| 1 | package HPCD::SLURM; | ||
| 2 | ### HPCD::SLURM.pm ######################################################################### | ||
| 3 | |||
| 4 | =head1 NAME | ||
| 5 | |||
| 6 | HPCD::SLURM | ||
| 7 | |||
| 8 | =head1 VERSION | ||
| 9 | |||
| 10 | Version 0.01 | ||
| 11 | |||
| 12 | =cut | ||
| 13 | |||
| 14 | our $VERSION = '0.01'; | ||
| 15 | |||
| 16 | =head1 SYNOPSIS | ||
| 17 | |||
| 18 | use HPCI; | ||
| 19 | |||
| 20 | my $group = HPCI->group(cluster => 'SLURM', ...); | ||
| 21 | |||
| 22 | =head1 DESCRIPTION | ||
| 23 | |||
| 24 | This is a driver module that can be used internally from HPCI, by | ||
| 25 | specifying a cluster type of 'SLURM'. It customizes the HPCI access | ||
| 26 | interface to control jobs submitted to an SLURM (Simple Linux Utility | ||
| 27 | for Resource Management) cluster. | ||
| 28 | |||
| 29 | The differences provided by the SLURM driver are: | ||
| 30 | |||
| 31 | =over | ||
| 32 | |||
| 33 | =item - | ||
| 34 | |||
| 35 | the following resource names have slurm-specific aliases | ||
| 36 | |||
| 37 | =over | ||
| 38 | |||
| 39 | =item h_time | ||
| 40 | time | ||
| 41 | |||
| 42 | =back | ||
| 43 | |||
| 44 | Using these aliases causes your code to be SLURM-specific. That makes | ||
| 45 | converting it to use an alternate type of cluster more painful. However, | ||
| 46 | if you do not expect to ever be moving your code to a different cluster | ||
| 47 | type, this might simplify the interface for people who already know the | ||
| 48 | SLURM-specific resource names and wish to avoid the confusion that might | ||
| 49 | come from using the generic names. | ||
| 50 | |||
| 51 | =item - | ||
| 52 | |||
| 53 | a stage attribute B<native_args_string> is available which specifies a string of | ||
| 54 | argument(s) that will be included in the srun/sbatch command that starts execution | ||
| 55 | of that stage | ||
| 56 | |||
| 57 | =item - | ||
| 58 | |||
| 59 | The result information hash returned by C<$stats = $group-E<gt>execute> | ||
| 60 | includes all of the info provided by the SLURM sacct command, not just | ||
| 61 | the exit status. | ||
| 62 | |||
| 63 | =item - | ||
| 64 | |||
| 65 | If a stage is terminated because it exceeds the requested memory resource | ||
| 66 | limit, then it will normally be retried using a higher limit. | ||
| 67 | The default is to first try with '2G' (if no mem resource limit | ||
| 68 | is explicitly specified). After mem failures, the next larger size in the | ||
| 69 | default sequence (2G 4G 8G 16G 32G) is attempted. | ||
| 70 | |||
| 71 | =back | ||
| 72 | |||
| 73 | =head1 SEE ALSO | ||
| 74 | |||
| 75 | =over | ||
| 76 | |||
| 77 | =item HPCI | ||
| 78 | |||
| 79 | Describes the generic HPCI iterface - this manual is only providing | ||
| 80 | the exceptions to that document (and its related documents). | ||
| 81 | |||
| 82 | =item HPCI::Group | ||
| 83 | |||
| 84 | Describes the interface common to all B<HPCI Group> | ||
| 85 | objects, regardless of the particular type of cluster that | ||
| 86 | is actually being used to run the stages. | ||
| 87 | |||
| 88 | =item HPCI::Stage | ||
| 89 | |||
| 90 | Describes the interface common to stage object returned | ||
| 91 | by all B<HPCI Stage> objects, regardless of the | ||
| 92 | particular type of cluster that is actually being used to | ||
| 93 | run the stages. | ||
| 94 | |||
| 95 | =back | ||
| 96 | |||
| 97 | =head1 AUTHOR | ||
| 98 | |||
| 99 | Christopher Lalansingh - Boutros Lab | ||
| 100 | |||
| 101 | John Macdonald - Boutros Lab | ||
| 102 | |||
| 103 | Anqi (Joyce) Yang - Boutros Lab | ||
| 104 | |||
| 105 | =head1 ACKNOWLEDGEMENTS | ||
| 106 | |||
| 107 | Paul Boutros, Phd, PI - Boutros Lab | ||
| 108 | |||
| 109 | The Ontario Institute for Cancer Research | ||
| 110 | |||
| 111 | =cut | ||
| 112 | |||
| 113 | 1; | ||
| 114 |