${SSH} ${ip} "sudo find /etc > /tmp/find.etc.txt"
${SSH} ${ip} "sudo find /opt/stack > /tmp/find.opt.stack.txt"
${SSH} ${ip} "sudo find /var > /tmp/find2.txt"
- rsync --rsync-path="sudo rsync" -arv --list-only ssh ${ip}:/var/ > ${folder}/rsync.var.txt
+ ${SSH} ${ip} "sudo find /var > /tmp/find.var.txt"
+ rsync --rsync-path="sudo rsync" --list-only -arvhe ssh ${ip}:/etc/ > ${folder}/rsync.etc.txt
+ rsync --rsync-path="sudo rsync" --list-only -arvhe ssh ${ip}:/opt/stack/ > ${folder}/rsync.opt.stack.txt
+ rsync --rsync-path="sudo rsync" --list-only -arvhe ssh ${ip}:/var/ > ${folder}/rsync.var.txt
scp ${ip}:/tmp/find.etc.txt ${folder}
scp ${ip}:/tmp/find.opt.stack.txt ${folder}
scp ${ip}:/tmp/find2.txt ${folder}
+ scp ${ip}:/tmp/find.var.txt ${folder}
}
function collect_logs () {
set +e # We do not want to create red dot just because something went wrong while fetching logs.
- for i in `seq 1 ${NUM_ODL_SYSTEM}`; do
- CONTROLLERIP=ODL_SYSTEM_${i}_IP
- echo "Lets's take the karaf thread dump again..."
- KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
- ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_after.log || true
- echo "killing karaf process..."
- ${SSH} "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh'
- done
-
cat > extra_debug.sh << EOF
echo -e "/usr/sbin/lsmod | /usr/bin/grep openvswitch\n"
/usr/sbin/lsmod | /usr/bin/grep openvswitch
# FIXME: Do not create .tar and gzip before copying.
for i in `seq 1 ${NUM_ODL_SYSTEM}`; do
CONTROLLERIP=ODL_SYSTEM_${i}_IP
+ echo "collect_logs: for opendaylight controller ip: ${!CONTROLLERIP}"
NODE_FOLDER="odl_${i}"
mkdir -p ${NODE_FOLDER}
- ${SSH} ${!CONTROLLERIP} "journalctl > /tmp/journalctl.log"
+ echo "Lets's take the karaf thread dump again..."
+ ${SSH} ${!CONTROLLERIP} "sudo ps aux > /tmp/ps.log"
+ KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
+ ssh ${!CONTROLLERIP} "jstack ${KARAF_PID}"> ${WORKSPACE}/karaf_${i}_threads_after.log || true
+ echo "killing karaf process..."
+ ${SSH} "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh'
+ ${SSH} ${!CONTROLLERIP} "sudo journalctl > /tmp/journalctl.log"
scp ${!CONTROLLERIP}:/tmp/journalctl.log ${NODE_FOLDER}
${SSH} ${!CONTROLLERIP} "dmesg -T > /tmp/dmesg.log"
scp ${!CONTROLLERIP}:/tmp/dmesg.log ${NODE_FOLDER}
# Control Node
for i in `seq 1 ${NUM_OPENSTACK_CONTROL_NODES}`; do
OSIP=OPENSTACK_CONTROL_NODE_${i}_IP
+ echo "collect_logs: for openstack control node ip: ${!OSIP}"
NODE_FOLDER="control_${i}"
mkdir -p ${NODE_FOLDER}
+ scp ${!OSIP}:/etc/dnsmasq.conf ${NODE_FOLDER}
scp ${!OSIP}:/etc/keystone/keystone.conf ${NODE_FOLDER}
scp ${!OSIP}:/etc/keystone/keystone-uwsgi-admin.ini ${NODE_FOLDER}
scp ${!OSIP}:/etc/keystone/keystone-uwsgi-public.ini ${NODE_FOLDER}
scp ${!OSIP}:/opt/stack/requirements/upper-constraints.txt ${NODE_FOLDER}
scp ${!OSIP}:/opt/stack/tempest/etc/tempest.conf ${NODE_FOLDER}
scp ${!OSIP}:/tmp/get_devstack.sh.txt ${NODE_FOLDER}
- scp ${!OSIP}:/var/log/httpd/keystone_access.log ${NODE_FOLDER}
- scp ${!OSIP}:/var/log/httpd/keystone.log ${NODE_FOLDER}
scp ${!OSIP}:/var/log/openvswitch/ovs-vswitchd.log ${NODE_FOLDER}
scp ${!OSIP}:/var/log/openvswitch/ovsdb-server.log ${NODE_FOLDER}
list_files "${!OSIP}" "${NODE_FOLDER}"
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/etc/hosts ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/usr/lib/systemd/system/haproxy.service ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/audit/audit.log ${NODE_FOLDER}
- rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/dmesg.log ${NODE_FOLDER}
+ rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/httpd/keystone_access.log ${NODE_FOLDER}
+ rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/httpd/keystone.log ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/messages ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/rabbitmq ${NODE_FOLDER}
rsync -avhe ssh ${!OSIP}:/opt/stack/logs/* ${NODE_FOLDER} # rsync to prevent copying of symbolic links
scp ${!OSIP}:/tmp/extra_debug.log ${NODE_FOLDER}
scp ${!OSIP}:/tmp/journalctl.log ${NODE_FOLDER}
scp ${!OSIP}:/tmp/*.xz ${NODE_FOLDER}
+ ${SSH} ${!CONTROLLERIP} "dmesg -T > /tmp/dmesg.log"
+ scp ${!CONTROLLERIP}:/tmp/dmesg.log ${NODE_FOLDER}
mv local.conf_control_${!OSIP} ${NODE_FOLDER}/local.conf
+ mv /tmp/qdhcp ${NODE_FOLDER}
mv ${NODE_FOLDER} ${WORKSPACE}/archives/
done
# Compute Nodes
for i in `seq 1 ${NUM_OPENSTACK_COMPUTE_NODES}`; do
OSIP=OPENSTACK_COMPUTE_NODE_${i}_IP
+ echo "collect_logs: for openstack compute node ip: ${!OSIP}"
NODE_FOLDER="compute_${i}"
mkdir -p ${NODE_FOLDER}
scp ${!OSIP}:/etc/nova/nova.conf ${NODE_FOLDER}
list_files "${!OSIP}" "${NODE_FOLDER}"
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/etc/hosts ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/audit/audit.log ${NODE_FOLDER}
- rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/dmesg.log ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/libvirt ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/messages ${NODE_FOLDER}
rsync --rsync-path="sudo rsync" -avhe ssh ${!OSIP}:/var/log/nova-agent.log ${NODE_FOLDER}
scp ${!OSIP}:/tmp/extra_debug.log ${NODE_FOLDER}
scp ${!OSIP}:/tmp/journalctl.log ${NODE_FOLDER}
scp ${!OSIP}:/tmp/*.xz ${NODE_FOLDER}/
+ ${SSH} ${!OSIP} "dmesg -T > /tmp/dmesg.log"
+ scp ${!OSIP}:/tmp/dmesg.log ${NODE_FOLDER}
mv local.conf_compute_${!OSIP} ${NODE_FOLDER}/local.conf
mv ${NODE_FOLDER} ${WORKSPACE}/archives/
done
# AccessRefused: (0, 0): (403) ACCESS_REFUSED - Login was refused using authentication mechanism AMQPLAIN. For details see the broker logfile.
# Compare that timestamp to this log in the control stack.log: sudo rabbitmqctl set_permissions -p nova_cell1 stackrabbit
# If the n-cpu.log is earlier than the control stack.log timestamp then the failure condition is likely hit.
-
-echo "Wait a maximum of 30m until rabbitmq is ready to allow the controller to create nova_cell1 before the computes need it"
-retry 30 60 "is_rabbitmq_ready ${OPENSTACK_CONTROL_NODE_1_IP}"
+WAIT_FOR_RABBITMQ_MINUTES=60
+echo "Wait a maximum of ${WAIT_FOR_RABBITMQ_MINUTES}m until rabbitmq is ready to allow the controller to create nova_cell1 before the computes need it"
+retry ${WAIT_FOR_RABBITMQ_MINUTES} 60 "is_rabbitmq_ready ${OPENSTACK_CONTROL_NODE_1_IP}"
rc=$?
if ((${rc} == 0)); then
echo "rabbitmq is ready, starting ${NUM_OPENSTACK_COMPUTE_NODES} compute(s)"
else
- echo "rabbitmq was not ready in "
+ echo "rabbitmq was not ready in ${WAIT_FOR_RABBITMQ_MINUTES}m"
collect_logs
exit 1
fi
Code Review / releng / builder.git / blobdiff