github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-07-23 19:56:01 +00:00
Code Issues Projects Releases Wiki Activity

Wait 15m after instead of before breaking nodes

Browse Source 
15m is enough for Cluster Autoscaler to remove empty nodes, so we need
to break them sooner than that. Instead, wait 15m after breaking them to
ensure Cluster Autoscaler will consider them as unready instead of still
starting.
This commit is contained in:
Daniel Kłobuszewski 2021-08-31 11:37:25 +02:00
parent f610eee161
commit 9808dd9a03
1 changed files with 10 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
test/e2e/autoscaling/cluster_size_autoscaling.go
View File

@ -892,7 +892,14 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
// node to be unhealthy unless it was created more than 15m
// ago. Within that 15m window, it'll assume node is just
// starting and not unhealthy.
time.Sleep(15 * time.Minute)
//
// However, waiting for 15m would allow scale down to kick in
// and remove recently added nodes, so here we just wait 2m for
// nodes to come up (1m should be enough, another 1m added as
// an extra buffer. Then, we break connectivity to a subset of
// nodes and only after that we wait for 15m, since scale down
// shouldn't happen when the cluster is unhealthy.
time.Sleep(2 * time.Minute)
ginkgo.By("Block network connectivity to some nodes to simulate unhealthy cluster")
nodesToBreakCount := int(math.Ceil(math.Max(float64(unhealthyClusterThreshold), 0.5*float64(clusterSize))))
@ -915,7 +922,8 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
} else {
ReserveMemory(f, "memory-reservation", 100, nodeCount*memAllocatableMb, false, defaultTimeout)
defer e2erc.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
time.Sleep(scaleUpTimeout)
// Wait for 15m to ensure Cluster Autoscaler won't consider broken nodes as still starting.
time.Sleep(15 * time.Minute)
currentNodes, err := e2enode.GetReadySchedulableNodes(f.ClientSet)
framework.ExpectNoError(err)
framework.Logf("Currently available nodes: %v, nodes available at the start of test: %v, disabled nodes: %v", len(currentNodes.Items), len(nodes.Items), nodesToBreakCount)