feat(nodeScaleDownTime): add a new metric to track unprocessed nodes during scaleDown #8614
Conversation
|
Adding the "do-not-merge/release-note-label-needed" label because no release-note block was detected, please follow our release note process to remove it. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
k8s-ci-robot
added
do-not-merge/work-in-progress
|
|
|
Welcome @shaikenov! |
k8s-ci-robot
added
cncf-cla: no
|
Hi @shaikenov. Thanks for your PR. I'm waiting for a kubernetes member to verify that this patch is reasonable to test. If it is, they should reply with Once the patch is verified, the new status will be reflected by the I understand the commands that are listed here. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
k8s-ci-robot
added
size/L
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: kada2004, shaikenov The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
… nodes during scaleDown
shaikenov
force-pushed
the
shaikenov-scaledown-unprocessed-node-tracking
branch
from
b9e7969 to
86a98d1
Compare
| } | ||
|
|
||
| type longestNodeScaleDownTime struct { | ||
| defaultTime time.Time |
There was a problem hiding this comment.
Let's maybe rename it to sth more meaningful: evalTime, lastEvalTime, ...?
| l.nodeNamesWithTimeStamps = make(map[string]time.Time) | ||
| metrics.ObserveLongestNodeScaleDownTime(0) |
There was a problem hiding this comment.
You should report the biggest time (if there were any carry over from the previous loop) before clearing it.
Also l.defaultTime should be updated here.
Please add a unit test for this scenario.
| nodes := make([]*apiv1.Node, tc.nodes) | ||
| for i := 0; i < tc.nodes; i++ { | ||
| nodes[i] = BuildTestNode(fmt.Sprintf("n%d", i), 1000, 10) | ||
| } | ||
| provider := testprovider.NewTestCloudProviderBuilder().Build() | ||
| provider.AddNodeGroup("ng1", 0, 0, 0) | ||
| for _, node := range nodes { | ||
| provider.AddNode("ng1", node) | ||
| } | ||
| context, err := NewScaleTestAutoscalingContext(config.AutoscalingOptions{ | ||
| NodeGroupDefaults: config.NodeGroupAutoscalingOptions{ | ||
| ScaleDownUnneededTime: 1 * time.Minute, | ||
| }, | ||
| ScaleDownSimulationTimeout: 1 * time.Hour, | ||
| MaxScaleDownParallelism: 10, | ||
| LongestNodeScaleDownTimeTrackerEnabled: true, | ||
| }, &fake.Clientset{}, nil, provider, nil, nil) | ||
| assert.NoError(t, err) | ||
| clustersnapshot.InitializeClusterSnapshotOrDie(t, context.ClusterSnapshot, nodes, nil) | ||
| deleteOptions := options.NodeDeleteOptions{} | ||
| p := New(&context, processorstest.NewTestProcessors(&context), deleteOptions, nil) |
There was a problem hiding this comment.
Wouldn't it be easier to just create instance with newLongestNodeScaleDownTime()?
| var timestamp time.Time | ||
|
|
||
| // take nodes "n0" - "n3" as unneeded | ||
| unneededNodeNames := tc.unneededNodes1 |
There was a problem hiding this comment.
Why do you need unneededNodeNames variable?
| // if a node is already in nodeNamesWithTimeStamps copy the last value | ||
| valueFromPrevIter := l.get(nodeName) | ||
| newNodes[nodeName] = valueFromPrevIter | ||
| if minimumTime.Compare(valueFromPrevIter) > 0 { |
There was a problem hiding this comment.
Nit: From what I see Compare() is not used here. Please use Before() or After().
| return | ||
| } | ||
| newNodes := make(map[string]time.Time, len(nodeNames)) | ||
| l.defaultTime = currentTime |
There was a problem hiding this comment.
I think you override l.defaultTime too early.
Let's imagine very simple scenario with one node:
| time | unneeded | reportedTime | nodeNamesWithTimeStamps |
|---|---|---|---|
| t | [] | 0 | |
| t+1 | [n1] | 1 | n1 : t |
| t+2 | [n1] | 2 | n1 : t |
| t+3 | [] | 3 | |
| t+4 | [] | 0 |
Seems like with your implementation you would at t +1 save n1: t+1 and report 0.
| Name: "longest_scale_down_eval", | ||
| Help: "Longest node evaluation time during ScaleDown.", |
There was a problem hiding this comment.
Please add a unit as suffix (eg. like function_duration_seconds).
Have you checked what buckets are used by default and if they are fine for this use case?
What type of PR is this?
/kind feature
What this PR does / why we need it:
This PR adds a new metric to the exported prometheus metrics list: LongestNodeScaleDownTime
We want to track all the nodes that were marked as unneeded, but were unprocessed during the ScaleDown. If a node was unneeded, but unprocessed multiple times consecutively, we store only the earliest time it happened. The difference between the current time and the earliest time among all unprocessed nodes will give the longest time. This time can give us an indication of possible throttling and helps to better monitor what happens during ScaleDown.
Which issue(s) this PR fixes:
None
Special notes for your reviewer: this is a draft PR, work is still in progress
Does this PR introduce a user-facing change?
Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:
None