New start, status, stop scripts
Redesign of the startup, stop scripts.
This script serves as an easy way to launch the GekkoFS daemon on multiple nodes via srun.
Script will feature the following functionality:
-
A configuration file allowing to set various environment variables and more (e.g., log level or log locations) -
A gkfs
shell script which has the following functions:-
start -
status -
stop
-
Script optional arguments
Most commonly used arguments are used to overwrite default values of config file. Values in <
>
are default values. (NOT FINAL):
-
[-h, --help]
help page -
[-r/--rootdir <config>]
rootdir path -
[-m/--mountdir <config>]
mountdir path -
[--srun <true>]
should srun be used to launch on multiple nodes? -
[-n/--numnodes <jobsize>]
number of nodes. If not given, reads srun environment.--srun
must be true. -
[-a/--auto-sm <false>]
toggle Mercury'sauto-sm
-
[-l/--listen <ib0>]
used network device -
[--cpuspertask <64>]
number of allowed cpus used by Slurm task. (This can severely impact daemon performance) -
[-v/--verbose <false>]
increase verbosity level of script -
[-p/--protocol <config>]
used network protocol -
[-c/--config <./gkfs_runtime.conf>
config file location if not in the same directory
Script positional arguments
-
start: TODO requirements -
run in foreground -
run in background -
run several instances on one node
-
-
status: TODO requirements no longer needed. can be added later -
stop: TODO requirements -
stop slurm task gracefully
-
Configuration file
The current configuration file contents are hosted here and can be collaboratively modified to meet our requirements: https://hedgedoc.lefthy.de/s/gkfs_config
Current issues
- Script takes CPUs per process from script launching node and not the deployment node
Edited by Marc Vef