Basically if I do next upgrade of ONAP components by using helm deploy and there are no changes in mariadb-galera charts pre-install script scales down maradb-galera stateful set to 0 replicas, then as there are no changes in charts helm doesn't do anything to deploy mariadb-galera cluster and after post-upgrade script removes temporary pod and that is causing lose of whole galera cluster and data.
needed components upgraded. ones without changes still running
ACTUAL RESULTS
mariadb-galera clusters gets scaled down by pre-upgrade hook job and as there are no changes no deployment is done. and when post-upgrade script kicks in it deletes additional galera node and all data is lost.
SUGGESTED FIX.
add additional logic to post-upgrade hook to scale up galera cluster back if it is size of 0. (will do test in my local environment)
ideal would be to see if any upgrade is going to happen on pre-upgrade hook, but after quick look couldn't find such option.
SUMMARY
Rolling upgrade of MariaDB-Galera ( https://jira.onap.org/browse/OOM-2316 ) doesn't work if there is no actual upgrade in charts.
Basically if I do next upgrade of ONAP components by using helm deploy and there are no changes in mariadb-galera charts pre-install script scales down maradb-galera stateful set to 0 replicas, then as there are no changes in charts helm doesn't do anything to deploy mariadb-galera cluster and after post-upgrade script removes temporary pod and that is causing lose of whole galera cluster and data.
OS / ENVIRONMENT
Kubernetes version:
1.15.6
Helm version:
2.16.6
OOM VERSION
frankfurt
CONFIGURATION
mariadb-galera with 3 replicas
STEPS TO REPRODUCE
deploy with command: helm deploy onap6 local/onap --namespace onap6 -f onap/resources/overrides/myoverride.yaml --verbose
upgrade with command: helm deploy onap6 local/onap --namespace onap6 -f onap/resources/overrides/myoverride.yaml --verbose
EXPECTED RESULTS
needed components upgraded. ones without changes still running
ACTUAL RESULTS
mariadb-galera clusters gets scaled down by pre-upgrade hook job and as there are no changes no deployment is done. and when post-upgrade script kicks in it deletes additional galera node and all data is lost.
SUGGESTED FIX.
add additional logic to post-upgrade hook to scale up galera cluster back if it is size of 0. (will do test in my local environment)
ideal would be to see if any upgrade is going to happen on pre-upgrade hook, but after quick look couldn't find such option.