Build, deploy and run XDMoD web as a Singularity/Apptainer container. Also demonstrate the automated XDMoD database ingestion from the accounting data which is periodically backup to the repository.
sudo singularity build xdmod_srv.sif ./def_xdmod_srv.def
overlay image is needed for storing the MySQL database (at least 100GB):
dd if=/dev/zero of=overlay.img bs=1M count=100000 && mkfs.ext3 overlay.img
sudo singularity sif add --datatype 4 --partfs 2 --parttype 4 --partarch 2 --groupid 1 xdmod_srv.sif overlay.img
Start the Singularity instance:
sudo singularity instance start -B /YOUE_HOME/SIF_XDMoD/init_load_data:/init_load_data --writable xdmod_srv.sif xdmod_web
sudo singularity run instance://xdmod_web
If the service needs to be shutdown, it must be done gracefully:
sudo singularity instance stop xdmod_web
The init_load_data directory contains SGE accounting data in tar.gz format for bunk loading. This data is not included in this repository, and should be prepared manually from the sge_accounting repository for the first deployment. Note down the commit hash when preping the tar.gz data, this hash will be come the first FROM_COMMIT_HASH used by automated new data upload.
# Example output of the init load task:
TASK [init_loader : Load data to XDMoD] *********************************************************************************************************************************************************
changed: [localhost] => (item=/init_load_data/CLUSTER60_accounting_2017_FIXED.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER62_accounting_ALL.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER60_accounting_2012-2014.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER60_accounting_2015.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER60_accounting_2016.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER62_accounting_ALL.tar.gz)
changed: [localhost] => (item=/init_load_data/CLUSTER70_accounting_init_load.tar.gz)
changed: [localhost] => (item=/init_load_data/Grandline70_accounting_init_load.tar.gz)
changed: [localhost] => (item=/init_load_data/Wolfpack70_accounting_init_load.tar.gz)
This process is implemented in Ansible:
cd /SIF_XDMoD
ansible-playbook init_loader.yml
Loading data to XDMoD is slow, it will take one or a few days depends the amount of data. After it's done, it brings the XDMoD database up to the commit FROM_COMMIT_HASH in the sge_accounting repository.
Clone the sge_accounting repository to the container so it can be used for automated new data upload:
cd /
eval $(ssh-agent -s)
ssh-add /xdmod_ansible/roles/common/files/dice_sif_rsa
git clone git@github.com:klin938/sge_accounting.git
Make sure it is cloned by SSH as required by passwordless git pull by cron (See below).
A special repo commit marker file needs to be created directly under the repository. The commit hash number is choosen according to which portion of the SGE data has been loaded in the init_load_data phase.
cd /sge_accounting
echo "FROM_COMMIT_HASH" > FROM_COMMIT_HASH
This only needs to be done once on the newly built container, it will be updated by the cron data load process automatically.
Once the initial data load is done, all future data will be loaded by cron fetching new data from the repository:
/etc/cron.daily/xdmod.cron.daily
crond is not activated by default, this is intentional. It should be activated manually after initial data load completed:
crond -s
IMPORTANT: make sure FROM_COMMIT_HASH has been created for the first run of the cron script.