TOP_DIR=$(cd $(dirname "$0") && pwd)
- ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
--os-tenant-name $DOCTOR_PROJECT"
compute_host_in_undercloud=${COMPUTE_HOST%%.*}
die_if_not_set $LINENO COMPUTE_HOST "Failed to get compute hostname"
- if is_installer apex; then
- COMPUTE_USER=${COMPUTE_USER:-heat-admin}
- COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
- "source stackrc; \
- nova show $compute_host_in_undercloud \
- | awk '/ ctlplane network /{print \$5}'")
- elif is_installer fuel; then
- COMPUTE_USER=${COMPUTE_USER:-root}
- node_id=$(echo $compute_host_in_undercloud | cut -d "-" -f 2)
- COMPUTE_IP=$(sshpass -p r00tme ssh 2>/dev/null $ssh_opts root@${INSTALLER_IP} \
- "fuel node|awk -F '|' -v id=$node_id '{if (\$1 == id) print \$5}' |xargs")
- elif is_installer local; then
- COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
- COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
- fi
+ get_compute_ip_from_hostname $COMPUTE_HOST
- die_if_not_set $LINENO COMPUTE_IP "Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
echo "COMPUTE_HOST=$COMPUTE_HOST"
echo "COMPUTE_IP=$COMPUTE_IP"
fi
}
- get_consumer_ip() {
+ # TODO(r-mibu): update this function to support consumer instance
+ # and migrate this function into installer lib
+ get_consumer_ip___to_be_removed() {
local get_consumer_command="ip route get $COMPUTE_IP | awk '/ src /{print \$NF}'"
if is_installer apex; then
CONSUMER_IP=$(sudo ssh $ssh_opts root@$INSTALLER_IP \
# avoid some network problems dpends on infra and installers.
# This tunnel will be terminated by stop_consumer() or after 10 mins passed.
if ! is_installer local; then
- if is_installer apex; then
- CONTROLLER_IPS=$(sudo ssh $ssh_opts $INSTALLER_IP \
- "source stackrc; \
- nova list | grep ' overcloud-controller-[0-9] ' \
- | sed -e 's/^.*ctlplane=//' -e 's/ *|\$//'")
- elif is_installer fuel; then
- get_controller_ips
- fi
-
- die_if_not_set $LINENO CONTROLLER_IPS "Could not get CONTROLLER_IPS."
for ip in $CONTROLLER_IPS
do
forward_rule="-R $CONSUMER_PORT:localhost:$CONSUMER_PORT"
[[ -n "$dev" ]] || dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $5}')
sleep 1
sudo ip link set $dev down
-echo "doctor set host down at" $(date "+%s.%N")
+echo "doctor set link down at" $(date "+%s.%N")
sleep 180
sudo ip link set $dev up
sleep 1
chmod +x disable_network.sh
scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
-}
-
-profile_performance_poc() {
- triggered=$(grep "^doctor set host down at" disable_network.log |\
- sed -e "s/^.* at //")
- vmdown=$(grep "doctor mark vm.* error at" inspector.log |tail -n 1 |\
- sed -e "s/^.* at //")
- hostdown=$(grep "doctor mark host.* down at" inspector.log |\
- sed -e "s/^.* at //")
-
- #calculate the relative interval to triggered(T00)
- export DOCTOR_PROFILER_T00=0
- export DOCTOR_PROFILER_T01=$(echo "($detected-$triggered)*1000/1" |bc)
- export DOCTOR_PROFILER_T03=$(echo "($vmdown-$triggered)*1000/1" |bc)
- export DOCTOR_PROFILER_T04=$(echo "($hostdown-$triggered)*1000/1" |bc)
- export DOCTOR_PROFILER_T09=$(echo "($notified-$triggered)*1000/1" |bc)
-
- python profiler-poc.py
+ # use host time to get rid of potential time sync deviation between nodes
+ triggered=$(date "+%s.%N")
}
calculate_notification_time() {
notified=$(grep "doctor consumer notified at" consumer.log |\
sed -e "s/^.* at //")
- if [[ "$PROFILER_TYPE" == "poc" ]]; then
- profile_performance_poc
- fi
-
echo "$notified $detected" | \
awk '{
d = $1 - $2;
-f value -c State | grep -q down' 240 5
}
+collect_logs() {
+ unset_forced_down_hosts
+ # TODO: We need to make sure the target compute host is back to IP
+ # reachable. wait_ping() will be added by tojuvone .
+ sleep 110
+ scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
+
+ # TODO(yujunz) collect other logs, e.g. nova, aodh
+}
+
+run_profiler() {
+ if [[ "$PROFILER_TYPE" == "poc" ]]; then
+ linkdown=$(grep "doctor set link down at " disable_network.log |\
+ sed -e "s/^.* at //")
+ vmdown=$(grep "doctor mark vm.* error at" inspector.log |tail -n 1 |\
+ sed -e "s/^.* at //")
+ hostdown=$(grep "doctor mark host.* down at" inspector.log |\
+ sed -e "s/^.* at //")
+
+ # TODO(yujunz) check the actual delay to verify time sync status
+ # expected ~1s delay from $trigger to $linkdown
+ relative_start=${linkdown}
+ export DOCTOR_PROFILER_T00=$(python -c \
+ "print(int(($linkdown-$relative_start)*1000))")
+ export DOCTOR_PROFILER_T01=$(python -c \
+ "print(int(($detected-$relative_start)*1000))")
+ export DOCTOR_PROFILER_T03=$(python -c \
+ "print(int(($vmdown-$relative_start)*1000))")
+ export DOCTOR_PROFILER_T04=$(python -c \
+ "print(int(($hostdown-$relative_start)*1000))")
+ export DOCTOR_PROFILER_T09=$(python -c \
+ "print(int(($notified-$relative_start)*1000))")
+
+ python profiler-poc.py >doctor_profiler.log 2>&1
+ fi
+}
+
cleanup() {
set +e
echo "cleanup..."
# TODO: We need to make sure the target compute host is back to IP
# reachable. wait_ping() will be added by tojuvone .
sleep 110
- scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
+ if is_set COMPUTE_IP; then
+ scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
+ fi
openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
sleep 1
check_host_status "(DOWN|UNKNOWN)"
calculate_notification_time
+collect_logs
+run_profiler
echo "done"
Code Review / doctor.git / commitdiff