This improvement request comes from this bug: https://jira.percona.com/browse/K8SPXC-585
If pmm-client cannot connect to pmm-server for any reason (network glitch, wrong password), pmm-client container crashes.
This means that monitoring server stability can affect the stability of the database cluster.
Container should not crash, but just throw an error.
Good example - mongodb-exporter. If it cannot connect to MongoDB database:
- it still runs
- on metrics page it tells that it cannot connect to MongoDB and states the reason
Default entrypoint shouldn't exit with error, but waiting in infinite loop with trying to apply failed commands and provide logs to show issue.
How to test:
- Use FB from https://github.com/Percona-Lab/pmm-submodules/pull/1924
- Enable DBaaS
- Set wrong public address in settings
- Create new DB Cluster
- Check that DB Cluster isn't failed because of failing pmm-client