airship/maas
Code Issues Proposed changes

[FIX] - Fixed maas-rack reschedule issue

Browse Source 
After a maas-rack pod rescheduling from one node to another,
only delete the node's local maas registration state, and
keep re-registering the maas-rack until success.

Change-Id: I8495ae61f38377d292a26ad09438ece40b2c53c6
This commit is contained in:
Ahmad Mahmoudi 2019-03-08 02:35:50 -06:00
parent 5e4ab93da8
commit 37dfe50954
1 changed files with 14 additions and 93 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

107
charts/maas/templates/bin/_register-rack-controller.sh.tpl
View File

@ -2,29 +2,8 @@
set -x set -x
# Path where the host's cloud-init data is mounted login_maas() {
# to source the maas system_id echo "Logging in to MAAS region."
HOST_MOUNT_PATH=${HOST_MOUNT_PATH:-"/host_cloud-init/"}
get_impacted_nets() {
system_id="$1"
maas local fabrics read | jq -cr 'map(.vlans) | map(.[]) | map(select(.primary_rack == "'"$system_id"'" or .secondary_rack == "'"$system_id"'")) | .[] | {vid, fabric_id}'
}
detach_rack_controller() {
system_id="$1"
for net in $(get_impacted_nets "$system_id");
do
vid=$(echo "$net" | jq -r .vid)
fid=$(echo "$net" | jq -r .fabric_id)
maas local vlan update "$fid" "$vid" primary_rack='' secondary_rack='' dhcp_on=false
done
}
unregister_maas_rack() {
sys_id="$1"
echo "Deregister this pod as MAAS rack controller ${sys_id}."
maas login local "$MAAS_ENDPOINT" "$MAAS_API_KEY" maas login local "$MAAS_ENDPOINT" "$MAAS_API_KEY"
if [[ $? -ne 0 ]]; if [[ $? -ne 0 ]];
@ -32,37 +11,21 @@ unregister_maas_rack() {
echo "Could not login to MAAS API." echo "Could not login to MAAS API."
exit $? exit $?
fi fi
}
detach_rack_controller "$sys_id" unregister_maas_rack() {
# In oder to avoid the issue with race condition in maas,
while [ 1 ]; # do not de-register the dead maas-controller from mass-region
do # just delete the local state of the maas-controller's.
maas local rack-controller delete "$sys_id" echo "Deregistering this pod's local state in /var/lib/maas directory."
rm -f /var/lib/maas/secret
if [[ $? -ne 0 ]]; rm -f /var/lib/maas/maas_id
then
echo "Could not delete rack controller."
sleep 10
else
break
fi
done
rm -f ~maas/maas_id
rm -f ~maas/secret
} }
register_maas_rack() { register_maas_rack() {
sys_id=${1:-""}
echo "register-rack-controller URL: ${MAAS_ENDPOINT}" echo "register-rack-controller URL: ${MAAS_ENDPOINT}"
if [[ ! -z "$sys_id" ]] # register forever until success
then
echo "Using provided system id ${sys_id}."
echo "$sys_id" > ~maas/maas_id
fi
# register forever
while [ 1 ]; while [ 1 ];
do do
if maas-rack register --url="${MAAS_ENDPOINT}" --secret="${MAAS_REGION_SECRET}"; if maas-rack register --url="${MAAS_ENDPOINT}" --secret="${MAAS_REGION_SECRET}";
@ -76,48 +39,6 @@ register_maas_rack() {
done; done;
} }
get_host_identity() { login_maas
# Check if the underlying host was deployed by MAAS unregister_maas_rack
if [[ -r "${HOST_MOUNT_PATH}/instance-data.json" ]] register_maas_rack
then
grep -E 'instance-id' "${HOST_MOUNT_PATH}/instance-data.json" | head -1 | tr -d ' ",' | cut -d: -f 2
else
echo ""
fi
}
get_pod_identity() {
if [[ -r ~maas/maas_id ]]
then
cat ~maas/maas_id
else
echo ""
fi
}
HOST_SYSTEM_ID=$(get_host_identity)
POD_SYSTEM_ID=$(get_pod_identity)
# This Pod state already has a MAAS identity
if [[ ! -z "$POD_SYSTEM_ID" ]]
then
# If the pod maas identity doesn't match the
# host maas identity, unregister the pod identity
# as a rack controller
if [[ "$HOST_SYSTEM_ID" != "$POD_SYSTEM_ID" ]]
then
unregister_maas_rack "$POD_SYSTEM_ID"
if $?;
then
echo "Unregister of $POD_SYSTEM_ID failed, exitting."
exit 1
fi
register_maas_rack "$HOST_SYSTEM_ID"
else
echo "Found existing maas_id, assuming already registered."
fi
exit 0
else
register_maas_rack
fi