creating a new image since commit adds additional overhead of potential logs and processes, not required for running the application in the container. This eventually makes images created by podman commit bigger due to the additional data stored there. </b></details>

True.

One evidence for that can be found in pulling images. Sometimes when you pull an image, you'll see a line similar to the following:

fa20momervif17: already exists

This is because it recognizes such layer already exists on the host, so there is no need to pull the same layer twice. </b></details>

Tags are mutable. This is mean that we can have two different images with the same name and the same tag. It can be very confusing to see two images with the same name and the same tag in your environment. How would you know if they are truly the same or are they different?

This is where "digests` come handy. A digest is a content-addressable identifier. It isn't mutable as tags. Its value is predictable and this is how you can tell if two images are the same content wise and not merely by looking at the name and the tag of the images. </b></details>

True. </b></details>

• Layers are compressed when pushed or pulled
• distribution hash is the hash of the compressed layer
• the distribution hash used when pulling or pushing images for verification (making sure no one tempered with image or layers)
• It's also used for avoiding ID collisions (a case where two images have exactly the same generated ID) </b></details>

1. A client makes a call to the registry to use a specific image (using an image name and optionally a tag)
2. A manifest list is parsed (assuming it exists) to check if the architecture of the client is supported and available as a manifest
3. If it is supported (a manifest for the architecture is available) the relevant manifest is parsed to obtain the IDs of the layers
4. Each layer is then pulled using the obtained IDs from the previous step </b></details>

docker manifest inspect <name> </b></details>

Look for "Cmd" or "Entrypoint" fields in the output of docker image inspec <image name> </b></details>

docker image history <image name>:<tag> </b></details>

1. Docker spins up a temporary container
2. Runs a single instruction in the temporary container
3. Stores the result as a new image layer
4. Remove the temporary container
5. Repeat for every instruction </b></details>

When you build an image for the first time, the different layers are being cached. So, while the first build of the image might take time, any other build of the same image (given that Containerfile/Dockerfile didn't change or the content used by the instructions) will be instant thanks to the caching mechanism used.

In little bit more details, it works this way:

1. The first instruction (FROM) will check if base image already exists on the host before pulling it
2. For the next instruction, it will check in the build cache if an existing layer was built from the same base image + if it used the same instruction
3. If it finds such layer, it skips the instruction and links the existing layer and it keeps using the cache.
4. If it doesn't find a matching layer, it builds the layer and the cache is invalidated.

Note: in some cases (like COPY and ADD instructions) the instruction might stay the same but if the content of what being copied is changed then the cache is invalidated. The way this check is done is by comparing the checksum of each file that is being copied. </b></details>

podman rmi IMAGE

It will fail if some containers are using it. You can then use --force flag for that but generally, it's better if you inspect the containers using the image before doing so.

To delete all images: podman rmi -a </b></details>

• Reduce number of instructions - in some case you may be able to join layers by installing multiple packages with one instructions for example or using && to concatenate RUN instructions
• Using smaller images - in some cases you might be using images that contain more than what is needed for your application to run. It is good to get overview of some images and see whether you can use smaller images that you are usually using.
• Cleanup after running commands - some commands, like packages installation, create some metadata or cache that you might not need for running the application. It's important to clean up after such commands to reduce the image size
• For Docker images, you can use multi-stage builds </b></details>

Pros:

• Smaller image
• Reducing number of layers (especially if the image has lot of layers) Cons:
• No sharing of the image layers
• Push and pull can take more time (because no matching layers found on target) </b></details>

# On the local host
podman save -o some_image.tar IMAGE
rsync some_image.tar SOME_HOST

# On the remote host
podman load -i some_image.tar

</b></details>

False. The image will still be available for use by potential containers in the future.

To remove the container, run podman rmi IMAGE </b></details>

docker image history <image name>:<tag> </b></details>

podman diff IMAGE_NAME </b></details>

True. For mounted files you can use podman inspec CONTAINER_NAMD/ID </b></details>

Registry </b></details>

Registry

• A registry is a service which stores container images and allows users to pull specified images to run containers.
• There are public registries (everyone can access them) and private (accessed only internally in the organization or specific network) </b></details>

A registry contains one or more repositories which in turn contain one or more images. </b></details>

Depends on the containers technology you are using. For example, in case of Docker, it can be done with docker info

> docker info
Registry: https://index.docker.io/v1

</b></details>

For podman, registries can be configured in /etc/containers/registries.conf this way:

[registries.search]
registries = ["quay.io"]

</b></details>

podman image pull ubuntu:latest </b></details>

podman push IMAGE

You can specify a specific registry: podman push IMAGE REGISTRY_ADDRESS </b></details>

• Use tags. Using latest is quite common (which can mean latest build or latest release)
  • tag like 3.1 can be used to reference the latest release/tag of the image like 3.1.6
• Don't use commit for creating new official images as they include the overhead of logs and processes and usually end up with bigger images
• For sharing the image, use a registry (either a public or a private one, depends on your needs) </b></details>

1. Create a Containerfile/Dockerfile and build an image out of it
2. Using podman commit on a running container after making changes to it </b></details>

Tags

Image tags are used to distinguish between multiple versions of the same software or project. Let's say you developed a project called "FluffyUnicorn" and the current release is 1.0. You are about to release 1.1 but you still want to keep 1.0 as stable release for anyone who is interested in it. What would you do? If your answer is create another, separate new image, then you probably want to rethink the idea and just create a new image tag for the new release.

In addition, it's important to note that container registries support tags. So when pulling an image, you can specify a specific tag of that image. </b></details>

podman tag IMAGE:TAG

for example: podman tag FluffyUnicorn:latest </b></details>

False. You can run podman rmi IMAGE:TAG. </b></details>

True. </b></details>

Containerfile

Different container engines (e.g. Docker, Podman) can build images automatically by reading the instructions from a Containerfile/Dockerfile. A Containerfile/Dockerfile is a text file that contains all the instructions for building an image which containers can use. </b></details>

In every Containerfile/Dockerfile, you can find the instruction FROM <image name> which is also the first instruction (at least most of the time. You can put ARG before).

It specifies the base layer of the image to be used. Every other instruction is a layer on top of that base image. </b></details>

• WORKDIR: sets the working directory inside the image filesystems for all the instructions following it
• EXPOSE: exposes the specified port (it doesn't adds a new layer, rather documented as image metadata)
• ENTRYPOINT: specifies the startup commands to run when a container is started from the image
• ENV: sets an environment variable to the given value
• USER: sets the user (and optionally the user group) to use while running the image </b></details>

• Include only the packages you are going to use. Nothing else.
• Specify a tag in FROM instruction. Not using a tag means you'll always pull the latest, which changes over time and might result in unexpected result.
• Do not use environment variables to share secrets
• Use images from official repositories
• Keep images small! - you want them only to include what is required for the application to run successfully. Nothing else.
• If are using the apt package manager, you might want to use 'no-install-recommends' with apt-get install to install only main dependencies (instead of suggested, recommended packages) </b></details>

Docker docs: "A build’s context is the set of files located in the specified PATH or URL" </b></details>

COPY takes in a source and destination. It lets you copy in a file or directory from the build context into the Docker image itself.

ADD lets you do the same, but it also supports two other sources. You can use a URL instead of a file or directory from the build context. In addition, you can extract a tar file from the source directly into the destination. </b></details>

RUN lets you execute commands inside of your Docker image. These commands get executed once at build time and get written into your Docker image as a new layer. CMD is the command the container executes by default when you launch the built image. A Containerfile/Dockerfile can only have one CMD. You could say that CMD is a Docker run-time operation, meaning it’s not something that gets executed at build time. It happens when you run an image. A running image is called a container. </b></details>

The following command is executed from within the directory where Dockefile resides:

docker image build -t some_app:latest . podman image build -t some_app:latest . </b></details>

One option is to use hadolint project which is a linter based on Containerfile/Dockerfile best practices. </b></details>

Instructions such as FROM, COPY and RUN, create new image layers instead of just adding metadata. </b></details>

Instructions such as ENTRYPOINT, ENV, EXPOSE, create image metadata and they don't create new layers. </b></details>

True </b></details>

ENTRYPOINT ["cmd", "param0", "param1"]
CMD ["param0"]

ENTRYPOINT cmd param0 param1
CMD param0

The first form is also referred as "Exec form" and the second one is referred as "Shell form".

The second one (Shell form) wraps the commands in /bin/sh -c hence creates a shell process for it.

While using either Exec form or Shell form might be fine, it's the mixing that can lead to unexpected results.

Consider:

ENTRYPOINT ["ls"]
CMD /tmp

That would results in running ls /bin/sh -c /tmp </b></details>

True but in case of ENTRYPOINT and CMD only the last instruction takes effect. </b></details>

The ENTRYPOINT from the base image is being used in such case. </b></details>

CMD </b></details>

<a name="questions-containers-storage"></a>

Storage

It means the contents of the container and the data generated by it, is gone when the container is removed. </b></details>

False. Containers are not built to store persistent data and even if it's possible with some implementations, it might not perform well in case of applications with intensive I/O operations. </b></details>

In order to reclaim the storage of a container, you have to remove it. </b></details>

CONTAINER_BINARY=podman
$CONTAINER_BINARY volume create some_volume

</b></details>

CONTAINER_BINARY=podman
mkdir /tmp/dir_on_the_host

$CONTAINER_BINARY run -v /tmp/dir_on_the_host:/tmp/dir_on_the_container IMAGE_NAME

In some systems you'll have also to adjust security on the host itself:

podman unshare chown -R UID:GUID /tmp/dir_on_the_host
sudo semanage fcontext -a -t container_file_t '/tmp/dir_on_the_host(/.*)?'
sudo restorecon -Rv /tmp/dir_on_the_host

</b></details>

<a name="questions-containerfile"></a>

<a name="questions-architecture"></a>

Architecture

Through the use of namespaces and cgroups. Linux kernel has several types of namespaces:

• Process ID namespaces: these namespaces include independent set of process IDs
• Mount namespaces: Isolation and control of mountpoints
• Network namespaces: Isolates system networking resources such as routing table, interfaces, ARP table, etc.
• UTS namespaces: Isolate host and domains
• IPC namespaces: Isolates interprocess communications
• User namespaces: Isolate user and group IDs
• Time namespaces: Isolates time machine </b></details>

• cgroups (Control Groups): used for limiting the amount of resources a certain groups of processes (and their children of course) use. This way, a group of processes isn't consuming all host resources and other groups can run and use part of the resources as well

• namespaces: same as cgroups, namespaces isolate some of the system resources so it's available only for processes in the namespace. Differently from cgroups the focus with namespaces is on resources like mount points, IPC, network, ... and not about memory and CPU as in cgroups

• SElinux: the access control mechanism used to protect processes. Unfortunately to this date many users don't actually understand SElinux and some turn it off but nonetheless, it's a very important security feature of the Linux kernel, used by container as well

• Seccomp: similarly to SElinux, it's also a security mechanism, but its focus is on limiting the processes in regards to using system calls and file descriptors </b></details>

Docker/Podman CLI passes your request to Docker daemon. Docker/Podman daemon downloads the image from Docker Hub Docker/Podman daemon creates a new container by using the image it downloaded Docker/Podman daemon redirects output from container to Docker CLI which redirects it to the standard output </b></details>

In short:

Cgroups = limits how much you can use; namespaces = limits what you can see (and therefore use)

Cgroups involve resource metering and limiting: memory CPU block I/O network

Namespaces provide processes with their own view of the system

Multiple namespaces: pid,net, mnt, uts, ipc, user

</b></details>

• cspaces

• namegroups

• namespaces

• cgroups

• ELlinux

• SElinux</summary>

  <b>

• namespaces

• cgroups

• SElinux </b></details>

True. The ephemeral storage layer is added on top of the base image layer and is exclusive to the running container. This way, containers created from the same base image, don't share the same storage. </b></details>

<a name="questions-docker-architecture"></a>

Docker Architecture

1. Runtime - responsible for starting and stopping containers
2. Daemon - implements the Docker API and takes care of managing images (including builds), authentication, security, networking, etc.
3. Orchestrator </b></details>

• Docker daemon
• containerd
• runc </b></details>

• The low level runtime is called runc
• It manages every container running on Docker host
• Its purpose is to interact with the underlying OS to start and stop containers
• Its reference implementation is of the OCI (Open Containers Initiative) container-runtime-spec
• It's a small CLI wrapper for libcontainer </b></details>

• The high level runtime is called containerd
• It was developed by Docker Inc and at some point donated to CNCF
• It manages the whole lifecycle of a container - start, stop, remove and pause
• It take care of setting up network interfaces, volume, pushing and pulling images, ...
• It manages the lower level runtime (runc) instances
• It's used both by Docker and Kubernetes as a container runtime
• It sits between Docker daemon and runc at the OCI layer

Note: running ps -ef | grep -i containerd on a system with Docker installed and running, you should see a process of containerd </b></details>

False. The Docker daemon performs higher-level tasks compared to containerd.

It's responsible for managing networks, volumes, images, ... </b></details>

docker.io/library/busybox:latest resolved to a manifestList object with 9 entries; looking for a unknown/amd64 match

found match for linux/amd64 with media type application/vnd.docker.distribution.manifest.v2+json, digest sha256:400ee2ed939df769d4681023810d2e4fb9479b8401d97003c710d0e20f7c49c6

pulling blob "sha256:61c5ed1cbdf8e801f3b73d906c61261ad916b2532d6756e7c4fbcacb975299fb Downloaded 61c5ed1cbdf8 to tempfile /var/lib/docker/tmp/GetImageBlob909736690

Applying tar in /var/lib/docker/overlay2/507df36fe373108f19df4b22a07d10de7800f33c9613acb139827ba2645444f7/diff" storage-driver=overlay2

Applied tar sha256:514c3a3e64d4ebf15f482c9e8909d130bcd53bcc452f0225b0a04744de7b8c43 to 507df36fe373108f19df4b22a07d10de7800f33c9613acb139827ba2645444f7, size: 1223534 </b></details>

1. The Docker client converts the run command into an API payload
2. It then POST the payload to the API endpoint exposed by the Docker daemon
3. When the daemon receives the command to create a new container, it makes a call to containerd via gRPC
4. containerd converts the required image into an OCI bundle and tells runc to use that bundle for creating the container
5. runc interfaces with the OS kernel to pull together the different constructs (namespace, cgroups, etc.) used for creating the container
6. Container process is started as a child-process of runc
7. Once it starts, runc exists </b></details>

False. While this was true at some point, today the container runtime isn't part of the daemon (it's part of containerd and runc) so stopping or killing the daemon will not affect running containers. </b></details>

True </b></details>

False. Once a container is created, the parent runc process exists. </b></details>

shim is the process that becomes the container's parent when runc process exists. It's responsible for:

• Reporting exit code back to the Docker daemon
• Making sure the container doesn't terminate if the daemon is being restarted. It does so by keeping the stdout and stdin open </b></details>

1. To remove one or more Docker images use the docker container rm command followed by the ID of the containers you want to remove.
2. The docker system prune command will remove all stopped containers, all dangling images, and all unused networks
3. docker rm $(docker ps -a -q) - This command will delete all stopped containers. The command docker ps -a -q will return all existing container IDs and pass them to the rm command which will delete them. Any running containers will not be deleted. </b></details>

Via the local socket at /var/run/docker.sock </b></details>

A Docker image is built up from a series of layers. Each layer represents an instruction in the image’s Containerfile/Dockerfile. Each layer except the very last one is read-only. Each layer is only a set of differences from the layer before it. The layers are stacked on top of each other. When you create a new container, you add a new writable layer on top of the underlying layers. This layer is often called the “container layer”. All changes made to the running container, such as writing new files, modifying existing files, and deleting files, are written to this thin writable container layer. The major difference between a container and an image is the top writable layer. All writes to the container that add new or modify existing data are stored in this writable layer. When the container is deleted, the writable layer is also deleted. The underlying image remains unchanged. Because each container has its own writable container layer, and all changes are stored in this container layer, multiple containers can share access to the same underlying image and yet have their own data state. </b></details>

<a name="questions-docker-compose"></a>

Docker Compose

Compose is a tool for defining and running multi-container Docker applications. With Compose, you use a YAML file to configure your application’s services. Then, with a single command, you create and start all the services from your configuration.

For example, you can use it to set up ELK stack where the services are: elasticsearch, logstash and kibana. Each running in its own container.

In general, it's useful for running applications which composed out of several different services. It let's you manage it as one deployed app, instead of different multiple separate services. </b></details>

• Define the services you would like to run together in a docker-compose.yml file
• Run docker-compose up to run the services </b></details>

Multi-stages builds allow you to produce smaller container images by splitting the build process into multiple stages.

As an example, imagine you have one Containerfile/Dockerfile where you first build the application and then run it. The whole build process of the application might be using packages and libraries you don't really need for running the application later. Moreover, the build process might produce different artifacts which not all are needed for running the application.

How do you deal with that? Sure, one option is to add more instructions to remove all the unnecessary stuff but, there are a couple of issues with this approach:

1. You need to know what to remove exactly and that might be not as straightforward as you think
2. You add new layers which are not really needed

A better solution might be to use multi-stage builds where one stage (the build process) is passing the relevant artifacts/outputs to the stage that runs the application. </b></details>

True. This allows us to eventually produce smaller images. </b></details>

By default, Docker uses everything (all the files and directories) in the directory you use as build context.

.dockerignore used for excluding files and directories from the build context </b></details>

<a name="questions-networking"></a>

Networking

CNM (Container Network Model):

• Requires distrubited key value store (like etcd for example) for storing the network configuration
• Used by Docker CNI (Container Network Interface):
• Network configuration should be in JSON format </b></details>

<a name="questions-docker-networking"></a>

Docker Networking

Docker is using the CNM (Container Network Model) design specification.

The implementation of CNM specification by Docker is called "libnetwork". It's written in Go. </b></details>

• Networks

• Endpoints

• Sandboxes</summary>

  <b>

• Networks: software implementation of an switch. They used for grouping and isolating a collection of endpoints.

• Endpoints: Virtual network interfaces. Used for making connections.

• Sandboxes: Isolated network stack (interfaces, routing tables, ports, ...) </b></details>

True. An endpoint can connect only to a single network. </b></details>

• Native service discovery
• ingress-based load balancer
• network control plane and management plane </b></details>

<a name="questions-security"></a>

Security

• Install only the necessary packages in the container
• Don't run containers as root when possible
• Don't mount the Docker daemon unix socket into any of the containers
• Set volumes and container's filesystem to read only
• DO NOT run containers with --privilged flag </b></details>

• Install only the necessary packages in the container
• Set volumes and container's filesystem to read only
• DO NOT run containers with --privilged flag </b></details>

<a name="questions-docker-in-production"></a>

Docker in Production

Images:

• Use images from official repositories
• Include only the packages you are going to use. Nothing else.
• Specify a tag in FROM instruction. Not using a tag means you'll always pull the latest, which changes over time and might result in unexpected result.
• Do not use environment variables to share secrets
• Keep images small! - you want them only to include what is required for the application to run successfully. Nothing else. Components:
• Secured connection between components (e.g. client and server) </b></details>

False. Communication between client and server shouldn't be done over HTTP since it's insecure. It's better to enforce the daemon to only accept network connection that are secured with TLS.

Basically, the Docker daemon will only accept secured connections with certificates from trusted CA. </b></details>

Restart Policies. It allows you to automatically restart containers after certain events. </b></details>

• always: restart the container when it's stopped (not with docker container stop)
• unless-stopped: restart the container unless it was in stopped status
• no: don't restart the container at any point (default policy)
• on-failure: restart the container when it exists due to an error (= exit code different than zero) </b></details>

<a name="questions-rootless-containers"></a>

Historically, user needed root privileges to run containers. One of the most basic security recommendations is to provide users with minimum privileges for what they need.

For containers it's been the situation for a long time and still for running some containers today from docker.io, you'll need to have root privileges. </b></details>

Yes, the full list can be found here.

Some worth to mention:

• No binding to ports smaller than 1024
• No images sharing CRI-O or other rootful users
• No support running on NFS or parallel filesystem homerdirs
• Some commands don't work (mount, podman stats, checkpoint, restore, ...) </b></details>

In rootless containers, user namespace appears to be running as root but it doesn't, it's executed with regular user privileges. If an attacker manages to get out of the user space to the host with the same privileges, there's not much he can do because it's not root privileges as opposed to containers that run with root privileges. </b></details>

Networking is usually managed by Slirp in rootless containers. Slirp creates a tap device which is also the default route and it creates it in the network namespace of the container. This device's file descriptor passed to the parent who runs it in the default namespace and the default namespace connected to the internet. This enables communication externally and internally. </b></details>

New drivers were created to allow creating filesystems in a user namespaces. Drivers like the FUSE-OverlayFS. </b></details>

<a name="questions-oci"></a>

OCI

OCI (Open Container Initiative) is an open governance established in 2015 to standardize container creation - mostly image format and runtime. At that time there were a number of parties involved and the most prominent one was Docker.

Specifications published by OCI:

• image-spec
• runtime-spec </b></details>

Create, Kill, Delete, Start and Query State. </b></details>

<a name="questions-containers-scenarios"></a>

Scenarios

podman commit can be a good choice for that. You can create a new image of the running container (with the issue) and share that new image with your team members.

What you probably want to avoid using:

• Using something as podman save/load as it applies on an image, not a running container (so you'll share the image but the issue might not be reproduced when your team members run a container using it)
• Modifying Containerfile/Dockerfile as you don't really want to add environment variables meant for debugging to the source from which you usually build images </b></details>

Use tags. You can distinguish between different releases of a project using image tags. There is no need to create an entire separate image for version/release of a project. </b></details>