cloudlab API

cloudlab

#!/bin/bash
# https://gitlab.flux.utah.edu/stoller/portal-tools
# git clone https://gitlab.flux.utah.edu/stoller/portal-tools
# cd portal-tools; python3 setup.py install
# sudo cp cl.sh /usr/local/bin/cl && sudo chmod +x /usr/local/bin/cl

# set -eux

PROJ=cache
USER=juncheng
PROFILE=compute
SITE=wisc
NODES_WISC=(d7525 d8545 sm220u sm110p c220g5 c240g5 c4130)
NODES_CLEMSON=(c6320 c8220 c8220x c6420 dss7500 r6525 r650 r7525)
NODES_UTAH=(flex01 flex02 flex03 flex04 d2950 d6515 c6525-25g c6525-100g dl360)

NODE1=""
N_NODE1=1

if [ $# -le 0 ]; then
    echo "Usage: $0 <node>"
    exit 1
fi
NODE1=$1

if [ $# -eq 2 ]; then
    N_NODE1=$2
fi

if [[ " ${NODES_WISC[@]} " =~ " ${NODE1} " ]]; then
    SITE=wisc
elif [[ " ${NODES_CLEMSON[@]} " =~ " ${NODE1} " ]]; then
    SITE=clemson
elif [[ " ${NODES_UTAH[@]} " =~ " ${NODE1} " ]]; then
    SITE=utah
else
    echo "Node not found in a site"
    exit 1
fi

EXP=${N_NODE1}-${NODE1}-${RANDOM}

#### create experiment ####
reset
echo "create experiment with ${N_NODE1} ${NODE1} in ${SITE} exp ${EXP} proj ${PROJ} profile ${PROFILE}"
startExperiment --project ${PROJ} --name ${EXP} --binding='{"nNode1": "'${N_NODE1}'", "site": "'${SITE}'", "nodeType1": "'${NODE1}'", "anyHardware": "False"}' --site ${SITE} ${PROJ},${PROFILE}

#### wait for the experiment to be ready ####
status=$(experimentStatus -j ${PROJ},${EXP} | jq -r '.status')
echo -n waiting for the experiment to be ready
while [[ "${status}" != "ready" ]]; do
    sleep 20
    echo -n .
    status=$(experimentStatus -j ${PROJ},${EXP} | jq -r '.status')
    if [[ "${status}" == "failed" || "${status}" == "" ]]; then
        echo
        echo "experiment failed"
        terminateExperiment ${PROJ},${EXP}
        exit 1
    fi
done
extendExperiment -m 'I need them longer' ${PROJ},${EXP} 144

# wait for the nodes to be ready
sleep 80
echo

#### obtain hostname ####
hostnames=$(experimentStatus -j ${PROJ},${EXP} | jq -r '.aggregate_status | to_entries[] | .value.nodes | to_entries[] | .value.hostname')
hostids=()
hostips=()

#### initiate the node ####
for hostname in ${hostnames[@]}; do
    hostid=$(ssh -q \
        -o StrictHostKeyChecking=no \
        -o UserKnownHostsFile=/dev/null \
        -o LogLevel=ERROR -o ConnectTimeout=10 \
        ${USER}@${hostname} "cat /var/emulab/boot/nodeid | tr -dc '[:print:]'")
    hostids+=($hostid)

    hostip=$(ssh -q \
        -o StrictHostKeyChecking=no \
        -o UserKnownHostsFile=/dev/null \
        -o LogLevel=ERROR -o ConnectTimeout=10 \
        ${USER}@${hostname} "cat /var/emulab/boot/myip | tr -dc '[:print:]'")
    hostips+=($hostip)

    (ssh -q -t \
        -o StrictHostKeyChecking=no \
        -o UserKnownHostsFile=/dev/null \
        -o ServerAliveInterval=1200 \
        ${USER}@${hostname} """bash /proj/${PROJ}-PG0/jason/scripts/init.sh""")

done
wait

#### set fastscp DNS name ####
echo
dns_name_base=cl
dns_name_start_idx=0

for i in $(seq 0 100); do
    dns_name=${dns_name_base}${i}
    (ssh -q -t \
        -o StrictHostKeyChecking=no \
        -o UserKnownHostsFile=/dev/null \
        -o LogLevel=ERROR -o ConnectTimeout=10 \
        ${USER}@${dns_name}.fastscp.com "ls >/dev/null") 2>/dev/null
    if [[ $? != 0 ]]; then
        dns_name_start_idx=$i
        break
    fi
done

n=${#hostips[@]}
n1=$((n - 1))
for idx in $(seq 0 ${n1}); do
    hostip=${hostips[idx]}
    hostid=${hostids[idx]}
    dns_name_idx=$((dns_name_start_idx + idx))
    dns_name=${dns_name_base}${dns_name_idx}
    echo "set DNS name ${dns_name} for ${hostip} ${hostid}"
    cf_dns ${dns_name} ${hostip}
done

echo "#############################################################################"
echo "############                                                    #############"
for hostid in ${hostids[@]}; do
    echo "########### finish preparing node ${hostid} ############"
done
echo "############                                                    #############"
echo "#############################################################################"

# terminateExperiment ${PROJ},${EXP}