We want to evaluate (and implement if feasible) a pre-upgrade risk warning for clusters upgrading from OCP 4.21 to 4.22, to alert users who may still be consuming operators from the Marketplace catalog (removed in 4.22).

Context / current limitation:

• Ideal detection would use a Subscription metric label for the subscribed catalog (catalogName), but catalogName is not included in the subscription_sync_total metric.

• Alternative approach is to infer Marketplace usage via the package label and known Marketplace package naming patterns.

Proposed approach (if feasible):

• Add a Cincinnati conditional risk (similar to prior blocked-edges/conditional-risk rules) based on PromQL matching the package label.

• Candidate PromQL:

 
 
count(last_over_time(subscription_sync_total{package=~"stackgres|flux|.*-rhmp"}[10m])) > bool 0 or vector(0)
Risk / accuracy notes:

• .*-rhmp is likely high-confidence for Marketplace, but it is not a hard requirement.

• flux (and potentially stackgres) may overlap with other catalogs, so this can produce false positives.

• Recommended mitigation: split into two rules/messages:

• • High-confidence rule for .*-rhmp

• • Lower-confidence rule for flux (and optionally stackgres) with wording that it may not be Marketplace-specific

References:

• More info: https://redhat-internal.slack.com/archives/C0A895F6PDH/p1768336579932609?thread_ts=1768323190.374519&cid=C0A895F6PDH
• OLM metric discussion (catalogName not present): https://github.com/operator-framework/operator-lifecycle-manager/blob/7890bdca55cc8fa1ce3cfb8997df8c6b81725752/pkg/metrics/metrics.go#L171-L177
• @joe shared: a gist for that promql query I mentioned earlier this week here: https://gist.github.com/joelanford/739e664ef84e16e09758ead894521973 

• Prior Cincinnati conditional risk example: https://github.com/openshift/cincinnati-graph-data/blob/5094c7f52e9b591f743c2719a0.../blocked-edges/4.15.10-OLMOperatorsInFailedState.yaml