Last modified 10 years ago

PSNC DRMAA for SLURM

  1. PSNC DRMAA for SLURM
    1. Introduction
    2. Download
    3. SVN access
    4. Installation
    5. Configuration
      1. Configuration file syntax
    6. Native specification
    7. Changelog
      1. Known bugs and limitations
      2. Authors
      3. Contact
      4. Developer tools
      5. Links
      6. License

Introduction

PSNC DRMAA for  Simple Linux Utility for Resource Management (SLURM) is an implementation of  Open Grid Forum  DRMAA 1.0 (Distributed Resource Management Application API)  specification for submission and control of jobs to SLURM. Using DRMAA, grid applications builders, portal developers and ISVs can use the same high-level API to link their software with different cluster/resource management systems.

This software also enables the integration of  QCG-Computing with the underlying SLURM system for remote multi-user job submission and control over Web Services.

Download

DRMAA for SLURM is distributed as a source package which can be downloaded via the Downloads section.

SVN access

  $ svn co https://apps.man.poznan.pl/svn/slurm-drmaa/

Please note the ./autogen.sh and ./autoclean.sh scripts which calls the autotools command chain in appropriate order.

note: You need some developer tools to compile the svn version. Also the trunk version may not always compile.

Installation

To compile and install the library just go to main source directory and type: 

$ ./configure [options] && make
$ sudo make install

The library was tested with Simple Linux Utility for Resource Management versions 2.1.13, 2.2 and 2.3.3. If you encountered any problems using the library on the different systems, please use the contact e-mails for reporting the problem.

Notable ./configure script options:

--with-slurm-inc SLURM_INCLUDE_PATH

Path to SLURM header files (i.e. directory containing slurm/slurm.h ). By default the library tries to guess the SLURM_INCLUDE_PATH and SLURM_LIBRARY_PATH based on location of the srun executable.

--with-slurm-lib SLURM_LIBRARY_PATH

Path to SLURM libraries (i.e. directory containing libslurm.a ).

--prefix INSTALLATION_DIRECTORY

Root directory where PSNC DRMAA for SLURM shall be installed. When not given library is installed in /usr/local.

--enable-debug

Compiles library with debugging enabled (with debugging symbols not stripped, without optimizations, and with many log messages enabled). Useful when you are to debug DRMAA enabled application or investigate problems with DRMAA library itself.

There are no unusual requirements for basic usage of library: ANSI C compiler and standard make program should suffice. If you have taken sources directly from SVN repository or wish to run test-suite you would need additional developer tools.

Configuration

During DRMAA session initialization (drmaa_init) library tries to read its configuration parameters from locations: /etc/slurm_drmaa.conf, ~/.slurm_drmaa.conf and from file given in SLURM_DRMAA_CONF environment variable (if set to non-empty string). If multiple configuration sources are present then all configurations are merged with values from user-defined files taking precedence (in following order: $SLURM_DRMAA_CONF, ~/.slurm_drmaa.conf, /etc/slurm_drmaa.conf).

Currently recognized configuration parameters are:

cache_job_state
According to DRMAA specification every drmaa_job_ps() call should query DRM system for job state. With this option one may optimize communication with DRM. If set to positive integer drmaa_job_ps() returns remembered job state without communicating with DRM for cache_job_state seconds since last update. By default library conforms to the specification (no caching will be performed).

Type: integer, default: 0

job_categories
Dictionary of job categories. Its keys are job categories names mapped to native specification strings. Attributes set by job category can be overridden by corresponding DRMAA attributes or native specification. Special category name default is used when drmaa_job_category job attribute was not set.

Type: dictionary with string values, default: empty dictionary

Configuration file syntax

Configuration file is in a form of a dictionary. Dictionary is set of zero or more key-value pairs. Key is a string while value could be a string, an integer or another dictionary. 

  configuration: dictionary | dictionary_body
  dictionary: '{' dictionary_body '}'
  dictionary_body: (string ':' value ',')*
  value: integer | string | dictionary
  string: unquoted-string | single-quoted-string | double-quoted-string
  unquoted-string: [^ \t\n\r:,0-9][^ \t\n\r:,]*
  single-quoted-string: '[^']*'
  double-quoted-string: "[^"]*"
  integer: [0-9]+

Native specification

DRMAA interface allows to pass DRM dependent job submission options. Those options may be specified directly by setting drmaa_native_specification job template attribute or indirectly by the drmaa_job_category job template attribute. The legal format of the native options looks like: 

  -A My_job_name -s -N 1=10

List of parameters that can be passed in the drmaa_native_specification attribute:

Native specification Description
-A, --account=\name Charge job to specified accounts
--acctg-freq Define the job accounting sampling interval
--comment An arbitrary comment
-C, --constraint=\list Specify a list of constraints
--contiguous If set, then the allocated nodes must form a contiguous set
--exclusive Allocate nodenumber of tasks to invoke on each nodes in exclusive mode when cpu consumable resource is enabled
--mem=\MB Minimum amount of real memory
--mem-per-cpu=\MB Maximum amount of real memory per allocated cpu required by a job
--mincpus=\n Minimum number of logical processors (threads) per node
-N, --nodes=\N Number of nodes on which to run (N = min[-max])
--ntasks-per-node=\n Number of tasks to invoke on each node
-p, --partition=\partition Partition requested
--qos=\qos Quality of Serice
--requeue If set, permit the job to be requeued
--reservation=\name Allocate resources from named reservation
-s, --share Job allocation can share nodes with other running jobs
-w, --nodelist=\hosts Request a specific list of hosts
-t, --time=\hours:minutes Set a maximum job wallclock time
-n, --ntasks=\n Number of tasks
--gres Specifies a comma delimited list of generic consumable resources
--no-kill Do not automatically terminate a job of one of the nodes it has been allocated fails
--licenses Specification of licenses
--mail-type Notify user by email when certain event types occur. Valid type values are BEGIN, END, FAIL, REQUEUE, and ALL (any state change)
--no-requeue Specifies that the batch job should not be requeued after node failure
-x, --exclude Explicitly exclude certain nodes from the resources granted to the job
--tmp Specify a minimum amount of temporary disk space

Description of each parameter can be found in man sbatch.

Changelog

  • 1.0.7 - user supplied (via DRMAA attribute) native specification now takes precedence over the native specification provided in configuration file.
  • 1.0.6 - added support for --gres, --no-kill, --licenses, --mail-type=, --no-requeue, --exclude, --tmp in native specification attribute. Implemented handling of missing jobs.
  • 1.0.5 - better handling of --time (-t) (thanks to Roman Valls Guimera) and added support for --ntasks (-n) in native specification attribute. Fixed DRMAA_V_EMAIL attribute handling
  • 1.0.4 - support for SLURM 2.3
  • 1.0.3 - the --time native option support
  • 1.0.2 - environment variables are now propagated from submission host to the worker nodes
  • 1.0.1 - added support for SLURM 2.2
  • 1.0.0 - first public release

Known bugs and limitations

Library covers all  DRMAA 1.0 specification with exceptions listed below. It was successfully tested with  Simple Linux Utility for Resource Management (SLURM) 2.1.13, 2.2 and 2.3.3. Known limitations:

  • drmaa_control options DRMAA_CONTROL_HOLD, DRMAA_CONTROL_RELEASE are only available for users being SLURM administrators (in version prior 2.2)
  • drmaa_control options DRMAA_CONTROL_SUSPEND, DRMAA_CONTROL_RESUME are only available for users being SLURM administrators
  • drmaa_wct_slimit not implemented
  • optional attributes drmaa_deadline_time, drmaa_duration_hlimit, drmaa_duration_slimit, drmaa_transfer_files not implemented
  • The SPANK client side (i.e. not remote) plugins chain is not invoked in DRMAA run job call. For this reason we advice you to use  TASK BLOCKS in the UseEnv SPANK plugin.

Authors

The library was developed by:

  • Michal Matloka - first implementation
  • Mariusz Mamonski - maintainer since version 1.0.3

This library relies heavily on the Fedstage DRMAA utils code developed by:

  • Lukasz Ciesnik.

Contact

In case of any problems or questions regarding the DRMAA for SLURM do not hesitate to contact us:

  • QosCosGrid Development Team - qcg(at)plgrid.pl

Developer tools

Although not needed for library user the following tools may be required if you intend to develop PSNC DRMAA for SLURM:

  • GNU autotools
    • autoconf (tested with version 2.67)
    • automake (tested with version 1.11)
    • libtool (tested with version 2.2.8)
    • m4 (tested with version 1.4.14)
  •  Bison parser generator,
  •  RAGEL State Machine Compiler,
  •  gperf gperf - a perfect hash function generator.

DRMAA:  http://www.drmaa.org/
Open Grid Forum:  http://www.gridforum.org/
DRMAA 1.0 specification:  http://www.ogf.org/documents/GFD.133.pdf
Official DRMAA test-suite:  http://drmaa.org/testsuite.php
Smoa Computing:  http://apps.man.poznan.pl/trac/smoa-comp
Simple Linux Utility for Resource Management (SLURM):  https://computing.llnl.gov/linux/slurm/
Bison:  http://www.gnu.org/software/bison/

License

Copyright (C) 2011 Poznan Supercomputing and Networking Center

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see  http://www.gnu.org/licenses/.