experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Clarify that a job step/srun is required

Browse Source
This commit is contained in:
Alexander Grund 2020-04-08 13:59:49 +02:00
parent f93710d728
commit 961e341274
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/python/distribute/cluster_resolver/README_Slurm.md
View File

@ -7,7 +7,8 @@ The resolution is done by determining job configuration through a number of Slur
By default everything is determined from the Slurm environment, hence for most uses case no manual setting of parameters is required.
## How it works
Slurm sets a number of variables for each job and job step.
`SlurmClusterResolver` reads the environment variables that are set inside a job step launched by Slurm.
This means it will only work correctly for applications launched via `srun`.
The process ID/rank is extracted from environment variable `SLURM_PROCID` and the total number of tasks launched is extracted from `SLURM_STEP_NUM_TASKS`.
The hostnames are resolved by inspection `SLURM_STEP_NODELIST`.