Issues

• [REFACTOR]: Share one select-and-reserve core between PushRouter and RoutingHost

  dynamo-llmdynamo-runtimelanguage::rustrefactorroutersig-router
  ai-dynamo/dynamo#17880 · Filter by author GuanLuo opened on Nov 2, 2026

• [BUG]: vLLM decode pre-admission rejection leaves prefill KV pinned until NIXL lease expiry

  backend::vllmbugdynamo-runtimefault tolerancenixl
  ai-dynamo/dynamo#17872 · Filter by author nnshah1 opened on Nov 2, 2026

• [BUG]: vLLM disaggregated decode holds slot and KV until lease expiry when client cancels before first token

  backend::vllmbugdynamo-runtimefault tolerancenixl
  ai-dynamo/dynamo#17870 · Filter by author nnshah1 opened on Nov 2, 2026

• Preserve agent routing context through Python chat processors

  agentsenhancementlanguage::pythonlanguage::rustroutersig-agents
  ai-dynamo/dynamo#17867 · Filter by author biswapanda opened on Nov 2, 2026
  

• fix(vllm): preserve self-benchmark prefix semantics on vLLM 0.33

  backend::vllmdynamo-llmlanguage::python
  ai-dynamo/dynamo#17852 · Filter by author tedzhouhk opened on Nov 2, 2026

• [BUG]: SGLang workers stall every stream while checking a long prompt's token IDs

  backend::sglangbuglanguage::pythonperformance
  ai-dynamo/dynamo#17849 · Filter by author jacobhinkle opened on Nov 2, 2026

• [CONTRIBUTION]: Group-aware router queue priority for RL rollouts without ThunderAgent

  contribution-requestdynamo-runtimeenhancementlanguage::rustrlroutersig-rl
  ai-dynamo/dynamo#17847 · Filter by author ShaneGZhu opened on Nov 2, 2026

• Feature the MatrixHub ModelExpress P2P article on the External publications page

  docsdocumentation
  ai-dynamo/dynamo#17842 · Filter by author carlory opened on Nov 2, 2026

• SGLang Python recipe preprocessing should preserve rendered image order

  backend::sglangbugfrontendlanguage::pythonsig-frontend
  ai-dynamo/dynamo#17803 · Filter by author xianlubird opened on Nov 2, 2026

• Remove stale Arc TODO from Context controller

  language::rustruntime
  ai-dynamo/dynamo#17790 · Filter by author xianlubird opened on Nov 2, 2026

• Document ToolCallCompletion parser-state and delivery semantics

  documentationfrontendlanguage::rusttool-calling
  ai-dynamo/dynamo#17787 · Filter by author xianlubird opened on Nov 2, 2026

• Sidecar mode: forward W3C trace context from the vLLM and TensorRT-LLM sidecars to the engine

  backend::trtllmbackend::vllmobservability
  ai-dynamo/dynamo#17780 · Filter by author tmonty12 opened on Nov 1, 2026

• [BUG]: Frontend benchmark report reads incorrect transport metric names

  bugfrontendobservabilitypythonsig-frontend
  ai-dynamo/dynamo#17773 · Filter by author anencore94 opened on Nov 1, 2026

• [BUG]: Planner speculative OP queries use logical counts with the packed AIS FPM interface

  backend::vllmbugdynamo-plannerperformanceplanner
  ai-dynamo/dynamo#17770 · Filter by author JasperChan929 opened on Nov 1, 2026

• [BUG]: Native GLM-4.7/5.x tool arguments are not streamed before closing tags

  frontendlanguage::rusttool-calling
  ai-dynamo/dynamo#17769 · Filter by author wangdongxuking61 opened on Nov 1, 2026

• [CONTRIBUTION]: Honor --tokenizer fastokens in the SGLang chat processor

  backend::sglangenhancementfrontendlanguage::pythonsig-frontend
  ai-dynamo/dynamo#17764 · Filter by author hbikki opened on Nov 1, 2026
  

• [BUG]: decode_video_nvdec returns copies of the last frame (host frames aliased), so H.264/H.265 video_url requests see the wrong frames

  bugmultimodal
  ai-dynamo/dynamo#17762 · Filter by author amrit110 opened on Oct 31, 2026

• [BUG]: vLLM chat processor can lose assistant reasoning history

  backend::vllmbugfrontendlanguage::pythontool-calling
  ai-dynamo/dynamo#17759 · Filter by author KrisLiu16 opened on Oct 31, 2026

• [BUG]: Failed Mooncake prefill advertises an unavailable KV transfer

  backend::vllmbugdynamo-llm
  ai-dynamo/dynamo#17757 · Filter by author wjinxu opened on Oct 31, 2026

• RFC: Programmable KV for SGLang direct linker and Mooncake

  backend::sglangdep:draftenhancementkv-memoryrouter
  ai-dynamo/dynamo#17755 · Filter by author magicYang1573 opened on Oct 31, 2026

• Explicit SGLang Mooncake-to-L1 prefetch control from KvRouter

  backend::sglangenhancementkv-memoryrouter
  ai-dynamo/dynamo#17747 · Filter by author magicYang1573 opened on Oct 31, 2026

• Bound parallel Operator upgrade scenarios within one Kind cluster

  cideployment::k8senhancementoperator
  ai-dynamo/dynamo#17744 · Filter by author xianlubird opened on Oct 31, 2026
  

• feat(vllm): serve audio transcription through the vLLM sidecar

  backend::vllmenhancementFeaturelanguage::rustmultimodal
  ai-dynamo/dynamo#17738 · Filter by author NurullahMorshed opened on Oct 30, 2026

• [BUG]: v2 route dispatches a wrapped <tool_call> quoted inside the thought as a second tool call (residual of #17663)

  bugfrontendsig-frontendtool-calling
  ai-dynamo/dynamo#17732 · Filter by author luckyq opened on Oct 30, 2026
  

• TensorRT-LLM sidecar: engine-loss handling without KV events

  backend::trtllmenhancementlanguage::rust
  ai-dynamo/dynamo#17721 · Filter by author JulienDarve opened on Oct 30, 2026