Yet another computing scheduler and cloud orchestration engine
Project description
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
.
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)
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
Example:
/srv/yadata
-
tasks_dir
Path to directory with tasks results.
Default:
tasks
underdata_dir
Example:
%(data_dir)s/tasks
-
keys_dir
Path to directory with SSH keys.
Default:
keys
underdata_dir
Example:
%(data_dir)s/keys
-
engines_dir
Path to directory with engines repository.
Default:
engines
underdata_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
underdata_dir
Example:
%(data_dir)s/tasks
-
engines_dir
Path to directory with engines on remote node.
Default:
engines
underdata_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 [cloud]
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:
cx51
-
hetzner_image_name
Image name for new nodes.
Default:
debian-10
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 withdeploy_local_archive
anddeploy_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 withdeploy_local_archive
anddeploy_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 withdeploy_local_archive
anddeploy_remote_archive
.Example:
https://example.org/dummyengine.tar.gz
Example:
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 alsocheck_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
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for yascheduler-1.2.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2f71c61e96e1c670a85f6701d1dd0c7cf98599a51650ed4af4df93e5d3bd1a58 |
|
MD5 | b661586ef0040c923d8909767555c566 |
|
BLAKE2b-256 | ce1d8be161fa909c5f81282cc83c7c18e7030bcc9eb6e057ec7992b49b4770d1 |