Lower alert severity for memory based alerts

[david-martin]

JIRA ID

https://issues.redhat.com/browse/INTLY-6614

Additional Information

Some discussion in https://issues.redhat.com/browse/INTLY-6614, the gist of which is a review of the critical alerts and if they should be critical (i.e. they page someone when they fire).

My thinking is that only problems that affect a potential service level objective (SLO) should be critical.
In this case, its:

• KeycloakInstanceNotAvailable if the service isn't available
• KeycloakAPIRequestDuration90PercThresholdExceeded & KeycloakAPIRequestDuration99.5PercThresholdExceeded if the service isn't responding as fast as expected

All other alerts would help with debugging & point at potential causes as to why one of the critical alerts are firing, but as long as the SLO is being met, no-one needs to be paged.

Verification Steps

1. Install operator in a cluster with prometheus-operator
2. Verify the alerts are created with the expected severity

Checklist:

• Verified by team member
• Comments where necessary
• Automated Tests
• Documentation changes if necessary

Additional Notes

[david-martin]

ping @pb82 @slaskawi

[coveralls]

Coverage remained the same at 41.127% when pulling a0a3e7a on david-martin:patch-1 into 3519447 on keycloak:master.

[slaskawi]

Thanks @david-martin !

[david-martin]

Wow, thanks for the quick turnaround @slaskawi 👍