Description of problem:
OLM's catalog operator makes excessive and unnecessary GRPC API requests to catalog source pods every time a subscription is reconciled. The observable outcome is that catalog pods have very high CPU usage, and depending on the number of subscriptions and the state of the system, the high CPU usage may be sustained permanently.
Version-Release number of selected component (if applicable):
4.15
How reproducible:
100%
Steps to Reproduce:
1. Get credentials for a 4.15 or higher cluster
2. Install an operator via a subscription against any catalog source
3. (On 4.17 or higher only) Check the logs of the catalog pod that serves the catalog, and see that it is receiving many ListBundles and GetPackage requests in a short period of time, especially when a subscription is performing an installation or upgrade.
Actual results:
~20 ListBundles calls performed while installing a single operator via a subscription
Expected results:
1 ListBundles call performed while installing a single operator via a subscription (other subscription reconciles should use a cached result until the cache is invalidated)
Additional info:
- is cloned by
-
OCPBUGS-48644 OLMv0: excessive catalog source snapshots cause severe performance regression [openshift-4.18.z]
-
- Closed
-
- is depended on by
-
-
- Closed
-
- is triggering
-
OPRUN-3968 Corrective Measure for OCPBUGS-48468: OLMv0: excessive catalog source snapshots cause severe performance regression
-
- Closed
-
- relates to
-
-
- New
-
- links to
-
RHEA-2024:11038
OpenShift Container Platform 4.19.z bug fix update
