UGE cluster with Tortuga on AWS with at least one compute node.
- run configuration script as
root:
# ./configure.sh
It will install qsub-wrapper.sh and scale-up.sh scripts to Tortuga binary directory.
- create input data directory for jobs:
$ mkdir -p ~/input1; touch ~/input1/infile1
$ mkdir -p ~/input2; touch ~/input2/infile2
- submit 2 jobs with qsub wrapper script (see
qsub-wrapper.sh -hfor details) from regular UGE user (centos):
$ qsub-wrapper.sh -src HOME/input1 -dest LOCAL -sync-back /home/$USER/out1:HOME/out1 -j y test/job.sh infile1 newfile1
$ qsub-wrapper.sh -src HOME/input2 -dest LOCAL -sync-back /home/$USER/out2:HOME/out2 -j y test/job.sh infile2 newfile2
For each job input file will be transferred to the remote cluster node (to be created by scale-up.sh script) and new file will be created by job script in the same outout directory. Upon job completion both file will be transferred back to the local cluster.
-
check that jobs are in pending state and get job ids (
qstat -f) -
run manually cluster scale script as
rootuser (could be invoked form Unisight rule engine or another periodic or event based trigger). For 2 jobs started above bycentosuser following invocation should be used:
# scale-up.sh <job_id>,<job_id> 1,1 centos,centos 0,0 2,2 all.q
This script will provision and configure new node[s] as well as transfer input data expected by jobs to the remote cluster nodes.
-
wait for completion of the jobs
-
check output of the jobs in submitter (
centos) home directory:
$ ls ~/out1
infile1 job.sh.o<job_id> newfile1
$ ls ~/out2
infile2 job.sh.o<job_id> newfile2
- clean output and submit one of the jobs again:
$ rm -rf ~/out1
$ qsub-wrapper.sh -src HOME/input1 -dest LOCAL -sync-back /home/$USER/out1:HOME/out1 -j y test/job.sh infile1 newfile99
With no other pending jobs the new job will start on the node with data (infile1) already present.
- observe outout:
$ ls ~/out1
infile1 job.sh.o<job_id> newfile99