dgr (pronounced "digg-er") is a command line utility designed to build and to configure at runtime App Containers Images (ACI) and App Container Pods (POD) based on convention over configuration.
dgr allows you to build generic container images for a service and to configure them at runtime. Therefore you can use the same image for different environments, clusters, or nodes by overriding the appropriate attributes when launching the container.
dgr will follow evolution of rkt. Since rkt will tend to support Open Container Initiative, dgr will support oci too. At some point, since rkt and docker will support oci, we will probably support building with and for both.
dgr is actively used at blablacar to build and run more than 300 different aci and pod to run all our platforms.
You can also have a look at examples that uses most of the features
dgr provides various resources to build and configure an ACI:
- scripts at runlevels (build, prestart...)
- templates and attributes
- static files
- images dependencies
Scripts are executed at the image build, before your container is started and more. See runlevels for more information.
Templates and attributes are the way dgr deals with environment-specific configurations. Templates are stored in the image and resolved at runtime ; attributes are inherited from different contexts (aci -> pod -> environment).
Static files are copied to the same path in the container.
Image dependencies are used as defined in APPC spec.
Download the executable from the releases page and put it somewhere. We currently only build for linux 64 bit.
$ dgr init # init a sample project
$ dgr build # build the image
$ dgr clean # clean the build
$ dgr clean build # just building, clean is always run before building
$ dgr clean install # clean, build and install aci in the local rkt
$ dgr clean push # clean, build and push aci to remote storage
$ dgr clean test # clean, build and test aci
$ dgr install # use already built aci in target directory to install in rkt
$ dgr push # use already built aci in target directory to push to remote storage
$ dgr test # run tests on already built aci
$ dgr try # run templating only to target/try (experimental)
$ dgr graph # generate graph image of .dot file of dependencies (app, builder and tester)
There is a lot of different flags on each command. use the helper to see them :
$ dgr --help
...
$ dgr build --help
...
Global configuration is optional to start as soon as you have rkt in $PATH
dgr global configuration is a yaml file located at ~/.config/dgr/config.yml
. Home is the home of starting user (the caller user if running with sudo).
targetWorkDir is used to indicate the target work directory where dgr will work to build and create the ACI push* contain informations on how to push the aci/pod to remote storage rkt if you are not using rkt in your path, or want to create specif config
Example of configuration:
targetWorkDir: /tmp/target # if you want to use another directory for all builds
rkt: # arguments to rkt. See rkt --help
path:
insecureOptions: [image]
dir: /var/lib/rkt
localConfig: /etc/rkt
systemConfig: /usr/lib/rkt
userConfig:
trustKeysFromHttps: false
noStore: false # can be set by command line
storeOnly: false # can be set by command line
Run the following commands to initialize a new complete sample project:
$ mkdir aci-myapp
$ cd aci-myapp
$ dgr init
It will generate the following file tree:
.
|-- attributes
| `-- attributes.yml # Attributes files that will be merged and used to resolve templates
|-- aci-manifest.yml # Manifest
|-- templates
| |-- etc
| | |-- templated.tmpl # template file that will end up at /etc/templated
| | `-- templated.tmpl.cfg # configuration of the targeted file, like user and mode (optional file)
| `-- header.partial # template part that can be included in template files
|-- files
| `-- dummy # Files to be copied to the same location in the target rootfs
|-- runlevels
| |-- builder
| | `-- 10.prepare.sh # Scripts to be run inside the builder to prepare the aci for build
| |-- build
| | `-- 10.install.sh # Scripts to be run when building inside aci's rootfs
| |-- build-late
| | `-- 10.build-late.sh # Scripts to be run when building inside aci's rootfs after the copy of files
| |-- inherit-build-early
| | `-- 10.inherit-build-early.sh # Scripts stored in ACI and executed while used as a dependency
| |-- inherit-build-late
| | `-- 10.inherit-build-late.sh # Scripts stored in ACI and executed while used as a dependency
| |-- prestart-early
| | `-- 10.prestart-early.sh # Scripts to be run when starting ACI before templating
| `-- prestart-late
| `-- 10.prestart-late.sh # Scripts to be run when starting ACI after templating
`-- tests
|-- dummy.bats # Bats tests for this ACI
`-- wait.sh # Script to wait until the service is up before running tests
This project is already valid which means that you can build it and it will result in a runnable ACI (dgr always adds busybox to the ACI). But you probably want to customize it at this point.
The only mandatory information is the aci-manifest.yml
, with only the aci name:
. You can remove everything else depending on your needs.
- builder runlevel with dependencies allow you build a project of any kind (java, php, go, node, ...) and release an aci without anything else than dgr and rkt on the host
- dgr will tell you if you are not using the latest version of a dependency and will tell you which version is the latest
- integrated test system that can be extended to support any kind of test system
- working with pods as a unit during build too
- build application version based on container name
- extract aci version from the version of the software during installation (templating of manifest)
dgr uses the builder information from the aci-manifest.yml to construct a rkt stage1. dgr then start rkt with this stage1 on an empty container with the final manifest of your aci (to have dependencies during build).
Inside rkt, the builder isolate the build process inside a systemd-nspawn on the builder's rootfs (with mount point on the final aci's rootfs and aci's home) and run the following steps :
- use internal dgr filesystem (busybox, openssl, wget, curl) for the builder if no dependencies (nothing in /usr/bin)
- run builder runlevel
- copy templater and inherit runlevels
- isolate on final rootfs and run build runlevels
- copy prestart, attributes, files, templates
- isolate on final rootfs and run build-late runlevels
To import dynamically environment vars to builder, you can use dgr --set-env=TOTO=TITI
. This is usefull to import proxy settings without explicitly knowing the proxy inside the build process.
Also all DGR_ENV_*
vars are automatically imported.
ex: having export DGR_ENV_no_proxy=127.0.0.1
on the host will end up having no_proxy=127.0.0.1
in builder.
The dgr manifest looks like a light ACI manifest with extra builder and tester info.
dgr will take the aci
part and convert it to the format defined in the APPC spec.
Example of a aci-manifest.yml:
name: example.com/myapp:0.1
builder:
dependencies:
- example.com/base:1
tester:
aci:
dependencies:
- example.com/base:1
aci:
dependencies:
- example.com/base:1
app:
exec:
- /bin/myapp
- -c
- /etc/myapp/myapp.cfg
mountPoints:
- name: myapp-data
path: /var/lib/myapp
readOnly: false
The name, well, is the name of the ACI you are building.
builder node is configuration of the filesystem you will use to build your ACI. By default, this filesystem only contain a busybox. When you set builder dependencies to handle specific build mechanism. (like archlinux or gentoo in the examples)
builder:
dependencies:
- example.com/aci-maven
- example.com/aci-java
mountPoints:
- {from: ../, to: /code}
- {from: ~/.m2, to: /root/.m2}
There is also a mountPoints
node to mount directories to the builder. This is usefull to have some external cache between builds (like .m2
maven directory) and also to trigger code build inside the builder, and release as an aci.
Here is an example of a builder script that can be used to do so.
#!/dgr/bin/busybox sh
set -e
source /dgr/bin/functions.sh
isLevelEnabled "debug" && set -x
mvn -f /code clean verify
cp /code/target/project.jar ${ROOTFS}/
mvn -f /code clean
Under the aci key, you can add every key that is defined in the APPC spec such as:
- exec which contains the absolute path to the executable your want to run at the start of the ACI and its args.
- mountPoints even though you can do it on the command line with recent versions of RKT.
- isolators...
Except handlers that are directly mapped to prestart runlevels
The scripts in runlevels/build
dir are executed during the build to install in the ACI everything you need. For instance if your dependencies are based on debian, a build script could look like:
#!/bin/bash
apt-get update
apt-get install -y myapp
You can create templates in your ACI. Templates are stored in the ACI as long as attributes and are resolved at start of the container.
Example:
templates/etc/resolv.conf.tmpl
You can also use templates/etc/resolv.tmpl.conf filename format to keep IDE language detection
{{ range .dns.nameservers -}}
nameserver {{ . }}
{{ end }}
{{ if .dns.search -}}
search {{ range .dns.search }} {{.}} {{end}}
{{end}}
templates/etc/resolv.conf.tmpl.cfg
uid: 0
gid: 0
mode: 0644
checkCmd: /dgr/bin/busybox true
checkCmd
is a command to run after the templating to check that the configuration is valid or fail container start.
When you have to reuse the same part in multiple templates, you can create a partial template like defined in the go templating.
templates/header.partial
{{define "header"}}
whatever
{{end}}
and include it in a template:
{{template "header" .}}
Templater provides functions to manipulate data inside the template. Here is the list:
Tables | Function | Description |
---|---|---|
base | path.Base | |
split | strings.Split | |
json | UnmarshalJsonObject | |
jsonArray | UnmarshalJsonArray | |
dir | path.Dir | |
getenv | os.Getenv | |
join | strings.Join | |
datetime | time.Now | |
toUpper | strings.ToUpper | |
toLower | strings.ToLower | |
contains | strings.Contains | |
replace | strings.Replace | |
orDef | orDef | if first element is nil, use second as default |
orDefs | orDefs | if first array param is empty use second element to fill it |
ifOrDef | ifOrDef | if first param is not nil, use second, else third |
add | add | add 2 numbers |
mul | mul | mutiply 2 numbers |
div | div | divide 2 numbers |
sub | sub | substract 2 numbers |
mod | mod | modulo of 2 numbers |
howDeep | HowDeep | deep level of an object in tree structure. usefull for indent |
isMapLast | isMapLast | is last element of a sorted keys map |
isMapFirst | isMapLast | is first element of a sorted keys map |
isString | isString | |
isArray | isArray | |
isMap | isMap | |
isKind | isKind | |
isType | isType |
It also provide all function defined by gtf project
We can add functions on demand
All the YAML files in the directory attributes are read by dgr. The first node of the YAML has to be "default" as it can be overridden in a POD or with a json in the env variable TEMPLATER_OVERRIDE in the cmd line.
attributes/resolv.conf.yml
default:
dns:
nameservers:
- "8.8.8.8"
- "8.8.4.4"
search:
- bla.com
myAttribute: "{{index .dns.nameservers 0}}" # example of templating in attributes
dgr uses the "pre-start" eventHandler of the ACI to customize the ACI rootfs before the run depending on the instance or the environment. It resolves at that time the templates so it has all the context needed to do that. You can also run custom scripts before (prestart-early) or after (prestart-late) this template resolution. This is useful if you want to initialize a mountpoint with some data before running your app for instance.
runlevels/prestart-late/init.sh
#!/bin/bash
set -e
/usr/bin/myapp-init
At this stage you should have a runnable aci. During build, dgr integrated into the aci a prestart that will take care of running templater using templates
and attributes
Templates and default attribute values are integrated into the aci.
At start you can change log level of prestart scripts and the templater with the environment variable --set-env=LOG_LEVEL=trace
.
default level is info. At debug
, prestart shell script will activate debug (set -x). At level trace
, templater will display the result of templating.
Default attributes values integrated in the aci can be overridden by adding a json tree in the environment variable TEMPLATER_OVERRIDE
# sudo rkt --set-env=LOG_LEVEL=trace --net=host --insecure-options=image run --interactive target/image.aci '--set-env=TEMPLATER_OVERRIDE={"dns":{"nameservers":["10.11.254.253","10.11.254.254"]}}'
dgr start by default with info log level. You can change this level with the -L
command line argument.
The log level is also propagated to all runlevels with the environment variable: LOG_LEVEL.
You can activate debug on demand by including this code in your scripts:
#!/dgr/bin/busybox sh
set -e
. /dgr/bin/functions.sh
isLevelEnabled "debug" && set -x
Build it
$ dgr -L debug build
You can also debug the start of your container (prestart, templates) the same way
$ rkt run --set-env=LOG_LEVEL=debug example.com/my-app
trace loglevel, will tell the templater to display the result
dgr is compatible with the appc push spec. Here is a example of how to test the push on local, without tls and signature
First you need an appc push spec compatible server. acserver is an official minimal implementation, but require aci signature. Here is a fork version where you can push non signed aci github.com/blablacar/acserver
Run the server :
$ mkdir /tmp/acserver
$ cd /tmp/acserver
$ wget https://github.com/blablacar/acserver/releases/download/0.1/acserver.tar.gz
$ tar xvzf acserver.tar.gz
$ rm acserver.tar.gz
$ sudo ./acserver -port 80 /tmp/acis admin password
2016/06/14 10:27:21 Listening on :80
Tell your system that aci.example.com is localhost :
/etc/hosts
...
127.0.0.1 aci.example.com
Tell dgr using rkt conf, how to access the server with authentication :
/etc/rkt/auth.d/aci.example.com.json
{
"rktKind": "auth",
"rktVersion": "v1",
"domains": ["aci.example.com"],
"type": "basic",
"credentials": {
"user": "admin",
"password": "password"
}
}
tell dgr that you do not support tls (and image signature) :
~/.config/dgr/config.yml
...
rkt:
insecureOptions: [http, image]
Init an aci that belong to aci.example.com
$ sudo dgr -W /tmp/aci-dummy init
...
Build and push the aci to your repository
$ sudo dgr -W /tmp/aci-dummy clean push
Run the aci fetching from repository
$ sudo rkt run --insecure-options http,image --no-store aci.example.com/aci-dummy:1
...
A pod is a group of aci that will build and run together as a single unit.
TODO
โโโ aci-elasticsearch # Directory that match the pod app shortname (or name)
โ โโโ attributes
โ โ โโโ attributes.yml # Attributes file for templating in this ACI
โ โโโ files # Files to be inserted into this ACI
โ ...
โโโ pod-manifest.yml # Pod Manifest
have a look at the examples/ directory where you can find aci for various distrib
Depending on distrib, package manager and what you want to do, you will not work the same way. but globally there is 2 way of building an aci.
This is what you will see everywhere else in docker or rkt. You use the build and build-late runlevels and run commands on the the final rootfs (like apt-get install...)
If you are using a package manager that support working outside of the target's rootfs or want to build a project, you will work outside of the stage1 directly inside the builder. For example if you are buiding an aci for a go project from sources. you will prepare a builder with go to be able to build the project on the stage1 and put the binary on the aci's rootfs (go is not needed to run the aci).
At this step, everybody can build any kind of project, since nothing on the host is used to build the project and the aci.
Also, if you are using a package manager like pacman
or emerge
, you can build and install packages on the final rootfs without build dependencies nor the package manager.
Most package manager are not design for overlay and are working with a db file for installed software. this means than when your aci have multiple dependencies on the aci, the db files will overlap and the package manager will only see half of package installed.
As far as I know only pacman
, that uses a file tree structure for install package, can support overlay.
If you are using a debian or similar. I recommand to limit the dependencies to only 2 layers. The base aci with debian minimal fs and one with the application you want.
A Dockerfile is purely configuration, describing the steps to build the container. It does not provide a common way of building containers across a team. It does not provide scripts levels, ending with very long bash scripting for the run option in the dockerfile. It does not handle configuration, nor at build time nor at runtime and does not support any kind of build outside of the container feature.
acbuild is a command line tools to build ACIs. It is more flexible than Dockerfiles as it can be wrapped by other tools such as Makefiles but like Dockerfiles it doesn't provide a standard way of configuring the images.
- rkt in your
$PATH
or configured in dgr global conf - being root is required to call rkt
- linux >= 3.18 with overlay filesystem
If you think your idea can be integrated directly in the core of dgr, please create an issue or a pull request.
If you want want to extend the way the builder is working (attributes, templates, files, ...), you can create a new stage1 builder and replace the internal one with :
...
builder:
image: blablacar.github.io/dgr/aci-builder:1
...
You can do the same for the tester.