This little project solves an annoying little issue, if you run microservices using Liquibase, e.g. in Spring, in kubernetes. Sometimes it happens, that during the start phase of a microservice, liquibase places a lock in the database, which never gets unlocked. Consequently, the microservice keeps on hanging forever. The healing operation as well as the observation is quite easy. It sometime might be a greater damage, if one of our services is crash-looping for a long time.
This guard solves the problem.
You can use the helm package in this project to directly install it in your cluster:
helm --namespace <NS_WHERE_YOUR_SERVICES_RUN> install liquibase-lock-guard helm/liquibase-lock-guardThen a small little pod appears and watches the evil constellation and cleans it up for you.
There are several reasons, why and how the deadlock appears. Sometimes it happens, when the service pod is killed for some reason shortly after placing the lock, or if multiple replicas of the same services run into a race-condition to the lock.
However, a typical liquibase managed database consists of a table called databasechangeloglock. This table holds the lock
to prevent actual race-conditions, it makes sense. But if the column locked is not set back to 'f' during start phase
all services will keep on hanging on that point forever. If you can determine, that the lock was granted more than an hour ago,
it usually means that liquibase caused a deadlock.
This issue mainly occurs in applications generated and deployed by JHipster, but can also occur for any Spring application using liquibase