From 4a454e79240687dbe10ed25a5bf7815d66962c65 Mon Sep 17 00:00:00 2001 From: Eric Weber Date: Tue, 3 Sep 2024 13:37:55 -0500 Subject: [PATCH] test: skeleton for don't orphan processes when node not ready Longhorn 6552 Signed-off-by: Eric Weber --- e2e/tests/regression/test_ha.robot | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/e2e/tests/regression/test_ha.robot b/e2e/tests/regression/test_ha.robot index 15cc9d1e5d..bc2121ace1 100644 --- a/e2e/tests/regression/test_ha.robot +++ b/e2e/tests/regression/test_ha.robot @@ -30,3 +30,18 @@ Disrupt Data Plane Traffic For Less Than Long Engine Replica Timeout And Wait for volume of statefulset 0 attached and degraded And Wait for volume of statefulset 0 healthy And Check statefulset 0 data in file data is intact + +Don't Orphan Processes When Node Not Ready + [Documentation] Don't orphan processes when a node becomes not ready. + ... + ... 1. Create a volume of any size and a number of replicas equal to the number of nodes. + ... 2. Attach the volume to a node. + ... 3. Stop kubelet on the attached node long enough the node to become not ready and its instance + ... manager to have state unknown. + ... NOTE: The timing is important here. The instance manager reaches state unknown after + ... approximately 30 seconds. Approximately 6 minutes later, Kubernetes evicts most pods on the + ... node, including the instance manager pod, so the instance manager transitions to state error. + ... The remaining steps must be executed before this happens. + ... 4. Detach the volume. + ... 5. Verify that no instance-manager CR has a status.instanceEngines[]. + ... 6. Verify that no instance-manager CR has a status.instanceReplicas[].