Getting Started#
This tutorial is a starter’s guide to run a dummy experiment with Autosubmit.
Dummy experiments run workflows with inexpensive empty tasks and therefore are ideal for teaching and testing purposes.
Real experiments instead run workflows with complex tasks. To read information about how to develop parameterizable tasks for Autosubmit workflows, refer to Developing a project.
Pre-requisites#
Autosubmit needs to establish password-less SSH connections in order to run and monitor workflows on remote platforms.
Ensure that you have a password-less connection to all platforms you want to use in your experiment. If you are unsure how to do this, please follow these instructions:
Open a terminal and prompt
ssh-keygen -t rsa -b 4096 -C "email@email.com" -m PEM
Copy the resulting key to your platform of choice. Via SCP or ssh-copy-key.
# Generate a key pair for password-less ssh, PEM format is recommended as others can cause problems
ssh-keygen -t rsa -b 4096 -C "email@email.com" -m PEM
# Copy the public key to the remote machine
ssh-copy-id -i ~/.ssh/id_rsa.pub user@remotehost
# Add your key to ssh agent ( if encrypted )
# If not initialized, initialize it
eval `ssh-agent -s`
# Add the key
ssh-add ~/.ssh/id_rsa
# Where ~/.ssh/id_rsa is the path to your private key
Description of most used commands#
Command |
Short description |
---|---|
expid |
Creates a new experiment and generates a new entry in the database by giving it a serial id composed of 4 letters. In addition, it also creates the folder experiment and the basic folder structure. |
create <expid> |
Generates the experiment workflow. |
run <expid> |
Runs the experiment workflow. |
monitor <expid> |
Shows the experiment workflow structure and status. |
inspect <expid> |
Generates Autosubmit scripts and batch scripts for inspection, by processing the tasks’ templates with the experiment parameters. |
refresh <expid> |
Updates the project directory. |
recovery <expid> |
Recovers the experiment workflow obtaining the last run complete jobs. |
setstatus <expid> |
Sets one or multiple jobs status to a given value. |
Create a new experiment#
autosubmit expid -dm -H "local" -d "Tutorial"
-dm: Generates a dummy experiment.
-H: Sets the principal experiment platform.
-d: Sets a short description for the experiment.
The output of the command will show the expid of the experiment and generate the following directory structure:
Experiment folder |
Contains |
---|---|
conf |
Experiment configuration files. |
pkl |
Workflow pkl files. |
plot |
Visualization output files |
tmp |
Logs, templates and misc files. |
proj |
User scripts and project code. (empty) |
Then, execute autosubmit create <expid> -np
and Autosubmit will generate the workflow graph.
Run and monitoring#
To run an experiment, use
`autosubmit run <expid>`
. Autosubmit runs experiments performing the following operations:
First, it checks the experiment configuration. If it is wrong, it won’t proceed further.
Second, it runs the experiment while retrieving all logs from completed or failed tasks as they run.
Third, it manages all the workflow steps by following the dependencies defined by the user until all jobs are in COMPLETED or FAILED status. There can be jobs left in WAITING status if their dependencies are in FAILED status.
While the experiment is running, it can be visualized via autosubmit monitor <expid>
.

illustrates the output of the autosubmit monitor. It describes all workflow jobs’ possible status and actual status.
Concurrently, the <expid>/tmp
gets filled with the cmd scripts generated by Autosubmit to run the local and remote tasks (in this case, they are sent and submitted to the remote platform(s)).
Autosubmit keeps logs at ASLOGS
and LOG_a000
folders, which are filled up with Autosubmit’s command logs and job logs.
Viewing the logs#
The autosubmit
commands such as expid
, run
, monitor
, all may produce
log files on the user’s file system. To save the user from having to navigate to the
log file, or to memorize the location of these files, Autosubmit provides the
autosubmit cat-log
command.
$ autosubmit cat-log a000
Autosubmit is running with 4.0.0b
2023-02-27 21:45:47,863 Autosubmit is running with 4.0.0b
2023-02-27 21:45:47,872
Checking configuration files...
2023-02-27 21:45:47,900 expdef_a000.yml OK
2023-02-27 21:45:47,904 platforms_a000.yml OK
2023-02-27 21:45:47,905 jobs_a000.yml OK
2023-02-27 21:45:47,906 autosubmit_a000.yml OK
2023-02-27 21:45:47,907 Configuration files OK
Note
The -f
(--file
) option is for the file type, not the file path.
See the complete help and syntax with autosubmit cat-log --help
for
a list of supported types, depending on whether you choose a workflow
log or a job log file. Note too that there is a -i
(--inspect
)
flag in the command to tell Autosubmit you want job files generated by
autosubmit inspect
, instead of job files generated by autosubmit run
.
Configuration summary#
In the folder
<expid>/conf
there are different files that define the actual experiment configuration.
File |
Content |
---|---|
|
|
|
|
|
|
|
|
|
|
Final step: Modify and run#
It is time to look into the configuration files of the dummy experiment and modify them with a remote platform to run a workflow with a few more chunks.
Open expdef.yml
DEFAULT:
# Don't change
EXPID: "a000"
# Change for your new main platform name, ej. marenostrum4
HPCARCH: "local"
# Locate and change these parameters, per ej. numchunks: 3
EXPERIMENT:
DATELIST: 20000101
MEMBERS: fc0
NUMCHUNKS: 1
(...)
Now open platforms.yml. Note: This will be an example for marenostrum4
PLATFORMS:
marenostrum4:
# Queue type. Options: ps, SGE, SLURM, PBS, eceaccess
# scheduler type
TYPE: slurm
HOST: mn1.bsc.es,mn2.bsc.es,mn3.bsc.es
# your project
PROJECT: bsc32
# <- your user
USER: bsc32070
SCRATCH_DIR: /gpfs/scratch
ADD_PROJECT_TO_HOST: False
# use 72:00 if you are using a PRACE account, 48:00 for the bsc account
MAX_WALLCLOCK: 02:00
# use 19200 if you are using a PRACE account, 2400 for the bsc account
MAX_PROCESSORS: 2400
PROCESSORS_PER_NODE: 48
SERIAL_QUEUE: debug
QUEUE: debug
autosubmit create <expid>** (without -np)
will generate the new workflow and autosubmit run <expid>
will run the experiment with the latest changes.
Warning
If you are using an encrypted key, you will need to add it to the ssh-agent before running the experiment. To do so, run ssh-add <path_to_key>
.