useful_scripts/fix_cloud.sh

207 lines
6.1 KiB
Bash
Executable File

#!/bin/bash
# Used for debugging
# set -ax
# This script is required after a reboot of the cloud after the cloud has been
# shut down
check_unit_status()
{
app_name=$1
status_check="$2"
unit_status=$(juju status --format json | jq -rc ".applications.${app_name}.units | to_entries[] | {sub:.key,status:.value[\"workload-status\"].message}")
app_units=$(echo ${unit_status} | jq .sub | sed s/\"//g)
num=0
for unit in ${app_units} ; do
this_unit_status=$(echo $unit_status | jq -rc . | grep ${unit} | jq .status | sed s/\"//g)
if [[ "${this_unit_status}" == "${status_check}" ]] ; then
(( num++ ))
fi
done
if [[ $num -ge 3 ]] ; then echo 1
else echo 0
fi
}
get_lead()
{
app_name=$1
cat ${juju_status} | jq -rc ".applications.${app_name}.units | to_entries[] | select(.value.leader == "true") | .key"
}
do_vault()
{
vault_vip=$(juju config vault vip)
echo export VAULT_ADDR="http://${vault_vip}:8200"
export VAULT_ADDR="http://${vault_vip}:8200"
echo " "
IPS=$(cat ${juju_status} | jq '.applications.vault.units | to_entries[] | .value."public-address"' | sed s/\"//g)
for ip in $IPS;do
echo export VAULT_ADDR=http://${ip}:8200;
export VAULT_ADDR=http://${ip}:8200;
for vault_key in $(head -n3 vault-secrets.txt | awk '{print $4}');do
echo vault operator unseal -tls-skip-verify $vault_key
vault operator unseal -tls-skip-verify $vault_key
done;
done;
juju run -a vault "hooks/update-status"
}
juju-wait -v
juju_status=$(mktemp)
juju status --format json > ${juju_status}
mysql_status=$(cat ${juju_status} | jq -rc ".applications.mysql.units | to_entries[] | {sub:.key,status:.value[\"workload-status\"].message}")
#{"sub":"mysql/0","status":"MySQL is down. Sequence Number: 102921. Safe To Bootstrap: 1"}
#{"sub":"mysql/1","status":"MySQL is down. Sequence Number: 102921. Safe To Bootstrap: 0"}
#{"sub":"mysql/2","status":"MySQL is down. Sequence Number: 102921. Safe To Bootstrap: 0"}
mysql_units=$(echo ${mysql_status} | jq .sub | sed s/\"//g)
bootstrap_unit=""
mysql_lead=$(get_lead mysql)
safe_to_bootstrap=$(echo $mysql_status | jq -rc . | grep "Safe To Bootstrap: 1" | jq .sub | sed s/\"//g)
if [[ -n "$safe_to_bootstrap" ]]
then
bootstrap_unit=$safe_to_bootstrap
else
seq_number=$(echo $mysql_status | jq -rc . | grep "Sequence Number" )
if [[ -n "${seq_number}" ]]
then
seqs=$(echo $seq_number | jq -rc ". | {sub:.sub,seq:(.status|split(\".\")[1]|split(\": \")[1])}")
uniq_seqs=$(echo $seqs| jq .seq | sed s/\"//g | sort -n | uniq)
seq_count=$(echo $uniq_seqs | xargs | wc -w)
highest_seq=$(echo "${seqs}"| jq .seq | sed s/\"//g | sort -n | uniq | tail -n 1)
lowest_seq=$(echo "${seqs}"| jq .seq | sed s/\"//g | sort -n | uniq | head -n 1)
if [[ ${seq_count} -eq 1 ]]
then # same seq numbers all round
if [[ ${highest_seq} -eq -1 ]]
then # if all seq numbers are -1
echo "The sequence number is -1 ... exiting"
exit 1
fi
bootstrap_unit=${mysql_lead}
else # we have different seq numbers
unit_high_seq=$(echo $seqs | jq -rc . | grep ${highest_seq} | jq .sub | sed s/\"//g | tail -n 1)
bootstrap_unit=${unit_high_seq}
fi
fi
fi
if [[ -n ${bootstrap_unit} ]]
then
juju run-action --wait ${bootstrap_unit} bootstrap-pxc
juju run --application mysql "hooks/update-status"
until [[ $(check_unit_status mysql "Unit waiting for cluster bootstrap") -eq 1 ]]
do
sleep 10
done
if [[ "${bootstrap_unit}" == "${mysql_lead}" ]] ; then
for unit in ${mysql_units}; do
if [[ "${unit}" != "${mysql_lead}" ]] ; then
juju run-action --wait ${unit} notify-bootstrapped
ran_bootstrap="true"
break
fi
done
else
juju run-action --wait ${mysql_lead} notify-bootstrapped
ran_bootstrap="true"
fi
juju run -a mysql "hooks/update-status"
until [[ $(check_unit_status mysql "Unit is ready") -eq 1 ]]
do
sleep 10
done
# This is so that nagios doesn't report that the mysql daemon is down
# although the process is running. juju will then automatically start
# the mysqld process
juju ssh ${bootstrap_unit} -- sudo reboot
fi
juju run -a nova-cloud-controller -- sudo systemctl restart nova-api-os-compute nova-conductor nova-consoleauth &
juju run -a heat -- sudo systemctl restart heat-engine &
juju run -a vault -- sudo systemctl restart vault &
wait
for app in nova-cloud-controller heat vault ; do
juju run -a $app "hooks/update-status" &
done
wait
# cleanup all crm resources
cat ${juju_status} | jq ".applications | to_entries[] | select(.value[\"charm-name\"] == \"hacluster\") | .key" | sed s/\"//g | xargs -i juju run --unit "{}"/leader -- 'sudo crm_resource -l | sed s/:.*//g | uniq | xargs -i sudo crm resource cleanup \"\{\}\"'
do_vault
# Wait 10 seconds, and ensure that vault is unsealed
echo "Sleeping 10 seconds to wait for vault to finalise unseal"
sleep 10
ceph_osd_apps=$(cat ${juju_status} | jq -rc ".applications | to_entries[] | select(.value[\"charm-name\"] == \"ceph-osd\") | .key")
ceph_osds=""
for apps in ${ceph_osd_apps}
do
ceph_osds="${ceph_osds} $(cat ${juju_status} | jq -rc ". | .applications[\"${apps}\"].units | to_entries[] | .key")"
done
for ceph_osd in ${ceph_osds}
do
juju ssh ${ceph_osd} -- 'sudo systemctl kill --all --type=service vaultlocker-decrypt@* ;
sudo systemctl start --all --type=service vaultlocker-decrypt@* ;
sudo systemctl start --all --type=service ceph-volume@*' &
done
wait
juju run -a ceph-osd "hooks/update-status"
lds_servers=$(cat ${juju_status} | jq -rc ". | .applications[\"landscape-server\"].units | to_entries[] | .key")
cat > /tmp/restart-landscape.sh << EOF
#!/bin/bash
sudo systemctl restart landscape-*
EOF
for lds_server in ${lds_servers}
do
juju scp /tmp/restart-landscape.sh ${lds_server}:.
juju ssh ${lds_server} chmod +x restart-landscape.sh
juju ssh ${lds_server} sudo ./restart-landscape.sh &
done
wait
juju run --all -- sudo systemctl restart systemd-resolved