-
Bug
-
Resolution: Duplicate
-
Major
-
None
-
4.18.z
-
Quality / Stability / Reliability
-
False
-
-
None
-
Critical
-
None
-
None
-
None
-
None
-
CORENET Sprint 277
-
1
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Description of problem: After upgrading to the cluster to 4.18.22 table nat list entries are not getting modified in event of pod deletion of replicaset.
Version-Release number of selected component (if applicable): 4.18.22
How reproducible: Upgrade the cluster from 4.17.X to 4.18.18 to 4.18.22
Steps to Reproduce:
1. Install cluster with 4.17.XX
2. Create 5 test project with sample deployment with random replicas across the project.
3. Upgrade the cluster to 4.18.18 (basically any version lesser than 4.18.22) then you will hit the known bug of stale entries. In order to resolve the same perform db rebuild.
4. Once Egress communication is stabilized upgrade cluster to 4.18.22
5. After upgrade Rollout restart of ovnkube-node pod once all pods are stabilize then verify the entries of available cluster.
6. Post that delete the replicas from existing deployment and you will see no new entries will get created and the new pod which get created is having issue while making communication due missing entries of new pod in table nat list inside ovn-db.
7. Also, It is observed that pod which is getting deleted no new entries for pod getting generated in ovn-db
Actual results: Communication is getting break and pod which is re-created is having communication issue because of missing nat entries in db.
Expected results: Entries should updated and deleted accordingly if there is and event of re-creation of pod. No communication issue should happen.
Additional info:
Scenario 1 : Where POD is deleted from existing replicas and new pod is created. For old pod entry still their and no entry created for newly re-created pod , for output refer comments.
Scenario 2 : No new pod entries are getting created if we delete the pod from existing replicas.
- duplicates
-
OCPBUGS-59371 [4.18] OCP 4.18.15 - EgressIP appears to remove and re-allocate endpoints periodically leading to packet loss (no egressIP migration between hosts observed)
-
- Closed
-