v0.19.0-rc0

What's Changed

• fix(release): improve release orchestrator with safety and automation fixes by @Jooho in #5423
• feat(llmisvc): propagate spec.labels and annotations to service by @neelabhsinha in #5365
• fix: remove incorrect crds in llmisvc-crd by @Jooho in #5428
• fix: normalize legacy deploymentMode in status by @jlost in #5427
• release: prepare release v0.18.0-rc1 by @Jooho in #5434
• fix: skip validation on update during deletion for all resource types by @jlost in #5420
• fix(llmisvc): allow stop when LLMInferenceServiceConfig is missing by @pierDipi in #5413
• fix: add approvers to reviewers in OWNERS by @Jooho in #5444
• fix(samples): add active-request-scorer and fix pluginRef typo by @dagrayvid in #5411
• ci: add --maxfail=10 to e2e pytest invocation by @maskarb in #5445
• feat(llmisvc): add heterogeneous GPUs load balancing sample by @pierDipi in #5374
• ci: add quick install e2e test for KServe + LocalModel + LLMISvc by @Jooho in #5429
• chore: change docker.io image to quay.io for seaweedfs by @Jooho in #5421
• fix: add timeout_keep_alive to work with Istio/Envoy by @yuzisun in #5448
• feat(llmisvc): add e2e autoscaling tests and CI for WVA scaling by @vivekk16 in #5407
• fix: de-hardcode torch wheels; upgrade torch/vision by @jlost in #5396
• fix(ci): add MEMBER association to cherry-pick workflow permissions by @Jooho in #5449
• feat(llmisvc): add migration logic for llm-d v0.6 components upgrade by @vivekk16 in #5433
• ci: add stale bot to warn 3 times before closing inactive issues and PRs by @spolti in #5395
• fix: release orchestrator rc1 improvements by @Jooho in #5460
• build: extend GOTAGS pass-through to the main controller Dockerfile by @bartoszmajsak in #5402
• feat(isvc): add dual-protocol (REST/gRPC) routing for Standard mode by @sivanantha321 in #5451
• fix(helm): align imagePullPolicy defaults with kustomize base by @Jooho in #5459
• release: prepare release v0.18.0 by @Jooho in #5461
• fix: oci support for storageUris (plural) by @brucearctor in #5261
• fix(llmisvc): prevent silent metric loss on non-inline config by @vivekk16 in #5471
• fix: upgrade the huggingface server base images to 2025.10, CUDA to 13.0.0 by @cjohannsen-cloudera in #5206
• fix(samples): use sha256_cbor for precise-prefix kv-cache routing by @KillianGolds in #5484
• fix: correct pyasn1 package name and migrate alibi explainer to uv by @spolti in #5463
• fix(ci): require OWNERS approver review before auto-merge by @spolti in #5495
• fix: huggingface-cpu image - torch version by @spolti in #5494
• chore: promote two reviewers to approvers and add new reviewer by @Jooho in #5504
• fix(e2e): disable vLLM V1 engine for CPU e2e tests by @cjohannsen-cloudera in #5509
• fix(localmodel): use NodeSelector for jobs to fix PVC access by @bartoszmajsak in #5502
• feat: add LocalModelCache support for LLMInferenceService by @VedantMahabaleshwarkar in #5318
• fix(ci): upgrade github-script to v8 and harden auto-merge checks by @spolti in #5511
• fix(crd): enable conversion webhooks in minimal installs by @bartoszmajsak in #5416
• feat(e2e/llmisvc): configurable test environment and new endpoint coverage by @bartoszmajsak in #5187
• fix: gate ClusterServingRuntime on CRD availability by @bartoszmajsak in #5392
• fix: improve checkpoint system consistency in release orchestrator by @Jooho in #5483
• fix(ingress): eliminate ExternalName service no-op update churn by @bartoszmajsak in #5512
• fix(cabundle): eliminate ConfigMap no-op update churn by @bartoszmajsak in #5513
• feat(llmisvc): add --scaling flag to kserve-install.sh by @vivekk16 in #5497
• chore(deps): upgrade WVA to v0.7.0 by @vivekk16 in #5498
• fix(ci): stabilize version resolution in precommit by @bartoszmajsak in #5523
• fix(ci): add explicit write permissions for auto-merge job by @spolti in #5514
• test(llmisvc): harden autoscaling e2e with pipeline health checks by @vivekk16 in #5493
• deps: upgrade Envoy AI Gateway to v0.6.0 and Envoy Gateway to v1.7.0 by @cjohannsen-cloudera in #5520
• fix(llmisvc): extend storage migration retry window by @bartoszmajsak in #5405
• feat(llmisvc): emit k8s events on llmisvc readiness transitions by @neelabhsinha in #5437
• feat(llmisvc): report observed routing topology in status by @bartoszmajsak in #5417
• feat(llmisvc): surface specific ConfigNotFound condition by @bartoszmajsak in #5409
• ci: add network-layer parameter support to e2e tests infrastructure by @spolti in #5506
• fix: move tidy after generate in precommit to fix go.sum instability by @font in #5538
• fix: prevent ClusterStorageContainer CRD deletion on helm upgrade by @Jooho in #5539
• feat(llmisvc): report observed workload references in status by @bartoszmajsak in #5414
• fix(router): correct off-by-one in splitter pickupRoute random range by @mwaykole in #5489
• feat(llmisvc): add reconciliation logic for static lora adapters by @andresllh in #5317
• feat(llmisvc): use targeted access-log flag with vLLM version fallback by @KillianGolds in #5507
• fix(charts): use nested if instead for some edge cases by @Jooho in #5542
• fix(CVE): fix vllm setup and pillow CVE by @agoston in #5553
• feat(llmisvc): add preStop hook and up terminationGracePeriod by @andresllh in #5485
• fix(llmisvc): use model-server-protocol-metrics by @pierDipi in #5545
• fix(ci): collect pod logs on fail, tune maxfail by @jlost in #5528
• fix(llmisvc): reduce logging noise in e2e tests by @pierDipi in #5552
• feat(llmisvc): track gateway origin on each discovered address by @bartoszmajsak in #5524
• feat(llmisvc): observed applied configs by @bartoszmajsak in #5418
• ci: add structured E2E test results (JUnit XML + JSON) by @jlost in #5378
• fix(ci): add manual debug path for prow automerge by @spolti in #5562
• feat(llmisvc): bubble up HPA/KEDA scaling status to service conditions by @vivekk16 in #5540
• feat(charts): make llmisvc GIE CRD creation optional by @neelabhsinha in #5544
• fix: pin azure-core>=1.38.0 to address CVE-2026-21226 by @spolti in #5508
• feat(llmisvc): add model name based routing by @pierDipi in #5521
• fix(llmisvc): set workload service appProtocol based on TLS config by @walnutkim in #5533
• feat(llmisvc): migrate non-zero threshold to prefix-based-pd-decider by @vivekk16 in #5560
• fix(llmisvc): wait for pod termination before starting next test by @andresllh in #5568
• chore: add bump-version Copilot agent and release issue template by @Jooho in #5577
• feat(llmisvc): adding vLLM shutdown-timeout by @andresllh in #5496
• release: prepare release v0.19.0-rc0 by @Copilot in #5588

New Contributors

• @neelabhsinha made their first contribution in #5365
• @dagrayvid made their first contribution in #5411
• @brucearctor made their first contribution in #5261
• @font made their first contribution in #5538
• @walnutkim made their first contribution in #5533
• @Copilot made their first contribution in #5588

Full Changelog: v0.18.0...v0.19.0-rc0