Yet another computing scheduler & cloud orchestration engine

Yascheduler is a simple job scheduler designed for submitting scientific calculations and copying back the results from the computing clouds.

Currently it supports several scientific simulation codes in chemistry and solid state physics. Any other scientific simulation code can be supported via the declarative control template system (see yascheduler.conf settings file). There is an example dummy C++ code with its configuration template.

Installation

Use pip and PyPI: pip install yascheduler.

By default, no cloud connectors are installed. To install the appropriate connector, use one of the commands:

• for Microsoft Azure: pip install yascheduler[azure]
• for Hetzner Cloud: pip install yascheduler[hetzner]
• for UpCloud: pip install yascheduler[upcloud]

The last updates and bugfixes can be obtained cloning the repository:

git clone https://github.com/tilde-lab/yascheduler.git
pip install yascheduler/

The installation procedure creates the configuration file located at /etc/yascheduler/yascheduler.conf. The file contains credentials for Postgres database access, used directories, cloud providers and scientific simulation codes (called engines). Please check and amend this file with the correct credentials. The database and the system service should then be initialized with yainit script.

Usage

from yascheduler import Yascheduler

yac = Yascheduler()
label = "test assignment"
engine = "pcrystal"
struct_input = str(...)  # simulation control file: crystal structure
setup_input = str(...)  # simulation control file: main setup, can include struct_input
result = yac.queue_submit_task(
    label, {"fort.34": struct_input, "INPUT": setup_input}, engine
)
print(result)

Or run directly in console with yascheduler (use a key -l DEBUG to change the log level).

Supervisor config reads e.g.:

[program:scheduler]
command=/usr/local/bin/yascheduler
user=root
autostart=true
autorestart=true
stderr_logfile=/data/yascheduler.log
stdout_logfile=/data/yascheduler.log

File paths can be set using the environment variables:

• YASCHEDULER_CONF_PATH

  Configuration file.

  Default: /etc/yascheduler/yascheduler.conf

• YASCHEDULER_LOG_PATH

  Log file path.

  Default: /var/log/yascheduler.log

• YASCHEDULER_PID_PATH

  PID file.

  Default: /var/run/yascheduler.pid

Configuration File Reference

Database Configuration [db]

Connection to a PostgreSQL database.

• user

  The username to connect to the PostgreSQL server with.

• password

  The user password to connect to the server with. This parameter is optional

• host

  The hostname of the PostgreSQL server to connect with.

• port

  The TCP/IP port of the PostgreSQL server instance.

  Default: 5432

• database

  The name of the database instance to connect with.

  Default: Same as user

Local Settings [local]

• data_dir

  Path to root directory of local data files. Can be relative to the current working directory.

  Default: ./data (but it's always a good idea to set up explicitly!)

  Example: /srv/yadata

• tasks_dir

  Path to directory with tasks results.

  Default: tasks under data_dir

  Example: %(data_dir)s/tasks

• keys_dir

  Path to directory with SSH keys. Make sure it only contains the private keys.

  Default: keys under data_dir

  Example: %(data_dir)s/keys

• engines_dir

  Path to directory with engines repository.

  Default: engines under data_dir

  Example: %(data_dir)s/engines

• webhook_reqs_limit

  Maximum number of in-flight webhook http requests.

  Default: 5

• conn_machine_limit

  Maximum number of concurrent SSH connection's connect requests.

  Default: 10

• conn_machine_pending

  Maximum number of pending SSH connection's connect requests.

  Default: 10

• allocate_limit

  Maximum number of concurrent task or node allocation requests.

  Default: 20

• allocate_pending

  Maximum number of pending task or node allocation requests.

  Default: 1

• consume_limit

  Maximum number of concurrent task's results downloads.

  Default: 20

• consume_pending

  Maximum number of pending task's results downloads.

  Default: 1

• deallocate_limit

  Maximum number of concurrent node deallocation requests.

  Default: 5

• deallocate_pending

  Maximum number of pending node deallocation requests.

  Default: 1

Remote Settings [remote]

• data_dir

  Path to root directory of data files on remote node. Can be relative to the remote current working directory (usually $HOME).

  Default: ./data

  Example: /src/yadata

• tasks_dir

  Path to directory with tasks results on remote node.

  Default: tasks under data_dir

  Example: %(data_dir)s/tasks

• engines_dir

  Path to directory with engines on remote node.

  Default: engines under data_dir

  Example: %(data_dir)s/engines

• user

  Default ssh username.

  Default: root

• jump_user

  Username of default SSH jump host (if used).

• jump_host

  Host of default SSH jump host (if used).

Providers [clouds]

All cloud providers settings are set in the [clouds] group. Each provider has its own settings prefix.

These settings are common to all the providers:

• *_max_nodes

  The maximum number of nodes for a given provider. The provider is not used if the value is less than 1.

• *_user

  Per provider override of remote.user.

• *_priority

  Per provider priority of node allocation. Sorted in descending order, so the cloud with the highest value is the first.

• *_idle_tolerance

  Per provider idle tolerance (in seconds) for deallocation of nodes.

  Default: different for providers, starting from 120 seconds.

• *_jump_user

  Username of this cloud SSH jump host (if used).

• *_jump_host

  Host of this cloud SSH jump host (if used).

Hetzner

Settings prefix is hetzner.

• hetzner_token

  API token with Read & Write permissions for the project.

• hetzner_server_type

  Server type (size).

  Default: cx52

• hetzner_location

  Location name.

• hetzner_image_name

  Image name for new nodes.

  Default: debian-11

Azure

Azure Cloud should be pre-configured for yascheduler. See Cloud Providers.

Settings prefix is az.

• az_tenant_id

  Tenant ID of Azure Active Directory.

• az_client_id

  Application ID.

• az_client_secret

  Client Secret value from the Application Registration.

• az_subscription_id

  Subscription ID

• az_resource_group

  Resource Group name.

  Default: yascheduler-rg

• az_user

  SSH username. root is not supported.

• az_location

  Default location for resources.

  Default: westeurope

• az_vnet

  Virtual network name.

  Default: yascheduler-vnet

• az_subnet

  Subnet name.

  Default: yascheduler-subnet

• az_nsg

  Network security group name.

  Default: yascheduler-nsg

• az_vm_image

  OS image name.

  Default: Debian

• az_vm_size

  Machine size.

  Default: Standard_B1s

UpCloud

Settings prefix is upcloud.

• upcloud_login

  Username.

• upcloud_password

  Password.

Engines [engine.*]

Supported engines should be defined in the section(s) [engine.name]. The name is alphanumeric string to represent the real engine name. Once set, it cannot be changed later.

• platforms

  List of supported platform, separated by space or newline.

  Default: debian-10 Example: mY-cOoL-OS another-cool-os

• platform_packages

  A list of required packages, separated by space or newline, which will be installed by the system package manager.

  Default: [] Example: openmpi-bin wget

• deploy_local_files

  A list of filenames, separated by space or newline, which will be copied from local %(engines_dir)s/%(engine_name)s to remote %(engines_dir)s/%(engine_name)s. Conflicts with deploy_local_archive and deploy_remote_archive.

  Example: dummyengine

• deploy_local_archive

  A name of the local archive (.tar.gz) which will be copied from local %(engines_dir)s/%(engine_name)s to the remote machine and then unarchived to the %(engines_dir)s/%(engine_name)s. Conflicts with deploy_local_archive and deploy_remote_archive.

  Example: dummyengine.tar.gz

• deploy_remote_archive

  The url to the engine arhive (.tar.gz) which will be downloaded to the remote machine and then unarchived to the %(engines_dir)s/%(engine_name)s. Conflicts with deploy_local_archive and deploy_remote_archive.

  Example: https://example.org/dummyengine.tar.gz

• spawn

  This command is used by the scheduler to initiate calculations.

  cp {task_path}/INPUT OUTPUT && mpirun -np {ncpus} --allow-run-as-root \
    -wd {task_path} {engine_path}/Pcrystal >> OUTPUT 2>&1
  

  Example: {engine_path}/gulp < INPUT > OUTPUT

• check_pname

  Process name used to check that the task is still running. Conflicts with check_cmd.

  Example: dummyengine

• check_cmd

  Command used to check that the task is still running. Conflicts with check_pname. See also check_cmd_code.

  Example: ps ax -ocomm= | grep -q dummyengine

• check_cmd_code

  Expected exit code of command from check_cmd. If code matches than task is running.

  Default: 0

• sleep_interval

  Interval in seconds between the task checks. Set to a higher value if you are expecting long running jobs.

  Default: 10

• input_files

  A list of task input file names, separated by a space or new line, that will be copied to the remote directory of the task before it is started. The first input is considered as the main input.

  Example: INPUT sibling.file

• output_files

  A list of task output file names, separated by a space or new line, that will be copied from the remote directory of the task after it is finished.

  Example: INPUT OUTPUT

Aiida Integration

See the detailed instructions for the MPDS-AiiDA-CRYSTAL workflows as well as the ansible-mpds repository. In essence:

ssh aiidauser@localhost # important
reentry scan
verdi computer setup
verdi computer test $COMPUTER
verdi code setup

License