The scenario of a minus task is a file named
scenario.yaml, located at the toplevel of the task directory.
The syntax of the file is YAML, a popular file format. The full YAML specification is quite large but only a small subset of YAML is needed for scenario files, and this subset is very intuitive, and easily understandable just by looking at a
scenario.yaml example. See for example the GNURadio benchmark example scenario. Otherwise, you can read the Grammar section below to get a more formal description of the scenario grammar.
Please bear in mind that YAML is indentation sensitive, just like python. YAML will follow whatever indentation system the scenario file uses, provided that it is consistent within the file. Note that in particular you should not mix spaces and tabs (beware, some text editors do that silently and need to be properly configured to avoid that). The suggested best practice is to use two spaces for indentation.
Comments in YAML start with a
The syntax of a yaml file may be checked with http://www.yamllint.com/
We're progressively adding new capabilities to the CorteXlab platform. There is now two different syntaxes for the scenario file, the legacy syntax, where commands are run on the nodes, and the new syntax, where docker containers are run on the nodes. Both syntaxes are allowed, even in the same scenario, but only one syntax is allowed per-node.
The scenario.yaml file is structured as follows:
- It MUST contain a
descriptionkey, value pair whose value is a string.
- It MUST contain a
durationkey, value pair, whose value is an integer, the max experiment duration expressed in seconds.
- It MUST contain a
nodeskey, whose value is a mapping (aka hash, or dictionnary). This mapping MUST contain:
- Nodes key, value pairs, whose keys are the node names (in the set
node40), and whose values are mappings:
- which MUST contain ONLY ONE of the following:
- (new syntax) a
containerkey, value pair, whose value is a mapping with the following key, value pairs:
imageMANDATORY a string, the name of the docker image, which must be accessible on a public docker repository (such as the docker hub). See docker images
commandOPTIONAL a string, the command to run in the container, which overrides the image default command (see https://docs.docker.com/engine/reference/run/#cmd-default-command-or-options https://docs.docker.com/engine/reference/builder/#cmd)
execOPTIONAL a string or array of strings: additionnal commands which are executed in the container (see https://docs.docker.com/engine/reference/commandline/exec/)
nameOPTIONAL a string for specifying the container's name. If not given, a default name will be automatically generated, with the minus task number and the container rank on the node.
- (old syntax) a
commandkey, value pair, whose value is the command line string which will be run on the corresponding node. Note that any filesystem path may be absolute or relative. If relative, it will be relative to the task top directory on the node. Note also that minus does NO magic with filesystem permissions, thus if a script or binary is to be executed, it needs to have executable permissions set before creating the task. The
commandmay also, if needed, contain a list of commands to be executed, instead of a single command. All the commands will be executed in parallel on the node. In this case, the syntax is multiline:
nodes: node1: command: - sleep 10000 - ./my_script
In the usual case where only one command is given, its stdout/stderr go to files
stderr.txt. In the case where more than one commands are given, their stdout/stderr go to files
stderr<X>.txt, where X is the command index (starting from zero).
- which MAY contain a
passivekey, value pair, whose value is a boolean (default is
false). A minus task will terminate when the first of the following conditions occur: its duration is exceeded or all its active nodes terminate.
An example scenario (new syntax):
description: Example of new syntax with docker containers duration: 120 nodes: node3: container: - image: m1mbert/cxlb-gnuradio-3.8:1.0 - image: m1mbert/cxlb-gnuradio-3.8:1.0 command: /usr/sbin/sshd -p 2223 -D node4: container: image: m1mbert/openbts-xenial:1.1 passive: true
Note that in this example, on node3, two containers are instanciated, based on the same image. The first container does not specify a
command, so the default command of the image is used, which is an ssh daemon listening on port 2222. The second container overrides the default command with an ssh daemon listening on port 2223 (if we try to instanciate two identical containers from this image, both ssh daemons will try to bind to port 2222 and the second one will fail)
An example scenario (old syntax):
description: Dummy scenario example duration: 60 nodes: node1: command: sleep 1000 # a bash command passive: true node2: command: ./myscript # relative path to a script embedded in the task node3: command: gnuradio-config-info -v # calling a gnuradio executable which is in the $PATH node4: command: /cortexlab/toolchains/current/share/gnuradio/examples/uhd/usrp_wfm_rcv.py # absolute path to a script or binary