machinery

machinery is a command-line tool to operate on a whole cluster of Docker Machine virtual machines. machinery uses a YAML definition of the whole cluster to create machines, bring them up or down, or remove them at will. In short, machinery is to docker-machine what docker-compose is to docker. In addition, machinery provides Docker Swarm and Compose integration. It will automatically arrange for the created virtual machines to join the swarm cluster, generate the token as needed or even manage the life-cycle of several compose projects to be run on the cluster. machinery can automatically bring up specific project files onto machines that it controls. machinery is able to substitute the value of local environment variables in the compose project files before bringing the components up. Together with conventions for the dynamic construction of network-related environment variables, this provides for a simple mechanism for service discovery.

In short machinery provides you with an at-a-glance view of your whole cluster, from all the (virtual) machines that build it up, to all the components that should be run, and on which machine(s).

Quick Tour

machinery reads its default configuration from the file cluster.yml in the local directory. YAML definition files have a straightforward syntax. For example, the following content would define 3 machines using the virtualbox driver, one with more memory and ready to be duplicated through using the YAML anchoring facilities, another one with more disk than the defaults provided by docker-machine and the last one as the master of the cluster. The description also defines some labels that can be used by swarm to schedule components on specific nodes and arrange for the machine called core to have access to your home directory. Finally, it arranges for the components pinpointed by a relative compose project file to automatically be started up when db is brought up and created.

wk01: &worker
  driver: virtualbox
  memory: 2GiB
  labels:
    role: worker
db:
  driver: virtualbox
  size: 40G
  labels:
    role: db
  compose:
    -
      file: ../compose/backend/db.yml
core:
  driver: virtualbox
  master: on
  labels:
    role: core
  shares:
    - $HOME

Given access to a cluster definition file such as the one described above, the following command would create all the configured machines and arrange for a swarm token to be created when first executed.

machinery up

And the following command would gently bring the machine called db down and then destroy it.

machinery destroy db

If you had a YAML compose project description file called myapp.yml describing the components to run on your cluster, you could schedule it for execution by calling:

machinery swarm myapp.yml

Do you want to try for yourself at once? Jump to the bottom of this documentation and read and try the example section. You might want to download a "compiled" binary to avoid having to solve the few dependencies machinery has yourself.

Operating on the cluster

machinery takes a number of global options (dash-led) followed by a command. These commands can be followed by command-specific options and patterns matching the names of one or several virtual machines, as specified in the YAML cluster description. Those patterns are glob-style patterns, where * matches any number of characters, ? a single character and [ and ] can be used to specify ranges of characters. UNIX shells may be inclined to resolve these patterns for you, so you might have to quote your arguments.

Commands

up

The command up will bring up one or several machines, which names matching patterns would follow on the command line. If no machines are specified, machinery will bring up all machines in the cluster. Machines that do not exist yet will be created prior to being brought up, and a swarm token for the cluster will be generated if necessary during that process. More details are available in the section on the supported YAML format. Note that instead of up, machinery also supports the synonym start.

machinery is able to mount local host shares within the virtual machines that runs the virtualbox driver. Mounting is made persistent so that future runs of docker-machine start with the same machine as an argument will properly mount the shares. On all other drivers, machinery will use rsync when bringing up and down the machines to keep the directories synchronised.

In addition, machinery keeps a hidden environment file with the networking information for all the machines of the cluster (see below). This file will have the same root name as the cluster YAML file, but with a leading . to hide it and the extension .env (see env command description).

halt

The command halt will bring down one or several machines, which name matching patterns would follow on the command line. If no machines are specified, machinery will bring down all machines in the cluster. Note that instead of halt, machinery also supports the synonym stop. All directories that should be synchronised will be copied back to the host before their respective machines are brought down.

destroy

The command destroy will destroy entirely and irrevocably one or several machines, which name matching patterns would follow on the command line. If no machines are specified, machinery will destroy all machines in the cluster. Synchronised directories will be copied before machine destruction.

sync

The command sync will recursively copy the content of VM directories that are marked as shared back to the host. Calling machinery sync from a cron job is a good way to ensure data is copied between restarts of the machines.

swarm

The command swarm will either schedule components to be run in the cluster or print out its current status. When called without arguments, swarm will print out current cluster status i.e. the virtual machines that are registered within the master and their details.

Arguments to swarm should be one or several path to YAML files. machinery recognises automatically two kinds of YAML files:

• Compose projects files will be substituted for environment variables and sent to the master of the cluster. You will have to use labels or other scheduling techniques if you want to pinpoint specific machines where to run these components.

• machinery also recognises list of indirections to compose project files. These have exactly the same syntax as the compose keys of the regular YAML syntax.

swarm also takes a number of options that should appear before its arguments. These specify what docker-compose operations will be executed on the specified files. You can specify several options in a row, for example to restart components. The supported options are:

• -stop matches the stop command of compose and will stop the components.

• -kill matches the kill command of compose and will kill the components.

• -rm matches the rm command of compose and will remove the components without asking at the prompt.

• -start matches the start command of compose and will start stopped components.

• -up is the default when nothing is specified. It matches the up command of compose and will create and start the components in the background. Additional option can be provided through the YAML syntax of indirecting files.

• -options is a way to pass instructions to machinery when parsing regular compose YAML project files, as if these files had been pointed at by the indirection YAML format. The argument to -options should be a ,-separated string where each token should take the form of k=v and where k is the name of a YAML indirection directive, such as substitution or project and where v is the overriding value.

token

The command token will (re)generate a swarm token for cluster. Cluster tokens are cached in hidden files in the same directory as the YAML file that was used to describe the cluster. These files will have the same root name as the cluster YAML file, but with a leading . to hide them and the extension .tkn. machinery token will print out the token on the standard output, which eases further automation. If a token for the cluster can be found in the cache, it will be returned directly. To force regeneration of the token, you can specify the option -force to the command.

Whenever a token needs to be generated, machinery will run swarm create in a component on the local machine. The component is automatically removed once the token has been generated. However, the image that might have been downloaded will remain on the local machine.

env

The command env will output all necessary bash commands to declare the environment variables for network discovery. In other words, running eval $(machinery env) at the shell should set up a number of environment variables starting with the prefix MACHINERY_ and describing the network details of all existing machines in the cluster.

ssh

The command ssh will execute a command within a virtual machine of the cluster and print out its result on the standard output. The machine needs to be started for the command to succeed. When run with no additional arguments, an interactive command-line prompt will be provided.

ps

When called without any argument, the command ps will ask the swarm master to return the list of all running components, as of docker-machine ps. When called with arguments, these should be name matching patterns and the command will return the list of running components for that/those machine(s) instead. Note that calling ps with no argument, is different from running ps "*". In the first case, only the components scheduled via swarm are printed out. In the second case, all components currently running in the cluster are printed out, including those that would be issued from compose files attached specifically attached to machines.

version

Print out the current version of the program on the standard output and exit.

ls

Print out a static and dynamic summary of all the machines that are comprised in the cluster.

search

forall

Interaction with system components

Networking Information

machinery keeps a hidden environment file with the networking information for all the machines of the cluster. This file will have the same root name as the cluster YAML file, but with a leading . to hide it and the extension .env. The file defines a number of environment variables. Given the full name of a machine, e.g. mycluster-mymachine, a whole uppercase prefix will be constructed by prefixing MACHINERY_ to the name of the machine in uppercase, followed by another _. So, the previous example would lead to the prefix: MACHINERY_MYCLUSTER_MYMACHINE_ (note that the dash has been replaced by an _). This prefix will be prepend to the following strings:

• IP, the main IPv4 address of the virtual machine.

• And, for each relevant network interface name, e.g. if, at maximum two variables with the name of the interface in uppercase followed by a _ will be constructed: one for the IPv4 address (suffix INET) and one for the IPv6 address (suffix INET6), when they are present.

To wrap it up, and for the same machine name example as above, you might find the following in the environment file:

MACHINERY_MYCLUSTER_MYMACHINE_DOCKER0_INET=172.17.42.1
MACHINERY_MYCLUSTER_MYMACHINE_DOCKER0_INET6=fe80::5484:7aff:fefe:9799/64
MACHINERY_MYCLUSTER_MYMACHINE_ETH0_INET=10.0.2.15
MACHINERY_MYCLUSTER_MYMACHINE_ETH0_INET6=fe80::a00:27ff:fec2:ea1/64
MACHINERY_MYCLUSTER_MYMACHINE_ETH1_INET=192.168.99.111
MACHINERY_MYCLUSTER_MYMACHINE_ETH1_INET6=fe80::a00:27ff:fe58:c217/64
MACHINERY_MYCLUSTER_MYMACHINE_LO_INET=127.0.0.1
MACHINERY_MYCLUSTER_MYMACHINE_LO_INET6=::1/128
MACHINERY_MYCLUSTER_MYMACHINE_IP=192.168.99.111

Interaction with docker-machine

At a meta-level, machinery is simply a high-level interface to docker-machine. It will indeed call docker-machine for most of its under operations, for example when creating machines, stopping them, starting them, etc. The power of mcahinery is that it is able to operate on the whole cluster at one time, but also to provide a single summary point for all the machines of a specific cluster through the YAML definition.

By default, machinery will look for a file called cluster.yml in the current directory. This mimics the operation of related commands such as docker-compose. However, machinery is able to take a specific cluster definition file through its -cluster global option and its behaviour will slightly change in that case. Whenever started with the -cluster option pointing at another file than the default cluster.yml, machinery will create virtual machines which names uses the rootname (sans the directory path) of the cluster definition file as a prefix. Supposed you had started machinery with a YAML file called mycluster.yml and that it contained the definition for a machine called db, running machinery -cluster mycluster.yml create db would lead to the creation of a virtual machine called mycluster-db. Note that since the context is fully specified, you could mention a reference to that machine with the name db at the command-line. However, calling docker-machine ls will show up its real name, i.e. mycluster-db. This behaviour will help you managing several clusters from the same directory, for example when staging or when running sub-sets of your architecture for development.

Note, that when neither a specifiy YAML file is pointed at using -cluster, nor a default file called cluster.yml is found in the current directory, machinery will try to find a good candidate YAML description file automatically. It will list all files ending with .yml in the current directory and will consider those which first empty line contains the marker #docker-machinery. Whenever, only one candidate is found, this file will be taken into consideration as if it had been specifically pointed at using the -cluster global option.

Interaction with docker-compose

The combination of machinery, docker-machine and docker-compose enables a single control point for an entire cluster. This eases service discovery as it makes possible to pass information between components and machines of the cluster. machinery provides a solution to this through extending the YAML format for docker-compose. When substitution is turned on for a YAML compose project file, any occurrence of a local environment variable will be replaced by its value before being passed to docker-compose. The syntax also provides for default values, e.g. ${MYVAR:default} will be replaced by the content of the environment variable MYVAR if it existed, or by default if it did not exist. The implementation will perform local substitution into a temporary file that will be passed to docker-compose. In particular, all the environment variables described above will be available for substitution prior to docker-compose.

Authoring YAML files this way does not follow the official syntax, but you should still be able to pass your files to docker-compose after having fed them through envsubst.

Interaction with VBoxManage

machinery supports a number of extra features when creating virtualbox-based machines. This requires proper access to the VBoxManage command on the host machine. These features are port forwarding and the ability to mount host path into the guest machines. The mounting of shares is made persistent as through modifying the bootlocal.sh script of the boot2docker underlying image. However, mounting operations are performed the first time the virtual machines are started. Technically, the implementation generates unique names for the shares and declare those using VBoxManage sharedfolder add the first time they are needed. Then, each time the machine is started, a mount operation will be performed using an docker-machine ssh.

In practice, mounting of shares should be transparent to you, independantly to how you start the machine: using the start sub-command of machinery, starting manually using docker-machine start).

Global Options

machinery takes a number of global options before the command that it should execute. These options will affect its behaviour and are detailed below.

-help

Print some help on the standard output and exit. This is the same exact behaviour as calling machinery help.

-machine

This option should take as a value the path to your locally installed docker-machine binary. It defaults to docker-machine, which will attempt to find and run docker-machine using your PATH.

-docker

This option is similar to docker-machine and can be used to pinpoint a specific docker binary to run.

-token

This option takes a swarm token string as an argument and will override the token that would otherwise be generated and/or read from the (hidden) cache.

-cluster

This options takes the path to a YAML definition for your cluster. The directory specified using this option will be the directory where the cache for the swarm token is searched for and stored in. It defaults to the file cluster.yml in the current directory, when present, or the only file ending with .yml and containing #docker-machinery as its first non-empty line.

-driver

This option takes the name of a docker-machine driver as an argument. This is the driver that will be used when machines in the YAML definition do not have any specific driver, but also to generate the token, whenever this is relevant.

-cache

This option takes a boolean as an argument and will conduct the behaviour of machinery when it installs images on the machines that it creates. Boolean values can be on, false or 1, for example. When the cache is used, machinery will use the docker on the host to fetch the image and will then copy a tar of the image onto the machine where it will be unpacked and installed as a docker image. This has the advantage of cutting down download times in some situations, but also to arrange for login credentials to private registries not being present on the machines. Instead, only local access to the images on the host is necessary. When the cache is turned off, machinery will ask the docker daemon running on the machine to pull the images.

-ssh

When this option is set to an empty string, machinery will try to introspect the raw ssh command to log into the machine from the debug output of docker-machine ssh. This will in turn be used as a parameter to rsync (whenever relevant) or when copying files with scp (when docker-machine does not support the scp sub-command).

When this option is set to a string, this should contain the complete command to log into virtual machines using ssh. As a cluster contains more than one machine, the command can contain a number of %-framed keys that will be dynamically be replaced with virtual machine details whenever the ssh command is necessary. These keys are:

• %user% will contain the username into the machine, i.e. usually docker.
• %host% will contain the hostname of the machine (or its IP address).
• %identity% will contain the path to the RSA identity file for SSH security handshaking. This file is usually generated by docker-machine when the machine is created.
• %port% will contain the port number on the machine for SSH access, it does not need to be 22.

machinery is able to accomodate both -l username and username@host constructs within the ssh command that is specified as an argument to the -ssh option. The second construct will automatically be translated to the first (i.e. using the -l option to ssh whenever necessary). When it needs to construct scp commands, machinery is able to convert between the option-space of ssh and the one of scp, as these programs do not exactly take the same options.

-config

This option takes the path to a file as an argument. When present, the file will be read and its content will be used to initialised the value of the options that can otherwise be specified as part of the command-line global options. However, command-line options, if present, will always have precedence over the content of the file. In the file, any empty line will be ignored, any line starting with a # will be considered as a comment and ignored. Meaningful lines should contain the name of a global option (the leading dash can be omitted) followed by its value, separated by whitespaces. The value can be put between quotes " if necessary.

YAML Specification

Clusters are described using a YAML definition file. These files should contain a list of YAML dictionaries, where each key should be the name of a virtual machine within that cluster. Note that, as described before, machinery will prepend the rootname of the YAML file to the virtual machine in docker-machine in most cases. For each VM-specifying dictionary, machinery recognises a number of keys as described below:

driver

The value of driver should be the docker-machine driver to be used when creating the machine. If none is provided, the driver that is specified at the command-line under the option -driver will be used. This defaults to virtualbox as it is available on all platforms.

master

master should be a boolean and the value of master will specify if this machine is the swarm master. There can only be one swarm master per cluster.

cpu

cpu should be an integer and specifies the number of CPUs to allocate to that virtual machine. This will be automatically translated to each driver-specific option whenever possible. A warning will be issued at creation time if the driver does not support that option.

size

In its simplest form, size is an integer and specifies the size of the virtual disk for the virtual machine. This should be expressed in MB (see below).

To make it simpler, human-readable strings are also understood, e.g. 20G would specify a size of 20 gigabytes. machinery is able to make the difference between units expressed using the International System of Units (SI) and the binary-based units from the International Electrotechnical Commission (IEC). Thus 1GB is written according to SI metrics and is 1 000 000 000 bytes, while 1GiB is written according to IEC metrics and is 1 073 741 824 bytes. Unit specifications are case insensitive.

The option will be automatically translated to each driver-specific option whenever possible, possibly making the translation between MB and GB, or similar. A warning will be issued at creation time if the driver does not support that option.

memory

memory should specify the amount of memory for that virtual machine, it defaults to being expressed in MiB (see discussion above). This will be automatically translated to each driver-specific option whenever possible. A warning will be issued at creation time if the driver does not support that option.

labels

labels should be itself a dictionary. The content of this dictionary will be used as labels for the docker machines that are created. These labels can be used to schedule components on particular machines at a later time.

options

options should be itself a dictionary and contain a number of driver-specific options that will be blindly passed to the driver at machine creation. The leading double-dash that precedes these options can be omitted to keep the syntax simpler.

ports

ports should be a list of port forwarding specifications. A specification is either a single port or a host port separated from a guest port using a colon. When there is a single port, this port will be used as both the host and the guest port. It can also contain a trailing protocol specification following a slash, that defaults to tcp. So 8080:80 would forward the local host port 8080 onto the guest port 80. And 20514:514/udp would forward port 20514 onto the standard syslog port 514 on UDP.

At present, port forwarding is only meaningful and supported on virtualbox based machines.

shares

shares should be a list of share mounting specifications. A specification is a either a single path or a host path separated from a guest path using a colon. In addition, there might be a trailing type following another colon sign. Recognised types are vboxsf and rsync. When there is a single path (or the guest path is empty), this path will be used as both the host and the guest path. In paths, any occurrence of the name of an environment variable preceded with the $-sign will be replaced by the value of that local variable. For example, specifying $HOME would arrange for the path to your home directory to be available at the same location within the guest machine; handy whenever you want to transfer development files or initiate components. Relative path are resolved to the directory hosting the cluster definition YAML file.

The default type is vboxsf on virtualbox-based machines and rsync on top of all other drivers. When using the rsynctype, you can bring back changes that would have occured within the virtual machine onto the host using the command sync. When proper mounting is possible, the mounting will persist restarts of the virtual machine, e.g. when doing docker-machine restart or similar.

images

images should be a list of images to automatically pull from registries once a virtual machine has been created, initialised and verified. This can be handy if you want to make sure images are already present when your machine is being put into action. For example, docker-compose will sometimes timeout the first time that it schedules components as image downloading takes too long.

The default behaviour is to download images on the host and to transfer them to the virtual machine using a combination of docker save and docker load, i.e. as tar files. This has two benefits:

1. You can easily push images of private nature onto the virtual machines without exporting any credentials or similar.
2. If you have many machines using the same image, this can be quicker than downloading from a remote registry.

To switch off that behaviour and download the images from the virtual machines instead, set the global option -cache to off, 0 or false.

registries

registries should be a list of dictionaries specifying (private) registries at which to login upon machine creation. These dictionaries should contain values for the following keys: server, username, password and email, where server is the URL to the registry and the other fields are self-explanatory. machinery will log into all specified registries before attempting to pre-download images as explained above.

compose

compose should be a list of dictionaries that will, each, reference a docker-compose project file. Each dictionary must have a key called file that contains the path to the compose project file. A relative path will be understood as relative to the directory containing the YAML description file, thus allowing you to easily copy and/or transfer entire hierarchies of files.

Additionally, a key called options can be specified and it will contain a list of additional options that will be passed to docker-compose up. By default, all project files are brought up with the option -d to start their components in the background.

Another optional key called substitution can be set to a boolean. When true, the YAML compose file will be substituted for local environment variables before being parsed by docker-compose. See above for more information.

Finally a key called project can be set and is a string. It will contain the name of the compose project and will replace the one that usually is extracted from the directory name.

addendum

addendum should be a list of dictionary that will, each, reference a program or script to be run once a machine has been completely initialised.

These dictionaries should at least contain a key called exec, which content points to the program or script to run. A relative path will be understood as relative to the directory containing the YAML description file for the cluster.

Additionally, the content of the optional key args will be given as arguments when starting the program. Finally, if a key called substitution is present and set to a positive boolean, substitution will occur in the run script, as for compose files above.

Giving it a quick test

The directory test contains a test cluster with a single machine. Try for yourself by running the following command from the main directory of the repository.

./machinery -cluster test/test.yml up

You should see an output similar to the following one on the terminal. Actually, what you will see is a colourised output without timestamps. machinery automatically segregates terminals from regular file descriptor and the following was captured using a file redirection.

[20150414 204739] [NOTICE] Generating new token
[20150414 204739] [INFO] Detaching from vm...
[20150414 204739] [INFO] Creating swarm token...
[20150414 204740] [NOTICE] Created cluster token 87c9e52eb6be5d0c794afa7053462667
[20150414 204740] [INFO] Token for cluster definition at test/test.yml is 87c9e52eb6be5d0c794afa7053462667
[20150414 204740] [NOTICE] Creating machine test-test
[20150414 204741] [INFO]   Creating SSH key...
[20150414 204741] [INFO]   Creating VirtualBox VM...
[20150414 204743] [INFO]   Starting VirtualBox VM...
[20150414 204743] [INFO]   Waiting for VM to start...
[20150414 204829] [INFO]   Configuring Swarm...
[20150414 204849] [INFO]   "test-test" has been created and is now the active machine.
[20150414 204849] [INFO]   To point your Docker client at it, run this in your shell: $(docker-machine env test-test)
[20150414 204849] [INFO] SSH to test-test working properly
[20150414 204849] [NOTICE] Tagging test-test with role=testing target=dev
[20150414 204849] [NOTICE] Copying local /tmp/profile-11494-395 to test-test:/tmp/profile-11494-395
[20150414 204856] [INFO]   Waiting for VM to start...
[20150414 204928] [NOTICE] Port forwarding for test-test as follows: 8080->80/tcp 20514->514/udp 9090->9090/tcp
[20150414 204929] [NOTICE] Mounting shares as follows for test-test: /home/emmanuel->/home/emmanuel
[20150414 204929] [INFO] Getting info for guest test-test
[20150414 204929] [NOTICE] Waiting for test-test to shutdown...
[20150414 204934] [NOTICE] Bringing up machine test-test...
[20150414 204935] [INFO]   Waiting for VM to start...
[20150414 205007] [INFO] Attaching to test-test
[20150414 205012] [INFO] Docker setup properly on test-test
[20150414 205012] [NOTICE] Pulling images in test-test: gliderlabs/alpine
[20150414 205012] [INFO] Attaching to test-test
[20150414 205013] [INFO]   Pulling repository gliderlabs/alpine
[20150414 205015] [INFO]   a5b60fe97da5: Pulling image (latest) from gliderlabs/alpine
[20150414 205015] [INFO]   a5b60fe97da5: Pulling image (latest) from gliderlabs/alpine, endpoint: https://registry-1.docker.io/v1/
[20150414 205016] [INFO]   a5b60fe97da5: Pulling dependent layers
[20150414 205016] [INFO]   511136ea3c5a: Download complete
[20150414 205016] [INFO]   a5b60fe97da5: Pulling metadata
[20150414 205017] [INFO]   a5b60fe97da5: Pulling fs layer
[20150414 205019] [INFO]   a5b60fe97da5: Download complete
[20150414 205019] [INFO]   a5b60fe97da5: Download complete
[20150414 205019] [INFO]   Status: Downloaded newer image for gliderlabs/alpine:latest

To check around, you could run the following command to check that the machine test-test has really been created:

docker-machine ls

You could also jump into the created machine using the following command:

docker-machine ssh test-test

At the prompt, you can perhaps get a list of the docker components that have been started in the machine using the following command and verify that there are two running components: one swarm master component and one swarm agent.

docker ps

You can also check which images have been downloaded using the following command. That should list at least 3 images: one for swarm, one for busybox (which is used to verify that docker runs properly at the end of the machine creation process) and finally one for Alpine Linux, which is downloaded as part of the test cluster definition file.

docker images

Finally, you can check that you can access your home directory at its usual place, as it is automatically mounted as part of the test cluster definition. A final note: jumping into the machine was not a necessary process, you would have been able to execute thos commands directly from the host command prompt after having run $(docker-machine env test-test).

Once done, return to the host prompt and run the following to clean everything up:

./machinery -cluster test/test.yml destroy

Comparison to Other Tools

machinery is closely related to Vagrant, and it evens provides a similar set of commands. However, being built on top of docker-machine provides access to many more providers through all the existing Docker Machine drivers.

Implementation

machinery is written in Tcl. It requires a recent version of Tcl (8.6 at least) and the yaml library to be able to parse YAML description files. As the yaml library is part of the standard tcllib, the easiest is usually to install the whole library using your package manager. For example, on ubuntu, running the following will suffice as Tcl is part of the core server and desktop installation.

apt-get install tcllib