What's Changed

• fix: correct the FQDN in configmap when using lwp or ccp by @NoobDream2568 in #471
• fix: validate warmup actions and target RBG failures by @googs1025 in #466
• fix: retry unhealthy update targets in Parallel rollouts by @jiangLLM in #470
• test(upgrade): add the v0.8.0 -> current hop to the upgrade e2e suite by @diw-zw in #476
• fix: isolate unhealthy timers by RoleInstanceSet by @cyf-2002 in #479
• chore(api): mark InPlaceOnly update strategy as deprecated by @NoobDream2568 in #472
• chore(ci): bump test-e2e global timeout from 30m to 45m by @cheyang in #485
• test: make restart backoff e2e deterministic by @googs1025 in #489
• doc: add doc best-practice/stress-testing-and-tuning by @diw-zw in #390
• fix: recover stateful rollout stalls by @diw-zw in #484
• fix: retain RoleInstance revision selector by @diw-zw in #495
• test(ci): split full E2E suite into parallel slices by @diw-zw in #494
• feat: enhance CustomizedAction execution reporting by @googs1025 in #488
• fix: do not claim a child workload the RBG does not control by @NoobDream2568 in #487
• fix: require active owned Pods for RoleInstance readiness by @riverzhang in #497
• perf: throttle stateful instance health pruning by @QIW9 in #501
• perf: scope instance health observations to each set by @cyf-2002 in #504
• fix(make): stop the default E2E label filter from splitting test-e2e by @cheyang in #506
• chore(rbgs): update Helm chart and default image versions to 0.9.0-alpha.2 by @cheyang in #505

New Contributors

• @jiangLLM made their first contribution in #470
• @riverzhang made their first contribution in #497
• @QIW9 made their first contribution in #501

Full Changelog: v0.9.0-alpha.1...v0.9.0-alpha.2

What's Changed

• feat(helm): record installed chart version by @diw-zw in #451
• Update crd-upgrade RBAC by @diw-zw in #456
• Updated Makefile version to 0.9.0 by @diw-zw in #452
• Implement kep-430 by @diw-zw in #434
• fix: align template ref and update strategy validation by @googs1025 in #436
• fix: validate RoleBasedGroup role dependencies at admission by @googs1025 in #447
• doc: add doc best-practice/port-allocation-and-service-discovery by @NoobDream2568 in #402
• docs: correct RoleInstance lifecycle metadata reference by @czczycz in #458
• doc: add doc best-practice/configuring-coordinated-policy by @NoobDream2568 in #400
• fix(api): default CoordinatedPolicy scaling progression to OrderScheduled by @NoobDream2568 in #462
• fix: count scheduled Pods in role replica units by @cyf-2002 in #459
• fix: leave Pod Ready condition to kubelet; controller writes only readiness gates by @NoobDream2568 in #461
• fix(api): heal legacy update strategy types via a mutating webhook by @diw-zw in #463
• test: select active pods in restart policy e2e by @googs1025 in #467
• test(upgrade): assert expected post-upgrade content instead of counting rewrites by @diw-zw in #469
• chore(rbgs): update Helm chart and default image versions to 0.9.0-alpha.1 by @cheyang in #468

New Contributors

• @czczycz made their first contribution in #458
• @cyf-2002 made their first contribution in #459

Full Changelog: v0.8.0...v0.9.0-alpha.1

What's Changed

• docs: add Model Express integration to Dynamo SGLang examples by @Syspretor in #356
• fix(rbg-controller): override role.Replicas from bound RBGSA by @sebest in #306
• fix: coordinatedPolicy rollingUpdate maxSkew stall RoleInstanceSet by @JasonHe-WQ in #308
• feat: add ua for every controller by @vie-serendipity in #364
• docs: update README release notes and add ecosystem projects by @Syspretor in #365
• kep: support inplace scheduling by @NoobDream2568 in #352
• refactor: extract patio sidecar to inference-engine-runtime project by @Syspretor in #366
• docs: add inference-engine-runtime to ecosystem projects in README by @Syspretor in #369
• feat(rbg): support inplace scheduling by @NoobDream2568 in #368
• test: add controller stress test framework with KWOK simulation by @Syspretor in #353
• Updated Makefile version to 0.8.0 by @cheyang in #372
• kep: rbg warmup by @TrafalgarZZZ in #129
• feat: support RoleBasedGroup Warmup (phase 1) by @TrafalgarZZZ in #335
• ci: build controller image from source in manifest e2e test by @cheyang in #373
• ci: run e2e-manifest on every PR and add /release-test workflow by @Syspretor in #374
• test: fix coordinated policy scaling test flake by @Syspretor in #378
• kep: refine warmup API definitions by @NoobDream2568 in #376
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #377
• feat: refactor and implement warmup by @NoobDream2568 in #379
• chore: update controller stress test by @diw-zw in #384
• chore: outdated field name in validation by @diw-zw in #385
• feat: compute maxUnavailable correctly when validating rollout strategy by @vie-serendipity in #382
• fix: handle GracePeriodSeconds not correctly working and respect UpdateStrategy.Type by @NoobDream2568 in #393
• doc: add doc best-practice/deploy-mooncake-store-with-rbg by @diw-zw in #391
• feat: add admission webhook for rbg by @NoobDream2568 in #395
• doc: add doc best-practice/using-role-templates by @diw-zw in #387
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #397
• Remove accidentally uploaded files. by @diw-zw in #398
• chore(rbgs): reorganize Helm chart templates into kubebuilder-style l… by @diw-zw in #403
• test(e2e): verify controller RBAC and webhook-cert bootstrap in a live cluster by @cheyang in #405
• test(e2e): stabilize flaky v1alpha2 specs on slow CI clusters by @diw-zw in #407
• refactor(helm): group values under global/controller/crdUpgrade and u… by @diw-zw in #406
• doc: add doc best-practice/deploy-inference-service by @diw-zw in #386
• doc: add doc best-practice/configuring-rolling-updates by @diw-zw in #388
• feat: add exponential backoff for restart policy by @Syspretor in #394
• feat: add exponential backoff for restart policy with API and docs improvements by @diw-zw in #409
• [WIP] doc: add doc best-practice/configuring-inplace-update-and-scheduling-policies by @NoobDream2568 in #399
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #412
• fix: worker pods get no DNS identity under sharedServiceSelection=All by @diw-zw in #418
• chore: update manifests for v0.8.0-alpha.4 release by @diw-zw in #419
• ci: drop the removed benchmark dashboard image from the nightly build by @diw-zw in #423
• ci: surface /release-test results in the PR merge box by @diw-zw in #420
• e2e: warmup merge multi-role actions - roles may be scheduled onto di… by @diw-zw in #417
• fix: restore restartPolicy as a string to keep v0.7.0 objects decodable by @diw-zw in #424
• feat: restrict v1alpha1 API compatibility RBAC via Helm toggle by @diw-zw in #416
• [WIP] chore: update manifests for v0.8.0-alpha.5 release by @diw-zw in #428
• fix: validate rolling strategy only when replicas gt 0 by @vie-serendipity in #431
• doc: add en doc best-practice/configuring-inplace-update-and-scheduling-policies by @NoobDream2568 in #415
• [WIP] add EPD by @liusy58 in #148
• KEP-430: Gang Scheduling for RBG by @diw-zw in #430
• chore(deps): upgrade dependencies, kubectl v1.36.4, and migrate golangci-lint v2 to fix CVEs by @diw-zw in #429
• Add in-place scheduling injection logs by @googs1025 in #432
• fix: add semantic revision equality check to prevent spurious rollouts by @diw-zw in #433
• fix: reject duplicate roleinstance component names by @googs1025 in #435
• Update crd-upgrade: use v1.37.0 kubectl by @diw-zw in #437
• Fix/upgrade spurious rollout and identity labels by @diw-zw in #439
• fix(e2e): caBundle verify race + apicompat update conflict flake by @cheyang in #443
• e2e: add a v0.7.0 -> current upgrade-compatibility suite to the release gate by @diw-zw in #444
• fix(e2e): gate the upgrade baseline on pods, not the Ready condition by @diw-zw in #446
• chore: update manifests for v0.8.0 release by @cheyang in #445

New Contributors

• @vie-serendipity made their first contribution in #364
• @liusy58 made their first contribution in #148
• @googs1025 made their first contribution in #432

Full Changelog: v0.7.0...v0.8.0

What's Changed

• ci: drop the removed benchmark dashboard image from the nightly build by @diw-zw in #423
• ci: surface /release-test results in the PR merge box by @diw-zw in #420
• e2e: warmup merge multi-role actions - roles may be scheduled onto di… by @diw-zw in #417
• fix: restore restartPolicy as a string to keep v0.7.0 objects decodable by @diw-zw in #424
• feat: restrict v1alpha1 API compatibility RBAC via Helm toggle by @diw-zw in #416
• chore: update manifests for v0.8.0-alpha.5 release by @diw-zw in #428

Full Changelog: v0.8.0-alpha.4...v0.8.0-alpha.5

Deprecated

This release is deprecated due to breaking API changes in restartPolicy.

What's Changed

• fix: worker pods get no DNS identity under sharedServiceSelection=All by @diw-zw in #418
• chore: update manifests for v0.8.0-alpha.4 release by @diw-zw in #419

Full Changelog: v0.8.0-alpha.3...v0.8.0-alpha.4

Deprecated

This release is deprecated due to breaking API changes in restartPolicy.

What's Changed

• Remove accidentally uploaded files. by @diw-zw in #398
• chore(rbgs): reorganize Helm chart templates into kubebuilder-style l… by @diw-zw in #403
• test(e2e): verify controller RBAC and webhook-cert bootstrap in a live cluster by @cheyang in #405
• test(e2e): stabilize flaky v1alpha2 specs on slow CI clusters by @diw-zw in #407
• refactor(helm): group values under global/controller/crdUpgrade and u… by @diw-zw in #406
• doc: add doc best-practice/deploy-inference-service by @diw-zw in #386
• doc: add doc best-practice/configuring-rolling-updates by @diw-zw in #388
• feat: add exponential backoff for restart policy by @Syspretor in #394
• feat: add exponential backoff for restart policy with API and docs improvements by @diw-zw in #409
• doc: add doc best-practice/configuring-inplace-update-and-scheduling-policies by @NoobDream2568 in #399
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #412

Full Changelog: v0.8.0-alpha.2...v0.8.0-alpha.3

What's Changed

• feat: refactor and implement warmup by @NoobDream2568 in #379
• chore: update controller stress test by @diw-zw in #384
• chore: outdated field name in validation by @diw-zw in #385
• feat: compute maxUnavailable correctly when validating rollout strategy by @vie-serendipity in #382
• fix: handle GracePeriodSeconds not correctly working and respect UpdateStrategy.Type by @NoobDream2568 in #393
• doc: add doc best-practice/deploy-mooncake-store-with-rbg by @diw-zw in #391
• feat: add admission webhook for rbg by @NoobDream2568 in #395
• doc: add doc best-practice/using-role-templates by @diw-zw in #387
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #397

Full Changelog: v0.8.0-alpha.1...v0.8.0-alpha.2

What's Changed

• docs: add Model Express integration to Dynamo SGLang examples by @Syspretor in #356
• fix(rbg-controller): override role.Replicas from bound RBGSA by @sebest in #306
• fix: coordinatedPolicy rollingUpdate maxSkew stall RoleInstanceSet by @JasonHe-WQ in #308
• feat: add ua for every controller by @vie-serendipity in #364
• docs: update README release notes and add ecosystem projects by @Syspretor in #365
• kep: support inplace scheduling by @NoobDream2568 in #352
• refactor: extract patio sidecar to inference-engine-runtime project by @Syspretor in #366
• docs: add inference-engine-runtime to ecosystem projects in README by @Syspretor in #369
• feat(rbg): support inplace scheduling by @NoobDream2568 in #368
• test: add controller stress test framework with KWOK simulation by @Syspretor in #353
• Updated Makefile version to 0.8.0 by @cheyang in #372
• kep: rbg warmup by @TrafalgarZZZ in #129
• feat: support RoleBasedGroup Warmup (phase 1) by @TrafalgarZZZ in #335
• ci: build controller image from source in manifest e2e test by @cheyang in #373
• ci: run e2e-manifest on every PR and add /release-test workflow by @Syspretor in #374
• test: fix coordinated policy scaling test flake by @Syspretor in #378
• kep: refine warmup API definitions by @NoobDream2568 in #376
• chore(rbgs): update Helm chart and default image versions to 0.8.0-al… by @cheyang in #377

New Contributors

• @vie-serendipity made their first contribution in #364

Full Changelog: v0.7.0...v0.8.0-alpha.1

What's Changed

• fix(stateful): skip correct unhealth instances by @JasonHe-WQ in #169
• feat(kep): add new kep to refine configmap content and configuration by @JasonHe-WQ in #134
• refactor(rbg): add api v1alpha2 by @Syspretor in #167
• fix(e2e): Fixed sporadic end-to-end (e2e) failures caused by slow pod… by @Syspretor in #170
• refactor(helm): support crd upgrader by @Syspretor in #175
• refactor(rbgs): init rbgs apiVersion v1alpha2 by @Syspretor in #173
• fix(pod group): fix instance workload calculation bug by @JasonHe-WQ in #174
• 修复PD分离部署中，手动减少prefill和decode的replicas的时候，patio无法正确通知router删除路由的问题。 by @zhaoweimiao in #177
• chore(coordinated-policy): init coordinated policy crd with apiversio… by @Syspretor in #183
• chore(rbg): update v1alpha2 constants by @Syspretor in #184
• fix(role-instance): restart policy does not take effect by @Syspretor in #185
• chore(role-instance): init apiversion v1alpha2 for roleInstance and r… by @Syspretor in #186
• chore(scalingadapter): init scalingAdapter apiversion v1alpha2 by @Syspretor in #189
• chore(engine-runtime): init engine runtime profile apiversion v1alpha2 by @Syspretor in #190
• chore(role-instance): implement roleInstance and roleInstanceSet as reconcile version by @Syspretor in #191
• chore(rbg): implement v1alpha2 as storage version by @Syspretor in #188
• fix(ut): fix base cases ut failure by @Syspretor in #192
• refactor(rbg): support customComponentsPatterns in v1alpha2 version by @Syspretor in #193
• feature(rbg): support role instance index env by @Syspretor in #194
• fix(instance): fix updated generation by @JasonHe-WQ in #197
• chore(role-instance): add roleinstance related rbac by @Syspretor in #199
• fix(rbg): fix inject env related ut by @Syspretor in #200
• fix(rbg): fix reconciler related ut by @Syspretor in #201
• chore(coordinated-policies): add rbac for coordinatedpolicies crd by @Syspretor in #202
• Fix(e2e): add crd check by @Syspretor in #198
• fix(rbg): solve the problem of patch race by @Syspretor in #203
• feature(rbg): support gang scheduling in v1alpha2 by @Syspretor in #195
• Feature(role-instanceset): support create pod in parallel by @Syspretor in #204
• chore(role-instanceset): change spec podManagement field type by @Syspretor in #207
• Refactor/unify constants by @Syspretor in #208
• chore(makefile): support to generate rbac in make manifests by @Syspretor in #209
• refactor(coordinated-policy): change strategies api field by @Syspretor in #205
• refactor(rbgs): change spec api field by @Syspretor in #206
• Ensure the inference service can be correctly re-register after the router is restarted through patio heartbeat by @zhaoweimiao in #187
• fix(env): fix missing env by @JasonHe-WQ in #214
• Change version of Makefile to 0.7.0 by @cheyang in #215
• fix(coordinated-policy): fix progression filed enum validation by @Syspretor in #217
• fix(rbg): fix rbg delete failed in foreground deletion by @Syspretor in #219
• [PATCH] feat: add arm64 multi-arch docker publishing by @cheyang in #218
• fix: generated manifests missing latest CRD by @Syspretor in #222
• fix(role-instance): fix update role instance status conflict by @Syspretor in #220
• Build: update rbg helm chart 0.7.0-alpha.1 by @cheyang in #224
• chore: add copyright 2026 by @Syspretor in #229
• Check copyright by @cheyang in #230
• fix(rbgs): missing re-generate client code for rbgs by @Syspretor in #236
• chore: add check for generated client-code by @Syspretor in #237
• refactor(coordinated-policy): add name field in policy by @Syspretor in #235
• chore: add check for yaml lint by @Syspretor in #238
• Controllerrevisions default by @JasonHe-WQ in #234
• update aiconfigurator dependency by @diw-zw in #228
• Update xxx-copyright-to-pr-go-files.sh - ignore vendor/ by @diw-zw in #239
• Cli implement by @diw-zw in #182
• fix: some typo and wrong kubebuilder comments by @Syspretor in #244
• fix(rbgsa-controller): use time.Second for RequeueAfter durations by @sebest in #240
• feat(scheduler): inherit podGroup annotations by @JasonHe-WQ in #233
• chore: update copyright header by @Syspretor in #245
• fix(rbg-controller): watch owned RoleBasedGroupScalingAdapter resources by @sebest in #241
• feat: pod port allocator (kep 171) by @NoobDream2568 in #210
• fix(port-allocator): add startup args to charts by @NoobDream2568 in #249
• feat(rbg): support enabling portAllocator on demand by @Syspretor in #250
• Build: update rbg helm chart 0.7.0-alpha.2 by @cheyang in #253
• Chore(rbg): add v1alpha2 related examples by @Syspretor in #231
• chore: use ARCH and remove TARGETARCH fallback by @NoobDream2568 in #256
• Security issue: unsafe json string in pulling job's annotations; unsa… by @diw-zw in #251
• chore(community): add examples for dynamo by @Syspretor in #255
• build: support patio arm64 image by @TrafalgarZZZ in #258
• Deal with codeQL alert: Uncontrolled data used in path expression by @diw-zw in #257
• feat(scaling-adapter): add Labels field for user-specified RBGSA labels by @sebest in #242
• Enhance service reconciler by @lx1036 in #252
• fix(rbgsa-controller): use Warning event type for failure events by @sebest in #262
• fix(readiness): re-evaluate pod Ready condition in removeNotReadyKey by @sebest in #246
• feat: support user-defined model configuration by @diw-zw in #259
• chore: disable unparam linter by @Syspretor in #266
• feat(rbg): support use templateRef in lwp by @Syspretor in #254
• feat(scaling-adapter): add readyReplicas to RBGSA with sole-writer pattern by @sebest in #243
• v1alpha2 conversion webhook by @diw-zw in #225
• chore: add examples for mooncake integration with v1alpha2 api by @Syspretor in #272
• rbgcli: adds multi-node LLM inference serving support by @diw-zw in #265
• build(deps): bump github.com/moby/spdystream from 0.5.0 to 0.5.1 by @dependabot[bot] in #278
• fix(rbgsa): api conversion failed caused by fields missing by @Syspretor in #283
• [CLI] Secure OSS secrets in cluster, fix engine port args, and add Qwen3.6 model by @diw-zw in #279
• refactor: remove workload field in v1alpha2 by @Syspretor in #281
• fix: deal with injectMetadataSave properly by @diw-zw in #286
• fix(cli): extractRBGStatus wrong by @diw-zw in #290
• chore: add cases for LWP env by @Syspretor in #287
• feat(cli): enhance llm svc run with two new override flags by @diw-zw in #289
• fix(rbgsa-controller): preserve and initialize readyReplicas by @sebest in #280
• fix(roleinstance): nil pointer dereference in in-place update by @sebest in https://githu...

What's Changed

• docs: update README with latest releases and PD disaggregation example by @Syspretor in #292
• feat(cli): allow custom all fields without pre-build model config by @diw-zw in #293
• chore: skip e2e test for cli changes by @Syspretor in #294
• docs: update documentation for v1alpha2 API by @Syspretor in #295
• Chore/update readme by @Syspretor in #297
• docs: update documents for rbg cli by @diw-zw in #296
• chore: update readme by @Syspretor in #298
• chore: skip e2e tests for non-code file changes by @Syspretor in #299
• feat: add --dev flag for local controller debugging by @Syspretor in #288
• docs: fix Deployment Patterns, Ecosystem descriptions and Mooncake fi… by @Syspretor in #300
• chore: rename run-dev to run-local and update developer guide by @Syspretor in #301
• chore(doc): update install ways in readme by @Syspretor in #311
• ci: add manifest-based E2E test for deploy changes by @Syspretor in #312
• chore(doc): update readme for cli by @Syspretor in #313
• ci: use server-side apply for manifest E2E and remove image build by @Syspretor in #314
• feat(cli): get storage/source/engine details by @diw-zw in #303
• feat(cli): add command 'llm svc model-configs' to list builtin/custom… by @diw-zw in #318
• Fix helm generate issue by @cheyang in #305
• fix(e2e): enable portallocator in e2e-manifest by @Syspretor in #321
• fix(controller): prevent RestartInProgress condition loss due to stal… by @Syspretor in #324
• fix(e2e): ensure manifests contain conversion webhook configuration by @Syspretor in #323
• fix: prevent RestartInProgress condition loss during RBG restart by @Syspretor in #325
• feat(cli): auto-benchmark ui by @diw-zw in #320
• feat(cli): new command - auto-benchmark by @diw-zw in #319
• feat(cli): enhancements to the auto-benchmark search algorithms by @diw-zw in #327
• feat(ui): enhances the benchmark viewer UI according to #327 by @diw-zw in #328
• cli: add --memory and --cpu flag to 'llm model pull' by @diw-zw in #326
• Remove cli by @diw-zw in #330
• fix(rbg): handle inactive pod states (Evicted/Failed) for RBG by @Syspretor in #329
• chore(rbg): deprecate and remove RecreateRBGOnPodRestart restart policy by @Syspretor in #340
• fix(rbg): handle inactive pod recreation and clean up dead code by @Syspretor in #337
• fix(rbg): use status.Replicas instead of spec.Replicas for RBG role s… by @Syspretor in #338
• fix(controller): fix stateful RoleInstanceSet surge rolling update by @NoobDream2568 in #333
• fix(controller): clean up stale RoleInstances on stateful/stateless pattern switch by @NoobDream2568 in #334
• feat: change default restartPolicy to RecreateRoleInstanceOnPodRestart by @Syspretor in #331
• fix(e2e): use RetryOnConflict for LWP update in revision test by @Syspretor in #344
• feat(rbg): enhanced customComponentsPatten lifcycle management by @Syspretor in #316
• feat(rbg): add leader only service by @JasonHe-WQ in #273
• feat: enhance networking to support per replica service by @JasonHe-WQ in #276
• feat: add annotation to exclude components from triggering restart po… by @Syspretor in #332
• feat: add kube-api-qps and kube-api-burst flags for API client tuning by @Syspretor in #347
• feat: add --enable-pprof flag for runtime profiling by @Syspretor in #346
• test(e2e): add v1alpha2 stability and update coverage by @Syspretor in #345
• chore(rbgs): update Helm chart and default image versions to 0.7.0-al… by @cheyang in #350

Full Changelog: v0.7.0-alpha.3...v0.7.0-alpha.4