In some cases, OLM fails to resolve a replacement chain in seemingly valid catalogs. Operator subscriptions get ResolutionFailed
condition with empty channel name listing recent versions of the operator. Example:

a unique replacement chain within a channel is required to determine the relative order between channel entries, but 2 replacement
chains were found in channel "" of package "devworkspace-operator": devworkspace-operator.v0.15.3, devworkspace-operator.v0.16.0

Once this occurs, subscription is stuck in this state, despite continuing to update the subscription; versions listed alternate
order every few seconds. Issue is quickly resolved by restarting catalog-operator pod.

It's unclear what triggers this, but some details about the cluster:

* The referenced operator is installed from a custom catalog source; the same operator and channel is present in redhat-operators
but has different versions available and a different replaces chain. However, neither of the versions listed above are available in
the redhat-operators catalog.
* The catalog is valid, and installing in a different cluster allows clean updates through all versions.
* Catalog is using file-based catalogs; bundle files available here: https://github.com/devfile/devworkspace-operator/tree/main/olm-catalog/release

reproduced last on OSD 4.11.5

No reproducer, unclear trigger. However, encountered at least twice in the last few months for this specific operator install.

1. Install catalog source using image quay.io/devfile/devworkspace-operator-index:release
2. Install devworkspace-operator from this catalog via subscription
3. (potentially, over time) operator updates will stop

Subscription gets ResolutionFailed condition

a unique replacement chain within a channel is required to determine the relative order between channel entries, but 2 replacement 
chains were found in channel "" of package "devworkspace-operator": devworkspace-operator.v0.15.3, devworkspace-operator.v0.16.0

Similar error is logged by catalog-operator pod. Note channel name is blank and versions shown alternate order every few seconds. 
Versions reflect current releases as of writing but has occurred with previous versions.

No error is expected. There is only one channel, named "fast", and output of opm render shows a clear replacement chain. Fresh 
installs allow upgrades as normal.

Workarounds:
* Restart catalog-operator pod in openshift-operator-lifecycle-manager namespace (delete running pod and let it be recreated)
* Re-install operator