|
| 1 | +# KEP-257: SubGroup Leader Only |
| 2 | + |
| 3 | +<!-- |
| 4 | +This is the title of your KEP. Keep it short, simple, and descriptive. A good |
| 5 | +title can help communicate what the KEP is and should be considered as part of |
| 6 | +any review. |
| 7 | +--> |
| 8 | + |
| 9 | +<!-- |
| 10 | +A table of contents is helpful for quickly jumping to sections of a KEP and for |
| 11 | +highlighting any additional information provided beyond the standard KEP |
| 12 | +template. |
| 13 | +
|
| 14 | +Ensure the TOC is wrapped with |
| 15 | + <code><!-- toc --&rt;<!-- /toc --&rt;</code> |
| 16 | +tags, and then generate with `hack/update-toc.sh`. |
| 17 | +--> |
| 18 | + |
| 19 | +<!-- toc --> |
| 20 | +- [Summary](#summary) |
| 21 | +- [Motivation](#motivation) |
| 22 | + - [Goals](#goals) |
| 23 | + - [Non-Goals](#non-goals) |
| 24 | +- [Proposal](#proposal) |
| 25 | + - [User Stories (Optional)](#user-stories-optional) |
| 26 | + - [Story 1](#story-1) |
| 27 | + - [Notes/Constraints/Caveats (Optional)](#notesconstraintscaveats-optional) |
| 28 | + - [Risks and Mitigations](#risks-and-mitigations) |
| 29 | +- [Design Details](#design-details) |
| 30 | + - [Subgroup Creation](#subgroup-creation) |
| 31 | + - [TPU Environment Injection](#tpu-environment-injection) |
| 32 | + - [Test Plan](#test-plan) |
| 33 | + - [Prerequisite testing updates](#prerequisite-testing-updates) |
| 34 | + - [Unit tests](#unit-tests) |
| 35 | + - [Integration tests](#integration-tests) |
| 36 | + - [e2e tests](#e2e-tests) |
| 37 | + - [Graduation Criteria](#graduation-criteria) |
| 38 | +- [Implementation History](#implementation-history) |
| 39 | +- [Drawbacks](#drawbacks) |
| 40 | +- [Alternatives](#alternatives) |
| 41 | +<!-- /toc --> |
| 42 | + |
| 43 | +## Summary |
| 44 | + |
| 45 | +<!-- |
| 46 | +This section is incredibly important for producing high-quality, user-focused |
| 47 | +documentation such as release notes or a development roadmap. It should be |
| 48 | +possible to collect this information before implementation begins, in order to |
| 49 | +avoid requiring implementors to split their attention between writing release |
| 50 | +notes and implementing the feature itself. KEP editors and SIG Docs |
| 51 | +should help to ensure that the tone and content of the `Summary` section is |
| 52 | +useful for a wide audience. |
| 53 | +
|
| 54 | +A good summary is probably at least a paragraph in length. |
| 55 | +
|
| 56 | +Both in this section and below, follow the guidelines of the [documentation |
| 57 | +style guide]. In particular, wrap lines to a reasonable length, to make it |
| 58 | +easier for reviewers to cite specific portions, and to minimize diff churn on |
| 59 | +updates. |
| 60 | +
|
| 61 | +[documentation style guide]: https://github.com/kubernetes/community/blob/master/contributors/guide/style-guide.md |
| 62 | +--> |
| 63 | +This KEP aims to extend the SubGroup feature by adding an API field that defines the type of Subgroup that will be created. The existing version |
| 64 | +will become "LeaderWorker", and a new type which will exclude the leader from any subgroup. |
| 65 | + |
| 66 | + |
| 67 | +## Motivation |
| 68 | + |
| 69 | +<!-- |
| 70 | +This section is for explicitly listing the motivation, goals, and non-goals of |
| 71 | +this KEP. Describe why the change is important and the benefits to users. The |
| 72 | +motivation section can optionally provide links to [experience reports] to |
| 73 | +demonstrate the interest in a KEP within the wider Kubernetes community. |
| 74 | +
|
| 75 | +[experience reports]: https://github.com/golang/go/wiki/ExperienceReports |
| 76 | +--> |
| 77 | + |
| 78 | +There are cases where the resource requirements for the leader are different from the ones by the worker, while still being able to guarantee exclusive placement |
| 79 | +on the workers. Having a `LeaderExcluded` subgroup means that the placement of the leader will be independent of the workers, and by adding the `subgroup-exclusive-topology`, the |
| 80 | +workers can be guaranteed to be scheduled on the same topology. |
| 81 | + |
| 82 | +### Goals |
| 83 | + |
| 84 | +<!-- |
| 85 | +List the specific goals of the KEP. What is it trying to achieve? How will we |
| 86 | +know that this has succeeded? |
| 87 | +--> |
| 88 | + |
| 89 | +Add an option exclude the leader from any subgroup, while still being able to create subgroups on the other workers. Therefore, it should be possible to do the following split: |
| 90 | +leader, (worker-1, worker-2), (worker-3, worker-4). If the desired effect is to have just one subgroup for the workers, subgroupSize should be set |
| 91 | +to the number of workers (so size - 1). |
| 92 | + |
| 93 | +### Non-Goals |
| 94 | +This KEP assumes that the leader will not request accelerator resources, and thus, subGroupSize will always be assumed to be an odd number. |
| 95 | + |
| 96 | +<!-- |
| 97 | +What is out of scope for this KEP? Listing non-goals helps to focus discussion |
| 98 | +and make progress. |
| 99 | +--> |
| 100 | + |
| 101 | +## Proposal |
| 102 | + |
| 103 | +<!-- |
| 104 | +This is where we get down to the specifics of what the proposal actually is. |
| 105 | +This should have enough detail that reviewers can understand exactly what |
| 106 | +you're proposing, but should not include things like API designs or |
| 107 | +implementation. What is the desired outcome and how do we measure success?. |
| 108 | +The "Design Details" section below is for the real |
| 109 | +nitty-gritty. |
| 110 | +--> |
| 111 | + |
| 112 | +### User Stories (Optional) |
| 113 | + |
| 114 | +<!-- |
| 115 | +Detail the things that people will be able to do if this KEP is implemented. |
| 116 | +Include as much detail as possible so that people can understand the "how" of |
| 117 | +the system. The goal here is to make this feel real for users without getting |
| 118 | +bogged down. |
| 119 | +--> |
| 120 | + |
| 121 | +#### Story 1 |
| 122 | +As a user, I should be able to create a subGroup that only includes the leader in order to be able to schedule the leader with different resource requirements |
| 123 | +still keeping exclusive topology on the workers. |
| 124 | + |
| 125 | +### Notes/Constraints/Caveats (Optional) |
| 126 | +As mentioned in Non-Goals, this KEP assumes that the leader does not request accelerator resources. |
| 127 | + |
| 128 | +<!-- |
| 129 | +What are the caveats to the proposal? |
| 130 | +What are some important details that didn't come across above? |
| 131 | +Go in to as much detail as necessary here. |
| 132 | +This might be a good place to talk about core concepts and how they relate. |
| 133 | +--> |
| 134 | + |
| 135 | +### Risks and Mitigations |
| 136 | + |
| 137 | +<!-- |
| 138 | +What are the risks of this proposal, and how do we mitigate? Think broadly. |
| 139 | +For example, consider both security and how this will impact the larger |
| 140 | +Kubernetes ecosystem. |
| 141 | +
|
| 142 | +How will security be reviewed, and by whom? |
| 143 | +
|
| 144 | +How will UX be reviewed, and by whom? |
| 145 | +
|
| 146 | +Consider including folks who also work outside the SIG or subproject. |
| 147 | +--> |
| 148 | + |
| 149 | +## Design Details |
| 150 | + |
| 151 | +<!-- |
| 152 | +This section should contain enough information that the specifics of your |
| 153 | +change are understandable. This may include API specs (though not always |
| 154 | +required) or even code snippets. If there's any ambiguity about HOW your |
| 155 | +proposal will be implemented, this is the place to discuss them. |
| 156 | +--> |
| 157 | + |
| 158 | +A new API field will be added under the SubGroupPolicy type. This API field will determine what type |
| 159 | +of subgroup is desired. The default is LeaderWorker, which is the existing implementation of subgroup. `SubGroupSize` |
| 160 | +will continue to be a required field, while `Type` will be optional. |
| 161 | + |
| 162 | +```golang |
| 163 | +type SubGroupPolicy struct { |
| 164 | + |
| 165 | + // +kubebuilder:validation:Enum={LeaderWorker,LeaderExcluded} |
| 166 | + // +kubebuilder:default=LeaderWorker |
| 167 | + Type SubGroupPolicyType `json:"subGroupPolicyType,omitempty"` |
| 168 | + |
| 169 | + SubGroupSize *int32 `json:"subGroupSize,omitempty"` |
| 170 | +} |
| 171 | + |
| 172 | +type SubGroupPolicyType string |
| 173 | + |
| 174 | +const ( |
| 175 | + SubGroupPolicyLeaderWorker SubGroupPolicyType = "LeaderWorker" |
| 176 | + |
| 177 | + SubGroupPolicyLeaderExcluded SubGroupPolicyType = "LeaderExcluded" |
| 178 | +) |
| 179 | +``` |
| 180 | + |
| 181 | +A new annotation will be created to determine whether or not the subgroup type is `LeaderExcluded` |
| 182 | + |
| 183 | +* `leaderworkerset.sigs.k8s.io/subgroup-policy-type` |
| 184 | + |
| 185 | +This annotation will only be injected in the leader pod. |
| 186 | + |
| 187 | +In order to keep backwards compatability, it will only be added if the type is `LeaderExcluded`. |
| 188 | + |
| 189 | +### Subgroup Creation |
| 190 | +Implementation wise, the only change needed is to not add the SubGroup labels on the leader if the SubGroupType is `LeaderExcluded`. Effectively, this means |
| 191 | +that the leader is not part of a subgroup at all. The only point of a pod being in a subgroup is to guarantee exclusive placement with the other pods |
| 192 | +in the subgroup. However, since this is a one pod subgroup, there is no use case for injecting the subgroup labels on the leader. |
| 193 | + |
| 194 | + |
| 195 | +```golang |
| 196 | +func (p *PodWebhook) Default() { |
| 197 | + if foundSubGroupSize && pod.Labels[leaderworkerset.SubGroupIndexLabelKey] == "" && !leaderOnlySubGroup { |
| 198 | + pod.Labels[leaderworkerset.SubGroupIndexLabelKey] = "0" |
| 199 | + subGroupUniqueKey := genGroupUniqueKey(pod.Name, "0") |
| 200 | + pod.Labels[leaderworkerset.SubGroupUniqueHashLabelKey] = subGroupUniqueKey |
| 201 | + if subEpKey, foundSubEpKey := pod.Annotations[leaderworkerset.SubGroupExclusiveKeyAnnotationKey]; foundSubEpKey { |
| 202 | + SetExclusiveAffinities(pod, subGroupUniqueKey, subEpKey, leaderworkerset.SubGroupUniqueHashLabelKey) |
| 203 | + } |
| 204 | + } |
| 205 | +} |
| 206 | +``` |
| 207 | + |
| 208 | +### TPU Environment Injection |
| 209 | +When injecting TPU environment variables, we will keep the same logic that exists for when the leader doesn't request TPU resources. |
| 210 | + |
| 211 | +### Test Plan |
| 212 | + |
| 213 | +<!-- |
| 214 | +**Note:** *Not required until targeted at a release.* |
| 215 | +The goal is to ensure that we don't accept enhancements with inadequate testing. |
| 216 | +
|
| 217 | +All code is expected to have adequate tests (eventually with coverage |
| 218 | +expectations). Please adhere to the [Kubernetes testing guidelines][testing-guidelines] |
| 219 | +when drafting this test plan. |
| 220 | +
|
| 221 | +[testing-guidelines]: https://git.k8s.io/community/contributors/devel/sig-testing/testing.md |
| 222 | +--> |
| 223 | + |
| 224 | +[X] I/we understand the owners of the involved components may require updates to |
| 225 | +existing tests to make this code solid enough prior to committing the changes necessary |
| 226 | +to implement this enhancement. |
| 227 | + |
| 228 | +##### Prerequisite testing updates |
| 229 | + |
| 230 | +<!-- |
| 231 | +Based on reviewers feedback describe what additional tests need to be added prior |
| 232 | +implementing this enhancement to ensure the enhancements have also solid foundations. |
| 233 | +--> |
| 234 | + |
| 235 | +##### Unit tests |
| 236 | + |
| 237 | + |
| 238 | +<!-- |
| 239 | +In principle every added code should have complete unit test coverage, so providing |
| 240 | +the exact set of tests will not bring additional value. |
| 241 | +However, if complete unit test coverage is not possible, explain the reason of it |
| 242 | +together with explanation why this is acceptable. |
| 243 | +--> |
| 244 | + |
| 245 | +<!-- |
| 246 | +Additionally, for Alpha try to enumerate the core package you will be touching |
| 247 | +to implement this enhancement and provide the current unit coverage for those |
| 248 | +in the form of: |
| 249 | +- <package>: <date> - <current test coverage> |
| 250 | +The data can be easily read from: |
| 251 | +https://testgrid.k8s.io/sig-testing-canaries#ci-kubernetes-coverage-unit |
| 252 | +
|
| 253 | +This can inform certain test coverage improvements that we want to do before |
| 254 | +extending the production code to implement this enhancement. |
| 255 | +--> |
| 256 | + |
| 257 | +##### Integration tests |
| 258 | + |
| 259 | +<!-- |
| 260 | +Integration tests are contained in k8s.io/kubernetes/test/integration. |
| 261 | +Integration tests allow control of the configuration parameters used to start the binaries under test. |
| 262 | +This is different from e2e tests which do not allow configuration of parameters. |
| 263 | +Doing this allows testing non-default options and multiple different and potentially conflicting command line options. |
| 264 | +--> |
| 265 | + |
| 266 | +<!-- |
| 267 | +This question should be filled when targeting a release. |
| 268 | +For Alpha, describe what tests will be added to ensure proper quality of the enhancement. |
| 269 | +
|
| 270 | +For Beta and GA, add links to added tests together with links to k8s-triage for those tests: |
| 271 | +https://storage.googleapis.com/k8s-triage/index.html |
| 272 | +--> |
| 273 | + |
| 274 | +- <test>: <link to test coverage> |
| 275 | + |
| 276 | +##### e2e tests |
| 277 | + |
| 278 | +<!-- |
| 279 | +This question should be filled when targeting a release. |
| 280 | +For Alpha, describe what tests will be added to ensure proper quality of the enhancement. |
| 281 | +
|
| 282 | +For Beta and GA, add links to added tests together with links to k8s-triage for those tests: |
| 283 | +https://storage.googleapis.com/k8s-triage/index.html |
| 284 | +
|
| 285 | +We expect no non-infra related flakes in the last month as a GA graduation criteria. |
| 286 | +--> |
| 287 | + |
| 288 | +- <test>: <link to test coverage> |
| 289 | + |
| 290 | +### Graduation Criteria |
| 291 | + |
| 292 | +<!-- |
| 293 | +
|
| 294 | +Clearly define what it means for the feature to be implemented and |
| 295 | +considered stable. |
| 296 | +
|
| 297 | +If the feature you are introducing has high complexity, consider adding graduation |
| 298 | +milestones with these graduation criteria: |
| 299 | +- [Maturity levels (`alpha`, `beta`, `stable`)][maturity-levels] |
| 300 | +- [Feature gate][feature gate] lifecycle |
| 301 | +- [Deprecation policy][deprecation-policy] |
| 302 | +
|
| 303 | +[feature gate]: https://git.k8s.io/community/contributors/devel/sig-architecture/feature-gates.md |
| 304 | +[maturity-levels]: https://git.k8s.io/community/contributors/devel/sig-architecture/api_changes.md#alpha-beta-and-stable-versions |
| 305 | +[deprecation-policy]: https://kubernetes.io/docs/reference/using-api/deprecation-policy/ |
| 306 | +--> |
| 307 | + |
| 308 | +## Implementation History |
| 309 | + |
| 310 | +<!-- |
| 311 | +Major milestones in the lifecycle of a KEP should be tracked in this section. |
| 312 | +Major milestones might include: |
| 313 | +- the `Summary` and `Motivation` sections being merged, signaling SIG acceptance |
| 314 | +- the `Proposal` section being merged, signaling agreement on a proposed design |
| 315 | +- the date implementation started |
| 316 | +- the first Kubernetes release where an initial version of the KEP was available |
| 317 | +- the version of Kubernetes where the KEP graduated to general availability |
| 318 | +- when the KEP was retired or superseded |
| 319 | +--> |
| 320 | + |
| 321 | +## Drawbacks |
| 322 | + |
| 323 | +<!-- |
| 324 | +Why should this KEP _not_ be implemented? |
| 325 | +--> |
| 326 | + |
| 327 | +## Alternatives |
| 328 | +Original implementation was to have the leader have the subGroupIndex be 0, and shift all other subGroup indices are moved one to the right. |
| 329 | + |
| 330 | +```golang |
| 331 | +if leaderOnly { |
| 332 | + return fmt.Sprint(((workerIndex - 1) / subGroupSize) + 1) |
| 333 | +} |
| 334 | +``` |
| 335 | + |
| 336 | +This implementation also requires to shift the subGroupIndex back to the left when injecting TPU environment variables |
| 337 | + |
| 338 | +```golang |
| 339 | +// Take worker 1 as an example with subGroupSize set to 2. It will now have subGroupIndex set to 1, which means that the start will be 3, and end will be 4. |
| 340 | +// Need to shift it down by 1 for the calculation to still work out. |
| 341 | +if leaderOnlyType && subGroupIndex != 0 { |
| 342 | + subGroupIndex -= 1 |
| 343 | +} |
| 344 | +start := subGroupSize*subGroupIndex + 1 |
| 345 | +end := subGroupSize * (subGroupIndex + 1) |
| 346 | +``` |
| 347 | + |
| 348 | +Since there is no use case for having the leaderPod on its own subgroup, this implementation was scratched as it adds unnecessary complexity |
| 349 | + |
| 350 | + |
| 351 | +<!-- |
| 352 | +What other approaches did you consider, and why did you rule them out? These do |
| 353 | +not need to be as detailed as the proposal, but should include enough |
| 354 | +information to express the idea and why it was not acceptable. |
| 355 | +--> |
0 commit comments