Absorb PR 11's final Para compatibility surface as an opt-in OpenAI serving layer rather than hard-coding business defaults into protocol models. The change adds server args for Para chat defaults, Kimi/GLM compatibility, tool-choice normalization, tool-role text flattening, and streaming first-chunk error preflight while preserving default upstream behavior unless explicitly enabled.
Reasoning token usage is also propagated through chat/completion usage paths, with GLM compatibility emitting completion_tokens_details.reasoning_tokens. Low-risk protocol fixes accept string image_url content parts and preserve GLM function-call argument value whitespace.
Constraint: Online Para-compatible deployments require request/response semantics that differ from default OpenAI serving behavior.
Constraint: Current CP/HiCache/bs>1 work must not be coupled to OpenAI serving compatibility changes.
Rejected: Merge PR 11 history directly | intermediate commits briefly hard-code chat max_tokens=32768 before later gating it by server args.
Rejected: Enable Para compatibility by default | would change non-Para OpenAI-compatible deployments.
Confidence: high
Scope-risk: moderate
Directive: Keep Para-specific serving policies behind explicit server args unless the business contract changes globally.
Tested: PYTHONPATH=python:. python -m unittest discover -s test/registered/unit/entrypoints/openai -p 'test_para_serving_protocol.py' -v (19 tests OK)
Tested: python -m py_compile modified OpenAI serving, tokenizer manager, server_args, function-call detector, and test files
Not-tested: Live router/prefill/decode OpenAI serving E2E after enabling Para flags.
Co-authored-by: OmX <omx@oh-my-codex.dev>
CP shared-KV bs>1 cache-hit loads already merge request load ops, but the host pool still rebuilt layer-invariant mapping work from the same host/device indices. Introduce a PreparedLoadDescriptor lifecycle around begin/end load, wire MLA KV and NSA index H2D loads through tai-kernel prepared submit when available, and add timing hooks plus regression coverage for descriptor reuse and explicit fallback logging. Record the P4/P6b design and benchmark results in the advanced feature notes.
Constraint: Radix residency and allocator decisions remain synchronous; only the data-transfer descriptor is prepared for per-layer async submit.
Constraint: Production fast path must not silently fall back when tai prepared H2D support is missing.
Rejected: Cross-batch descriptor reuse | descriptor lifetime and tensor ownership are only safe within one load operation.
Rejected: Change L2->L1 scheduling to layer-ahead prefetch in this commit | that is a separate lifecycle change after descriptor reuse is stable.
Confidence: medium
Scope-risk: moderate
Directive: Keep LayerDoneCounter per-layer readiness semantics; do not replace with all-layer waits.
Tested: python -m py_compile python/sglang/srt/mem_cache/memory_pool_host.py python/sglang/srt/managers/cache_controller.py
Tested: Remote g0034:cjy-glm5-new PYTHONPATH=python python -m pytest -q test/registered/unit/managers/test_hicache_controller_cp.py (88 passed)
Tested: Remote tai-kernel prepared descriptor CUDA test (6 passed) and P4 benchmark full matrix (90 rows)
Not-tested: ETE replay/GSM8K cache-hit correctness after this commit
Not-tested: Layer-ahead L2->L1 prefetch scheduling
Co-authored-by: OmX <omx@oh-my-codex.dev>
CP shared-KV bs>1 batching was only bounded by request count, extend tokens, and cached tokens. That left temporary GPU buffers such as MLA/index materialization, remap metadata, logits windows, and transfer descriptors implicit, and raw extend-token limits could exceed the active chunked-prefill budget.\n\nThis adds an explicit max-buffer-size admission gate with a CPU-only stream-aware estimator, wires it through PrefillAdder/Scheduler, performs a startup CUDA smoke allocation when configured, and reports the estimate in the scheduler admission benchmark. When chunked prefill is active, the effective CP extend-token limit is capped by the current chunk budget so the CP path does not advertise unreachable batch capacity or lift max-prefill-tokens too far.\n\nConstraint: Admission estimation must stay CPU-only on the scheduler hot path; CUDA allocation is limited to startup smoke checking.\nConstraint: Single oversized requests must still be allowed to run alone to avoid scheduler deadlock.\nRejected: Rely only on --max-prefill-tokens | it does not reliably bound the first oversized request and does not model cache-hit/load-back pressure.\nRejected: Let CP extend limit exceed chunked-prefill size | it creates an unreachable effective capacity and misleading budget lift.\nConfidence: medium\nScope-risk: moderate\nDirective: If bs>1 L1 prefetch is enabled later, update CPSharedKVPrefillBufferEstimatorContext.bs_gt1_l1_prefetch_enabled and include the live prefetch dense buffers in overlap windows.\nTested: local py_compile for touched files\nTested: local PYTHONPATH=python pytest -q test/registered/unit/managers/test_cp_shared_kv_prefill_buffer_estimator.py (4 passed)\nTested: remote cjy-glm5-new targeted pytest for new server_args, PrefillAdder, estimator, and benchmark cases (10 passed)\nTested: remote cjy-glm5-new PYTHONPATH=python pytest -q test/registered/unit/managers/test_cp_shared_kv_prefill_buffer_estimator.py test/registered/unit/managers/test_prefill_adder.py test/registered/unit/managers/test_prefill_scheduler_admission_bench.py (29 passed before chunk cap, then test_prefill_adder.py 21 passed after chunk cap)\nNot-tested: full server_args suite because existing TestPrepareServerArgs tries to reach HuggingFace and fails under container DNS/network\nNot-tested: GLM5 ETE smoke with --cp-shared-kv-prefill-max-buffer-size
Add an offline benchmark that reuses PrefillAdder to model how L1 cached tokens, L2 HiCache hits, and extend tokens shape CP shared-KV prefill batch admission. The tool makes scheduler stop reasons and fake L2 load-back capacity pressure observable without starting a model.
Constraint: The benchmark must stay CPU/offline and avoid depending on CUDA execution or live services.
Constraint: L2 cached tokens are modeled as host_hit_length, so successful load-back both increases prefix_len and consumes fake L1 capacity.
Rejected: Build an ETE benchmark first | too slow for isolating scheduler admission behavior.
Rejected: Reimplement scheduler logic from scratch | would drift from PrefillAdder semantics.
Confidence: high
Scope-risk: narrow
Directive: Treat duration_us as Python admission overhead only; it is not an ETE latency metric.
Tested: Remote pytest test/registered/unit/managers/test_prefill_scheduler_admission_bench.py: 4 passed as part of 6 targeted tests.
Tested: Remote synthetic benchmark run with --cp-max-total-cached-tokens showed second 4096-token cached request stopped with OTHER.
Not-tested: Real traffic trace import from production logs.
Record the staged design for batch-aware L1 shared-KV prefetch under CP + HiCache. The plan separates L2->L1 load two-layer lookahead from L1 prefetch one-layer lookahead, keeps target first and draft/EAGLE out of the first phase, and calls out that current suffix reuse remains synchronous until the request-aware span contract is extended.
Constraint: Existing CP shared-KV correctness failures came from batch-global remap/req-id ambiguity, so bs>1 prefetch must carry explicit per-request spans and ids.
Rejected: Reuse the bs=1 scalar prefix_pages prefetcher for bs>1 | it aliases request rows and cannot represent mixed prefix/current spans.
Rejected: Make draft/EAGLE prefetch part of the first phase | target path must be validated first before coupling draft lifecycle.
Confidence: medium
Scope-risk: narrow
Directive: Do not open bs>1 L1 prefetch until req-id fail-fast and per-request slot span tests cover both MLA and index paths.
Tested: git diff --cached --check
Not-tested: Runtime implementation; this is a design document only.
The CP shared target+draft HiCache budget was still estimating NSA target
bytes per token as if every target layer had an index cache. After index
skip compacts L1 and L2 index buffers to active layers, that stale estimate
kept host/L2 token capacity artificially low even though the physical host
index allocation was already smaller.
Use the pool's index_active_layer_ids when estimating NSA HiCache bytes per
token, with the existing full-layer fallback for pools that do not expose
compact metadata. This keeps the shared target+draft capacity computation
consistent with the actual L2 host layout.
Constraint: Index skip compacts NSA index cache by active logical layers, but MLA KV remains full-layer.
Rejected: Keep full-layer HiCache estimate | preserves correctness but wastes the L2 capacity recovered by compact index allocation.
Confidence: high
Scope-risk: narrow
Directive: Keep HiCache byte estimates aligned with NSATokenToKVPool allocation metadata; otherwise target+draft token budgets will drift from real host memory use.
Tested: Remote container targeted pytest for active-index HiCache estimate and shared-budget tests -> 3 passed.
Tested: Remote container P3-P6 regression suite plus new capacity tests -> 203 passed, 2 subtests passed.
Not-tested: Full ETE replay/GSM8K after this budget correction.
Co-authored-by: OmX <omx@oh-my-codex.dev>
Index skip reduces the number of target layers that own NSA index state,
but PD transfer and HiCache still assumed dense full-layer state buffers.
This change carries explicit state layer IDs through prefill/decode
registration, compacts device and host index buffers to active layers,
and maps logical layer IDs to compact slots on transfer paths.
The PD side fails fast when prefill/decode disagree on NSA state layer
identity instead of silently truncating or copying mismatched buffers.
Host direct tests now use the same CPU-index descriptor contract required
by the TAI cudaMemcpyBatchAsync path, and host registered memory is
unregistered on tensor finalization to avoid stale cudaHostRegister state
across CUDA tests.
Constraint: CP shared-KV with index_topk skip must keep target/draft state identity explicit before compacting buffers
Constraint: Direct HiCache TAI transfer rejects CUDA indices to avoid hidden D2H copies on the control path
Rejected: Keep full-layer L1/L2 index buffers | wastes the memory/bandwidth that index skip is meant to save
Rejected: Infer state buffer order by count only | can silently corrupt cache when active layer sets differ
Confidence: high
Scope-risk: moderate
Directive: Do not compact or reorder NSA state buffers without carrying logical layer IDs through PD registration and validating both sides
Tested: Remote container py_compile for touched runtime files
Tested: Remote container pytest: test_nsa_pool_host_unit.py, test_model_runner_kv_cache_mixin.py, test_cp_shared_kv_transfer_mapping.py, test_pd_state_layer_ids.py, test_cp_per_layer_transfer.py, test_cp_shared_kv_runtime.py -> 200 passed, 2 subtests passed
Not-tested: Full ETE GSM8K/replay after compacted P3-P6 changes
Co-authored-by: OmX <omx@oh-my-codex.dev>
Record the phased design for turning NSA index-topk sharing from a compute-only optimization into HiCache and PD-transfer savings. The plan deliberately separates metadata validation from allocation compaction so prefill/decode state-slot mismatches fail before any bandwidth reduction can corrupt cache contents.
Constraint: CP shared-KV cache corruption previously appeared only on cache-hit paths, so state-layer mapping must be explicit before compact transfer or allocation.
Rejected: Compact L1/L2 index buffers first | without state_layer_ids this can silently map different logical layers to the same transfer slot across prefill and decode.
Confidence: high
Scope-risk: narrow
Directive: Do not skip the state_layer_ids phase before compacting NSA index state buffers.
Tested: Documentation placeholder scan and referenced source-path existence check.
Not-tested: Runtime behavior; this commit is documentation only.
CP shared-KV HiCache transfers are per-layer, so the host layout should match the access pattern instead of forcing page-major strides through every layer. This adds a direct-only layer_page_first layout, routes per-layer KV and NSA index backup/load through the TAI LF<->LPF direct kernels, and keeps storage/page-buffer metadata paths fail-fast until their page-level contract is redesigned.\n\nThe direct controller keeps host indices in caller order for both page_first_direct and layer_page_first because the TAI direct path requires CPU index descriptors and owns descriptor coalescing. All-layer backup intentionally loops over per-layer direct kernels rather than using the sgl-kernel all-layer direct ABI.\n\nConstraint: layer_page_first is currently host-only CP HiCache; storage backends assume page-major contiguous page metadata.\nConstraint: TAI LPF direct kernels require CPU int64 page indices and complete page spans.\nRejected: silently fallback to SM copy when TAI LPF kernels are missing | that hides production performance regressions.\nRejected: support storage page metadata in this commit | LPF requires a layer-page-level storage contract, not a one-pointer-per-page contract.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not enable storage or kernel backend for layer_page_first without redesigning page-buffer metadata and adding remote ETE coverage.\nTested: local py_compile for touched runtime files.\nTested: remote py_compile in g0034 container for touched runtime files.\nTested: remote targeted pytest: 5 passed for parser/storage/layout/move_indices smoke coverage.\nNot-tested: full CP HiCache ETE with --hicache-mem-layout layer_page_first after this commit step.\nNot-tested: combined CUDA roundtrip tests in one pytest process; previous independent runs passed but combined run exposed a host-memory registration lifecycle issue.
CP shared-KV compute padding creates per-request lane slots, so valid rows are not a simple prefix/suffix mask. DeepEP MoE was still seeing dummy rows and using scalar non-padded semantics, which let padding participate in gate/topk and corrupted cache-hit tiny-extend inference.\n\nThe fix compacts CP-local valid rows before MoE dispatch and restores the compact output back to the compute-padded row layout before downstream layer communication. The local GSM8K investigation ledger is now removed from the tracked tree and ignored so future debug notes stay local.\n\nConstraint: CP shared-KV compute-padding layout must keep downstream communicator shapes stable.\nRejected: Disable bs>1/current reuse/cache-hit fast paths | hides the semantic bug and loses the intended performance path.\nRejected: Use num_token_non_padded for MoE under compute padding | valid rows are interleaved with dummy lane slots, not suffix-padded.\nConfidence: high\nScope-risk: moderate\nDirective: Do not feed compute-padded dummy rows into sparse MoE gate/topk; compact valid rows at the MoE boundary and restore shape afterward.\nTested: python -m py_compile python/sglang/srt/layers/attention/nsa/utils.py python/sglang/srt/models/deepseek_v2.py\nTested: remote focused CP utils tests passed, 4 tests.\nTested: remote GSM8K 50-question smoke accuracy 0.960; 200-question runs accuracy 0.955 and 0.965; full 1319-question run accuracy 0.952.\nNot-tested: Long-running production traffic beyond GSM8K after this commit.
The bs>1 warm-cache accuracy investigation produced several negative results, reproducer details, and the final duplicate-logical-page slot-aliasing hypothesis. Keeping the ledger prevents repeated log archaeology after context compaction and documents which probes were removed from runtime code.
Constraint: The document is an investigation ledger, not active runtime instrumentation
Rejected: Drop the tensor-dump history entirely | it explains why later work focused on request-slot remap instead of decode transfer
Confidence: medium
Scope-risk: narrow
Directive: Treat historical dump sections as evidence only; do not infer that dump hooks are still present in runtime code
Tested: Documentation-only; runtime verification covered by the adjacent code/test commits
Not-tested: Rendered documentation formatting outside plain Markdown
Warm-cache GSM8K failures needed request-to-log correlation across scheduler, prefill transfer, and Mooncake CP filtering. The added diagnostics are gated by existing CP shared-KV debug envs and include bounded token signatures plus transfer page summaries so future debugging can identify whether a failed request hit L1, L2, or transfer truncation paths.\n\nThe findings document records the ruled-out hypotheses and the RAGGED current-row contract failure, preventing repeated log archaeology after context compaction.\n\nConstraint: Production hot paths must not emit these logs unless SGLANG_CP_SHARED_KV_BS_GT1_DEBUG or existing CP shared-KV debug gates are enabled.\nRejected: Add a new debug env | reuse of existing bs>1 debug gate avoids more runtime switches.\nRejected: Store full token ids in logs | bounded signatures are enough for correlation without huge logs.\nConfidence: medium\nScope-risk: narrow\nDirective: Keep these diagnostics gated and bounded; do not convert them to unconditional INFO logs.\nTested: Remote container py_compile as part of the RAGGED fix validation.\nNot-tested: Long-running production log-volume impact with debug enabled.
The cache-hit accuracy drop is not isolated to one obvious kernel boundary, so the scheduler/radix/HiCache path needs request-correlated evidence. Add default-off, rate-limited CP shared-KV debug logs for prefix matching, valid-tail flooring, pending backup attach/commit, load-back planning, and unfinished-request cache insertion. The investigation notes capture current GSM8K pass1/pass2 evidence and rejected interpretations to avoid repeated log archaeology.
Constraint: Logs must be default-off and bounded because these paths are scheduler/control hot paths.
Rejected: Always-on INFO tracing | would distort the exact latency and cache-hit behavior under investigation.
Confidence: medium
Scope-risk: moderate
Directive: Remove or keep env-gated only after the GSM8K cache-hit root cause is closed; do not leave unbounded hot-path logs.
Tested: python -m py_compile on changed runtime files.
Not-tested: Local pytest blocked before collection by missing orjson dependency; fresh restarted GSM8K verification still pending.
(cherry picked from commit bfc6d1473dc2a5d72bc3a8d6fca1e2429537be0e)
Cache-hit GSM8K regressions only appeared after the second pass reused request-specific suffix pages, so this change adds fail-fast transfer validation, masks stale rectangular page-table tails, and extends CUDA/unit coverage across FP8 CP shared-KV write, load, top-k, and materialization paths. The temporary ledger records eliminated hypotheses to prevent re-debugging the same L2 and persistent-cache paths.\n\nConstraint: CP shared KV stores physical pages but scheduler-visible semantics must remain valid-token/page-bounded.\nConstraint: bs>1 FP8 prefill must preserve existing CP shared-KV fast paths without silent fallback.\nRejected: Blame raw HiCache L2 load without tests | L2 KV and index backup/load/materialize roundtrips pass on remote CUDA.\nRejected: Disable current/partial reuse broadly | hides the cache-hit contract regression and costs performance.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not weaken CP shared-KV fail-fast or rectangular-tail masking without rerunning second-pass cache-hit accuracy tests.\nTested: remote CUDA pytest for fused FP8 MLA store, fused persistent index store, L2-loaded FP8 KV materialize, L2-loaded index materialize, ragged top-k offset, TAI batched index MQA prepare.\nTested: local py_compile for touched test files and git diff --check.\nNot-tested: full second-pass GSM8K accuracy after these diagnostic tests; root cause remains under investigation.
Repeated GSM8K/cache-hit traffic can drain a CP owner lane while exact
allocation still succeeds. This changes L1/L2 free-room logic so the trigger
accounts for the pending allocation, then evicts a minimum target chunk or the
exact shortage, whichever is larger. CP load-back also performs best-effort
owner-lane free-room eviction before admitting tiny cache hits that would
otherwise skew one lane.
The commit also adds gated bs>1 prefill timing markers around
recv/bootstrap/batch/inflight polling so future hangs expose the scheduler
boundary instead of only surfacing as a watchdog shutdown. The post-fix repeat
GSM8K failure is recorded as an active regression to continue investigating,
not as old-process noise.
Constraint: Free-room policy must reduce repeated eviction and owner-lane starvation without reserving required+target pages after every allocation
Rejected: Evict to required+target availability | wastes L1/L2 residency under many short cache hits
Rejected: Treat free-room misses as fatal on load-back | exact capacity should remain the strict admission condition
Confidence: medium
Scope-risk: moderate
Directive: Do not remove the gated bs>1 timing markers until repeat-GSM8K hangs have a direct failing boundary
Tested: git diff --check
Tested: python -m py_compile for touched Python files
Tested: remote earlier test_cp_shared_kv_layout.py and test_cp_hicache_metadata.py passed 157 tests after this free-room formula
Not-tested: Two full repeat GSM8K runs after this commit; latest repeat still killed prefill and requires follow-up root-cause work
CP shared-KV bs>1 exposed two separate overhead sources: HiCache load-back could thrash near capacity, and partial-current sync compose could all-reduce row-major page-table gaps between request prefixes. Keep the intended batch-plan path for bs=1, but make host/L1 free-room handling less reactive and teach the MLA/index sync compose path to use exact per-request prefix slot spans instead of one bounding span.\n\nThe exact-span path preserves the single-span IPC fast path for bs=1/single-span cases, while avoiding over-communication for heterogeneous cache-hit batches. The HiCache metadata tests cover host/L1 free-room propagation and load-back batching behavior.\n\nConstraint: bs=1 using CPSharedKVBatchPlan is the expected steady-state path and must not be treated as a regression.\nConstraint: Remote production-like validation runs inside g0034 container /sgl-workspace/sglang-tai.\nRejected: Disable batch-plan for bs=1 | user confirmed this is intended behavior and it would hide the actual bs>1 overhead.\nRejected: Keep one bounding prefix span for bs>1 | row-major page tables can include current/gap slots and inflate per-layer all-reduce work.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not replace exact prefix spans with a single row-major bounding span unless ETE data proves collective launch count dominates gap over-communication.\nTested: g0034 docker py_compile for changed Python/test files.\nTested: g0034 docker PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py test/registered/unit/mem_cache/test_cp_hicache_load_back_owner_lanes.py test/registered/unit/mem_cache/test_cp_hicache_metadata.py — 241 passed, 5 warnings, 2 subtests passed.\nNot-tested: Full ETE bs>1 throughput after the exact-prefix-span change.\nNot-tested: CUDA kernel benchmark and live traffic replay.
Burst decode can sit in disaggregation prealloc/transfer queues while health probes still need an immediate scheduler-alive response, and EAGLE prebuilt metadata must outlive the synthetic prebuilt step until the first real decode consume. The transfer path also needs to send final aux metadata even when there are no new KV pages in the final chunk.
This keeps decode metadata slot ownership narrow instead of cloning EAGLE tensors, adds fail-safe cleanup for prebuilt exceptions/finished prebuilt requests, fixes final empty chunk metadata transfer, and records the investigation ledger for future debugging.
Constraint: bs=1 historically worked, so decode compute/sampling behavior must not be changed without direct evidence.
Rejected: Clone EAGLE metadata tensors on transfer commit | avoids lifetime issues but adds hot-path CPU/GPU memory traffic.
Rejected: Treat prefill AbortReq logs as root cause | current evidence shows they can be downstream of decode/router aborts.
Confidence: medium
Scope-risk: moderate
Directive: Do not move EAGLE metadata slot release earlier than first real decode result processing without proving the H2D/spec_info consume point changed.
Tested: g0034 docker py_compile for touched scheduler/disagg/test files; PYTHONPATH=python python -m pytest -q test/registered/unit/disaggregation/test_decode_queue_compaction.py test/registered/unit/mem_cache/test_req_to_token_pool.py test/registered/unit/managers/test_scheduler_health_check.py -> 24 passed
Not-tested: Fresh end-to-end burst traffic after restart; decode node_rank=1 persistent log capture.
Batch-planning metadata can now be attached even for single-request CP prefill, which routed FP8 flashmla_sparse through the batch cp_index path. That path used compact MQA-buffer row bases for score lookup but did not override the final ragged topk coordinate base consumed by attention, so topk indices could point at the wrong KV rows and produce low accept length or meaningless output.\n\nThis keeps ordinary long page tails out of compute padding, reserves compute padding for truly tiny suffixes, and makes cp_index RAGGED topk emit request-ragged offsets while preserving the compact buffer descriptors used for score materialization. The debug ledger records the rejected intermediate diagnoses and the confirmed coordinate-space failure.\n\nConstraint: CP shared-KV cache residency is page-granular, but attention/index compute must not consume synthetic long-tail rows.\nConstraint: FP8 CP prefill uses flashmla_sparse/RAGGED, where fused topk output is consumed directly as attention page_table_1.\nRejected: Disable current reuse or batch planning | would hide the regression and lose the intended bs>1 fast path.\nRejected: Treat all page tails as compute padding | regresses bs=1 semantics and can corrupt query/topk row contracts.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not change cp_index RAGGED topk offset handling without verifying score-buffer row bases and final attention KV coordinate space independently.\nTested: python -m py_compile on touched Python/test files; git diff --check; remote targeted ragged cp_index offset regression test; remote test_nsa_cp_utils.py; remote test_cp_shared_kv_runtime.py; user-reported ETE output recovered after restart.\nNot-tested: Agent-driven full ETE traffic run; broad multi-request bs>1 production soak.
CP shared KV current-only and partial-current paths were previously unsafe because owner-local current rows could be filled into dense page slots without synchronizing the current slots across CP ranks. This preserves the fast path and fixes the missing synchronization at the actual contract boundary: prefix slots remain materialized by the existing IPC/collective path, while freshly produced current slots are composed from current forward tensors and reduced only over the current slot ranges.\n\nThe change also keeps page-tail compute padding explicit, restores current-only index compose, fixes TAI index descriptor dtype consistency, and records the investigation ledger to avoid repeating discarded hypotheses.\n\nConstraint: CP shared KV must preserve current reuse and TAI materialize fast paths; blanket disable is not an acceptable fix.\nConstraint: Prefix slots must not be reduced twice after IPC/prefix materialization.\nRejected: Disable current-only MLA/index reuse | hides the owner-local composition bug and regresses the intended fast path.\nRejected: Disable TAI materialize globally | avoids symptoms without proving the byte/layout contract.\nConfidence: medium\nScope-risk: broad\nDirective: Do not remove current-slot range synchronization unless replacing it with an equivalent owner-aware P2P/IPC gather contract.\nTested: Local py_compile for touched files.\nTested: Remote g0034 py_compile for touched runtime/test files.\nTested: Remote test_cp_shared_kv_runtime.py: 111 passed, 5 warnings, 2 subtests passed.\nTested: Remote targeted current-slot compose tests and direct current-only index compose script.\nNot-tested: Full ETE output quality and decode accept len after restart.\nNot-tested: Full test_nsa_cp_utils.py collection on remote due incomplete installed sgl_kernel import/fake-op environment.
Mooncake previously truncated source prefill pages when the selected decode destination page list was shorter. That made an invalid prefill/decode page mapping continue as an incomplete KV transfer, which can surface later as decode garbage instead of the original mapping error.\n\nThis changes the transfer contract to require exact source/destination page-count equality and records compact page summaries in the fail-fast error. The helper is shared so the page-count contract can be unit-tested without constructing the transfer worker thread.\n\nConstraint: CP shared-KV page ownership requires a one-to-one prefill source page to decode destination page mapping.\nRejected: Keep warning-and-truncate | masks page-map corruption and can silently drop KV.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not restore source-page truncation; fix the mapping producer if this fail-fast triggers.\nTested: git diff --check; python -m py_compile python/sglang/srt/disaggregation/utils.py python/sglang/srt/disaggregation/mooncake/conn.py test/registered/unit/disaggregation/test_cp_shared_kv_transfer_mapping.py\nNot-tested: pytest blocked locally by missing orjson; remote g0034 was unavailable during this pass.
CP shared-KV now uses CP-local current rows consistently across MLA/index current reuse, passes fp8 current-index K through the tai-kernel uint8 ABI, and clears the transient EAGLE CP-local hidden marker after draft capture. The disaggregation bootstrap also fingerprints the runtime source contract so prefill/decode mismatches fail fast instead of silently exchanging incompatible KV metadata.
Constraint: CP shared-KV batch paths flatten current K/V rows in CP-rank-local valid order, not global request order.
Constraint: tai-kernel current-index prepare validates current_index_k as uint8 bytes for fp8 payloads.
Rejected: Keep using global extend offsets for bs>1 current-index reuse | corrupts request-local bases once current_index_kv is CP-local.
Rejected: Infer CP-local EAGLE hidden semantics from tensor shape | static padding and bs>1 can make shape-based inference unsafe.
Confidence: medium
Scope-risk: moderate
Directive: Do not reintroduce forward_batch.out_cache_loc slicing in CP shared-KV current reuse without verifying CP-local owner-lane layout.
Tested: Remote container py_compile for touched runtime/test files.
Tested: Remote PYTHONPATH=python pytest -q test/registered/unit/layers/test_nsa_cp_utils.py test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py test/registered/unit/disaggregation/test_common_conn_runtime_fingerprint.py (198 passed, 2 subtests passed).
Not-tested: Full remote ETE traffic after this commit; accept length and garbage-output recovery still require a fresh prefill/decode run.
Co-authored-by: OmX <omx@oh-my-codex.dev>
CP shared-KV bs>1 exposed three distinct padding domains: valid cache rows, CP page-tail compute rows, and MLP-sync flattened static padding. The previous implementation mixed these domains in direct-write and index top-k paths, so real requests failed when q/out_cache_loc lengths matched valid rows while metadata aliases described compute rows.\n\nThis change makes compute split strip only proven flattened static padding, keeps valid cache writes strict except for extend_num_tokens-proven static tails, marks CP-local EAGLE draft hidden state explicitly, and selects NSA index top-k query metadata by the actual q/weight row count.\n\nConstraint: CP shared-KV cache writes must never persist dummy page-tail or MLP static padding rows.\nConstraint: EAGLE draft hidden state can be CP-local before full CP metadata is visible in prepare_mlp_sync_batch.\nRejected: Use compute_padding_enabled as direct-write truncation proof | it silently accepts unknown out_cache_loc tails.\nRejected: Always consume compute q metadata in index top-k | actual q/weights can be valid-only after CP split.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not collapse valid rows, CP compute padding, and MLP static padding into one length condition; use explicit provenance.\nTested: remote py_compile for touched NSA files\nTested: remote targeted CP shared-KV padding/top-k regressions\nTested: remote pytest test_nsa_cp_utils.py test_cp_shared_kv_layout.py test_cp_shared_kv_runtime.py -k 'not test_tai_current_slot_fill_sparse_page_self_test_passes_on_installed_kernel' => 228 passed, 1 deselected, 5 warnings, 2 subtests passed\nNot-tested: full ETE replay after the final index top-k fix\nNot-tested: TAI current-index fast path dtype fallback
CP shared-KV prefill batching can present multiple requests to HiCache write-through in one forward. Keeping host free-room admission per request causes repeated small L2 evictions and defeats the free-room policy, while request metadata and radix attachment still need per-request ownership.\n\nThis change adds a batch prepare path that aggregates required host tokens by CP owner lane, performs one host admission/eviction step for the batch, and then reserves/submits each request independently. SessionAwareCache and the scheduler prefer the batch API when available.\n\nConstraint: Radix nodes, host slots, draft slots, rollback, and ack semantics remain per request.\nRejected: Merge reservations or radix nodes across requests | would complicate rollback and writing_check ack ownership.\nRejected: Add collective synchronization for host admission | local owner-lane logic already provides deterministic capacity accounting.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not reintroduce per-request host free-room eviction in the scheduler path without profiling host-full workloads.\nTested: local py_compile for touched Python files\nTested: remote g0034 docker PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_hicache_metadata.py => 117 passed, 5 warnings\nNot-tested: full ETE bs>1 HiCache run under production traffic
The scheduler can now admit multi-request NSA in-seq CP shared-KV prefill batches only when the shared-KV bs>1 flag is explicitly enabled. The gate is still disabled by default and is scoped to CP shared-KV so ordinary CP is not widened accidentally.
Batch admission is bounded by optional request-count and page-aligned extend-token limits while real memory capacity remains allocator-owned. This keeps bf16 and fp8 on the same scheduler path because dtype differences are already reflected in KV pool token/page capacity.
Constraint: bs>1 runtime paths remain guarded by existing CP shared-KV fail-fast checks.
Constraint: Scheduler must not duplicate bf16/fp8 byte-level capacity estimation.
Rejected: Open the old CP gate unconditionally | ordinary CP would inherit an unverified shared-KV-specific batching path.
Rejected: Treat the extend-token cap as a hard per-request limit | a single large request could deadlock the scheduler.
Confidence: medium
Scope-risk: moderate
Directive: Keep CP shared-KV batching gated until ETE validates EAGLE accept length, output length, and HiCache load/backup behavior under real traffic.
Tested: local py_compile for server_args, schedule_policy, scheduler, prefill_adder tests, and server_args tests.
Tested: remote g0034 py_compile for the same files.
Tested: remote g0034 pytest target set: 5 passed for parser, parameter validation, default single-request CP gate, enabled bs>1 gate, and page-aligned extend cap.
Tested: remote g0034 pytest test_prefill_adder.py => 13 passed.
Not-tested: full server_args test file has an unrelated HuggingFace DNS/config-download failure in TestPrepareServerArgs.test_prepare_server_args.
Not-tested: ETE production traffic with --enable-cp-shared-kv-prefill-bs-gt1.
EAGLE draft shared-KV is supposed to mirror the target CP layout, so bs>1 must not fall back to legacy full-input or padded-hidden behavior when required batch metadata is missing or inconsistent. This change keeps the existing bs=1 compatibility path but makes batched CP draft fail fast on missing/mismatched spec hidden states, embedding pad metadata, or input embed shapes. The docs record the current W7 boundary: draft prefill follows target metadata, while scheduler admission and ETE remain gated.
Constraint: CP draft KV must mirror target layout and must not silently diverge under bs>1 shared-KV.
Rejected: Allow bs>1 to use the old full-input fallback | it can hide wrong owner/page metadata and corrupt accept length.
Confidence: medium
Scope-risk: moderate
Directive: Do not open the scheduler bs>1 CP gate until EAGLE accept length/output length are verified with this fail-fast path enabled.
Tested: Remote g0034 targeted EAGLE fail-fast unit test passed; remote full test/registered/unit/layers/test_nsa_cp_utils.py passed 70 tests.
Not-tested: EAGLE bs>1 ETE, because scheduler CP bs>1 admission gate remains closed.
Tiny extend requests can leave most CP lanes without query work, which has been tied to hangs and accept-length regressions. This change introduces a dual valid/compute metadata contract: forward paths may materialize compute-padded rows, while cache, current reuse, direct write, HiCache backup, and load remain valid/page based.
The implementation keeps radix/HiCache/device allocation on real page extents, filters dummy compute rows before MLA/index cache writes and current reuse, makes top-k/index consume compute rows while compacting valid rows, and opens tiny CP shared-KV in-seq split through compute padding. The accompanying plan document records the contract and P1-P7 evidence.
Constraint: CP shared KV and HiCache must stay page-granular; dummy compute rows must not allocate, write, backup, or load KV cache.
Constraint: Avoid silent fallback and avoid adding collectives on hot paths.
Rejected: Pad cache allocations to cp_size pages | would waste KV capacity and pollute radix/HiCache state.
Rejected: Keep tiny suffixes out of CP split | preserves the zero-lane behavior that compute padding is meant to remove.
Confidence: medium
Scope-risk: broad
Directive: Do not route compute-padded dummy rows into out_cache_loc, current reuse, HiCache reservation, or backup descriptors; keep valid/cache metadata explicit.
Tested: Remote g0034 container targeted P7 tests: 3 passed, 3 warnings.
Tested: Remote g0034 container full unit slice: PYTHONPATH=python python -m pytest -q test/registered/unit/layers/test_nsa_cp_utils.py test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py test/registered/unit/mem_cache/test_cp_shared_kv_layout.py => 214 passed, 5 warnings, 2 subtests passed.
Tested: Local py_compile for touched P7 test file.
Not-tested: Latest CUDA/ETE traffic validation for dummy top-k rows, accept len, output len, and detokenizer hang behavior.
CP HiCache write reservations must stay per radix node, but the transfer descriptor does not need to be per request. This changes the layer-end hook to group pending write states for the same source and layer, so bs>1 prefill emits one target D2H descriptor and one draft D2H descriptor per layer while preserving per-node metadata, rollback, and ack semantics.\n\nConstraint: CP shared-KV HiCache metadata, host slots, and radix acknowledgements remain per request/node.\nConstraint: TAI direct transfer kernels already accept flattened page descriptors, so no tai-kernel change is required.\nRejected: Merge HiCache reservations or radix nodes | would complicate rollback and split handling.\nRejected: Add collective synchronization for grouped backup | grouping is local descriptor construction and must not add rank-level sync.\nConfidence: high\nScope-risk: moderate\nDirective: Keep target and draft source notifications separate; final ack must wait for both when draft HiCache is attached.\nTested: local py_compile for cache_controller.py and test_hicache_controller_cp.py\nTested: local git diff --check\nTested: remote pytest test/registered/unit/managers/test_hicache_controller_cp.py test/registered/unit/mem_cache/test_cp_hicache_load_back_owner_lanes.py => 85 passed, 3 warnings\nNot-tested: full ETE bs>1 CP HiCache replay with admission gate removed\nNot-tested: Nsight/throughput validation of reduced D2H submit count
The bs>1 path needs index top-k, shared-index prepare, current-index compact, and current-slot compose to consume flattened batch descriptors instead of falling back to per-request or per-segment Python/Torch work. This change wires SGLang to the new TAI batch prepare kernels, keeps fallback explicit, and records the remaining HiCache/load-backup gaps in the bs>1 workstream docs.
Constraint: CP shared-KV bs>1 must reuse fast paths rather than adding slow batch-only fallbacks
Constraint: No new collective operations were introduced
Rejected: Leave current-only cp_index as Python slice/cat | it keeps per-segment overhead in the short-extend bs>1 case
Rejected: Infer max segment lengths from CUDA descriptor tensors | .item() would add CPU synchronization on the hot path
Confidence: medium
Scope-risk: moderate
Directive: Do not remove the explicit fallback warnings without verifying the corresponding TAI symbols are present in production
Tested: local py_compile for touched SGLang files
Tested: remote g0034 test_nsa_cp_utils.py passed, 53 tests
Tested: remote g0034 test_fill_current_index_page_slots_uses_tai_kernel_when_available passed
Not-tested: full ETE bs>1 traffic with HiCache load/backup and draft/EAGLE enabled
PR #10 provides the W2 owner-lane allocation side needed by the current bs>1 CP shared-KV work: supported multi-request extends now derive page owners per request, preserve flattened request order, and use owner-lane allocation instead of legacy page allocation.\n\nThe merge applies cleanly on top of the W3/W4 target current-reuse work because the touched runtime surface is allocator/mem_cache only. Focused remote tests cover the new owner-lane planner/allocation behavior and the existing W4 target reuse regressions.\n\nConstraint: CP shared-KV page ownership is request-relative and page-granular.\nConstraint: Supported bs>1 CP shared-KV allocation must not silently fall back to legacy allocation.\nRejected: Treat multiple requests as one concatenated extend | would assign owners relative to the wrong request boundary.\nRejected: Keep multi_batch fallback | breaks downstream direct-write owner assumptions.\nConfidence: medium\nScope-risk: moderate\nDirective: Preserve request-order flattening when changing owner allocation; do not reorder by lane for convenience.\nTested: Local py_compile for allocator.py, common.py, cp_shared_kv_compute_owner.py, and test_cp_shared_kv_layout.py.\nTested: Local git diff --check --cached.\nTested: Remote g0034 py_compile for touched mem_cache files and test_cp_shared_kv_layout.py.\nTested: Remote g0034 pytest test_cp_shared_kv_layout.py => 34 passed, 3 warnings.\nTested: Remote g0034 pytest test_nsa_cp_utils.py test_cp_shared_kv_runtime.py => 157 passed, 5 warnings, 2 subtests passed.\nNot-tested: Full ETE/perf run; production HiCache load/back up with bs>1 allocation under sustained traffic.
Batch-size > 1 cache-hit traffic must not lose the current/partial-current reuse fast path. This change extends the target MLA/index sync-correct path to validate batched current suffix rows, compose page-aligned prefix spans, and route batched index top-k through current-only or partial-current reuse without falling back to scalar guards.\n\nThe implementation keeps page as the minimum cache unit: prefix cache coverage is page-aligned, while current suffix rows are sliced by valid extend lengths so padded tail rows are not exposed to attention. The index top-k batch path now mirrors the single-request contract: current-only reuses current index KV directly, partial-current materializes once and shares the dense buffer across request segments.\n\nConstraint: CP shared-KV supports in-seq zigzag only; round-robin remains fail-fast for shared-KV.\nConstraint: No new collectives are introduced; this is a sync correctness path, not a new communication scheme.\nRejected: Keep bs>1 current_index_kv fail-fast | disables the cache-hit path that W4 is meant to restore.\nRejected: Pad requests to a common batch length | violates the page-granular contract and exposes padded tails to consumers.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not reintroduce batch_size_not_one or batch_gt1 current-reuse guards without proving an equivalent batched fast path exists.\nTested: Local py_compile for touched runtime/test files.\nTested: Remote g0034 pytest test_nsa_cp_utils.py test_cp_shared_kv_runtime.py => 157 passed, 5 warnings, 2 subtests passed.\nTested: Remote g0034 pytest test_cp_shared_kv_layout.py => 27 passed, 3 warnings.\nNot-tested: Full ETE with live traffic; CUDA kernel-backed batched descriptors; L1 prefetch, HiCache load/backup, and draft/EAGLE bs>1 current reuse.
Route supported multi-request CP shared-KV extend allocation through request-local owner plans so downstream direct-write paths can trust flattened out_cache_loc owner lanes. Unsupported page plans now fail fast instead of falling back to legacy allocation.
Include the normative govctl RFC and completed work-item trail in the tracked repository; keep transient superpowers planning outside the sglang branch.
Constraint: RFC-0001 W2 requires no legacy multi_batch fallback for supported bs>1 owner-lane allocation
Rejected: concatenating requests before owner planning | CP ownership is request-relative
Rejected: committing only docs/superpowers plan | it omits the normative RFC and work-item trail
Confidence: high
Scope-risk: moderate
Directive: Preserve flattened request order when changing CP shared-KV allocation; run govctl from the sglang repository root
Tested: remote uv test_nsa_cp_utils.py 39 OK; remote uv test_cp_shared_kv_layout.py 34 OK; remote uv test_alloc_pages_with_owners.py 10 OK; govctl check; git diff --check
Not-tested: CUDA ETE/perf paths beyond focused W1/W2 unit suites
W4-1 needs target index/top-k sync correctness before current/partial-current reuse can be restored. Batch-size>1 in-seq CP produces local q/weights in request-segment order, so top-k must consume req0_prev, req0_next, req1_prev, req1_next rather than treating the flattened batch as one scalar prev/next pair.
The implementation adds a batch dispatch for _get_topk_in_seq_cp_pair, reuses one synchronous shared-index materialization per layer, and calls _get_topk_ragged_with_cp per request segment with an explicit batch_idx. The scalar bs=1 path remains unchanged.
Constraint: This is W4-1 target index/top-k sync correctness; original W4 current/partial-current reuse remains a separate follow-up.
Constraint: Phase W4-1 must not enable bs>1 index prefetch, current reuse, partial-current reuse, or the cp_index multi-batch branch.
Rejected: Use cp_index branch for multi-batch | source marks that path as having accuracy issues.
Rejected: Pad batch requests to max length | wastes compute and violates packed/ragged batch contract.
Confidence: high
Scope-risk: moderate
Directive: Keep bs>1 target top-k ordered by request segment unless a later fused descriptor proves identical ordering and correctness.
Tested: Remote g0034 py_compile for nsa_indexer.py
Tested: Remote g0034 PYTHONPATH=python pytest test/registered/unit/layers/test_nsa_cp_utils.py -> 45 passed
Tested: Remote g0034 PYTHONPATH=python pytest test_nsa_cp_utils.py test_cp_shared_kv_layout.py test_cp_shared_kv_runtime.py -> 172 passed, 2 subtests passed
Not-tested: Full ETE bs>1 serving run with live traffic
W3 needs batch-size>1 extends to use packed valid tokens while preserving per-request page boundaries. The local out_cache_loc planner now validates against the batch plan's request lengths instead of the first request's scalar split metadata, then reuses the existing batch split helper to produce this rank's logical/physical cache locs.
Direct-write failures inside the CP shared-KV contract now fail fast instead of silently falling back to legacy index/MLA stores. This exposes allocator, owner-lane, page-alignment, and shape-contract bugs early for both bs=1 and bs>1.
Constraint: bs>1 batching must not pad short requests to the longest request; only per-request page-boundary padding is allowed.
Rejected: Keep bs=1 compatibility fallback | it hides CP shared-KV contract violations and caused repeated slow-path ambiguity.
Rejected: Pad batch to max request length | wastes compute and complicates cache validity for short extends.
Confidence: high
Scope-risk: moderate
Directive: CP shared-KV contract errors should stay fail-fast; do not reintroduce silent direct-write fallback without ETE evidence and explicit warning semantics.
Tested: Remote g0034 py_compile for utils.py nsa_indexer.py forward_mla.py
Tested: Remote g0034 PYTHONPATH=python pytest test/registered/unit/layers/test_nsa_cp_utils.py -> 43 passed
Tested: Remote g0034 PYTHONPATH=python pytest test_nsa_cp_utils.py test_cp_shared_kv_layout.py test_cp_shared_kv_runtime.py -> 170 passed, 2 subtests passed
Not-tested: Full ETE bs>1 serving run with live traffic
Batch-size support needs request-first CP metadata; treating a batch as one long sequence breaks page ownership, top-k ranges, and phase1 compact output collection. This adds a batch CP plan that records per-request page-aligned splits, rank-local offsets, kv/actual-seq metadata, last-token owners, and flattened descriptors for downstream allocator/runtime workstreams.
The scalar full-rerange path now fail-fasts for batch metadata so bs>1 cannot silently discard the narrow-output optimization or restore hidden states with single-request assumptions.
Constraint: CP shared-KV cache state is page-owned and must preserve request boundaries under bs>1.
Rejected: Let bs>1 fall back to scalar full hidden rerange | it loses the phase1 communication reduction and uses wrong single-request metadata.
Rejected: Add a collective to confirm batch plans | all ranks can derive the same plan from CPU metadata and config.
Confidence: medium
Scope-risk: moderate
Directive: Do not remove batch fail-fast guards until W2/W3 consumers use CPSharedKVBatchPlan end-to-end.
Tested: python -m py_compile python/sglang/srt/layers/attention/nsa/utils.py test/registered/unit/layers/test_nsa_cp_utils.py
Tested: remote g0034 container PYTHONPATH=python python -m pytest -q test/registered/unit/layers/test_nsa_cp_utils.py -> 39 passed
Not-tested: full ETE bs>1 CP shared-KV runtime; W2/W3 allocator/direct-write consumers are not implemented yet
The existing CP shared-KV notes still reflected the earlier multi-slot-first conclusion, while current short cache-hit extends require true ForwardBatch>1 planning. This commit records the batch-aware contract, sequential phase plan, and parallel workstream split so implementation can proceed without rediscovering single-request blockers or accidentally removing guards into unsafe fallback paths.
Constraint: Current NSA in-seq CP shared-KV paths assume batch_size == 1 across metadata, owner-lane allocation, direct write, current reuse, prefetch, top-k, and EAGLE/draft collection.
Rejected: Remove batch_size guards directly | would treat multiple requests as one long sequence and corrupt request boundaries and page-owner state.
Rejected: Keep only the old multi-slot recommendation | online extend sizes around 200-2000 tokens make true batching a required throughput direction.
Confidence: high
Scope-risk: narrow
Directive: Implement bs>1 by per-request page-aligned planning before flattening; do not add planner collectives or silent legacy fallbacks.
Tested: Placeholder scan and referenced-path existence check for all four docs; git diff --cached --check.
Not-tested: Runtime behavior; documentation-only change.
Co-authored-by: OmX <omx@oh-my-codex.dev>
The missing index prefetcher state is expected when MLA/index prefetch is disabled. Logging it as an index-prefetch fallback made the normal synchronous materialize path look like a failed fast path and created noisy reports.
This keeps consume-miss warnings for enabled prefetchers, but suppresses missing-prefetcher warnings when the prefetch feature gate is off.
Constraint: Index prefetch is controlled by the same MLA prefetch enable gate in the current pipeline.
Rejected: Suppress all index prefetch fallback warnings | consume misses still indicate an enabled prefetch pipeline failed to serve a requested buffer.
Confidence: high
Scope-risk: narrow
Directive: Missing prefetcher is not a fallback when SGLANG_CP_SHARED_KV_ENABLE_MLA_PREFETCH=0; do not re-add warning noise without a separate enable gate.
Tested: Local py_compile for nsa_indexer.py and test_cp_shared_kv_runtime.py.
Tested: Remote g0034 targeted regression set passed 4 tests: disabled sync materialize no warning, consume miss warning, first-layer no-warning, enabled create-skip warning.
Not-tested: Full CUDA ETE run with logs.
Host HiCache reservations were paying token-level free-slot scans when trying to preserve page contiguity. The allocator now keeps a lazy page-extent index so availability checks and contiguous-preferred allocations avoid materializing the full 220GB-equivalent free-slot metadata path.
The companion benchmark models steady-state L2 churn near full occupancy, including burn-in and historical node-size effects, so LPF/RDMA descriptor quality can be separated from ETE noise.
Constraint: CP HiCache host allocations are page-shaped, but existing callers may still read free_slots directly.
Rejected: Sort and scan free_slots on each alloc_contiguous_preferred call | measured ms-level CPU overhead on 220GB-equivalent metadata.
Rejected: Remove free_slots compatibility | storage/tests still rely on the public tensor surface.
Confidence: medium
Scope-risk: moderate
Directive: Do not reintroduce per-allocation full free_slots scans on HostKVCache; preserve page-extent metadata or benchmark before changing allocator shape.
Tested: Local py_compile for memory_pool_host.py, allocator benchmark, and related tests.
Tested: Local test_cp_hicache_allocator_bench.py 10 passed.
Tested: Remote g0034 test_hicache_controller_cp.py 67 passed; test_cp_hicache_allocator_bench.py 10 passed.
Tested: Remote 220GB-equivalent host_churn benchmark: contiguous path reduced from ms-level to ~30-292us p50 depending on fragmentation.
Not-tested: Full CUDA ETE run after allocator change.
Not-tested: Production long-run fragmentation behavior under live traffic.
The CP shared-KV allocator was still doing total-cache-sized CPU work in the scheduler hot path. That cannot be hidden by GPU overlap, so owner-lane allocation now maintains per-owner free/release buckets and consumes request-sized prefixes instead of rebuilding masks over the full free-page tensor on each request.\n\nThe benchmark was extended to isolate L1 stats, selection, and allocation costs, and the CPU layout tests now install a complete sgl_kernel stub before importing SGLang helpers so remote unit collection does not abort in native extension loading.\n\nConstraint: Allocator CPU work blocks scheduler progress and cannot overlap with GPU forward execution.\nConstraint: CPU unit tests must not load native sgl_kernel on remote images where the loader can SIGABRT.\nRejected: Keep contiguous-run search over full free_pages | still scales with cache capacity and measured multi-ms overhead.\nRejected: Treat remote collection abort as an environment-only issue | it prevented allocator regression coverage and was fixable with a test-local stub.\nConfidence: high\nScope-risk: moderate\nDirective: CP owner-lane allocation is bucket-based; do not reintroduce full free_pages scans on the hot path without benchmark evidence.\nTested: Local py_compile for touched files\nTested: Local benchmark unit test, 6 passed\nTested: Remote benchmark unit test, 6 passed\nTested: Remote test_alloc_pages_with_owners.py, 10 passed\nTested: Remote test_cp_shared_kv_layout.py, 27 passed\nTested: Remote production allocator microbench shows select/alloc p50 reduced from ms-scale to sub-ms scale\nNot-tested: Full ETE traffic run after allocator bucket change
CP shared KV now keeps explicit L1 and host free-room targets so pressure is handled by planned eviction instead of repeated capacity-edge retries. The host allocator gains contiguous-preferred page reservation, L1 owner-lane allocation prefers contiguous physical pages, and CP HiCache metadata preserves pending backup safety for page-granular radix updates. Mooncake transfer stats and allocator microbenchmarks are included to make the remaining transfer bottlenecks measurable rather than inferred.
Constraint: CP shared KV uses decode CP size 1 with all prefill CP ranks participating in transfer, so L1/L2 cache residency must remain page-granular and avoid extra collectives.\nConstraint: Production HiCache can be hundreds of GB, so allocator metadata overhead must be visible before enabling aggressive contiguous allocation broadly.\nRejected: Evict only the exact deficit | this keeps the cache at the cliff and causes repeated evict/allocate pressure.\nRejected: Rely on allocator scans alone for contiguity | remote microbenchmarks show fragmented 220GB-equivalent host metadata can make contiguous-preferred scans multi-ms.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not increase L1/L2 free-room defaults or add new CP collectives without ETE evidence and transfer/allocator measurements.\nTested: python -m py_compile on touched runtime/test/benchmark files.\nTested: PYTHONPATH=. python -m pytest -q test/registered/unit/benchmark/test_cp_hicache_allocator_bench.py => 4 passed, 1 warning.\nTested: Remote g0034 log /mnt/beegfs/cjy/log/sglang_cp_hicache_20260601_233723.log shows active prefill process with L1/L2 free-room args, 702 HTTP 200 chat completions, 6272 prefill batches, and no fatal scheduler traceback in latest scan.\nTested: User-reported L1/L2 cache ETE validation passed on remote run.\nNot-tested: Full local pytest suite; local environment is missing several runtime dependencies.\nNot-tested: CUDA allocator microbenchmark during active production prefill process.\nNot-tested: Mooncake straggler fix; stats show transfer tail latency remains a separate bottleneck.
Chunked prefill was skipping prepared CP HiCache backup entirely while still inserting unfinished valid-tail radix state. That combination caused middle chunks to miss reusable host cache and repeatedly hit stale-tail split fallback paths.\n\nThis keeps CP HiCache page-granular: middle chunks now reserve/write only completed pages, and unfinished radix inserts expose only page-complete prefixes while preserving the request-owned sub-page tail in prefix_indices for later chunks.\n\nConstraint: CP HiCache radix and host residency are managed at page granularity, but chunked prefill can end on a sub-page tail.\nRejected: Continue skipping chunked backup | loses middle-chunk reuse and triggers fallback storms.\nRejected: Insert sub-page chunk tails into CP HiCache | creates stale-tail prune/split conflicts against request-owned KV.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not reintroduce chunked_req backup skip; keep CP HiCache chunked inserts page-complete unless the radix/host cache contract becomes sub-page aware.\nTested: python -m py_compile python/sglang/srt/mem_cache/radix_cache.py python/sglang/srt/mem_cache/hiradix_cache.py test/registered/unit/mem_cache/test_cp_hicache_metadata.py\nTested: g0034 targeted chunked CP HiCache regression tests, 2 passed\nTested: g0034 PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_hicache_metadata.py, 107 passed\nNot-tested: Fresh chunked-prefill ETE traffic run after process restart
Prepared per-layer CP HiCache backups can enqueue their final write ack before the radix node is attached. Chunked prefill exposed this window when a separate catch-up backup made writing_check drain the ack queue and pop an unattached node id.
Keep ready but unattached prepared acks queued until insert either attaches the prepared backup or rollback removes the orphan ack. Also document the reactive host free-room eviction plan separately from this state-machine fix.
Constraint: CP HiCache prepared backup transfer can complete before radix insertion attaches node state
Rejected: Drop unknown ack ids | would orphan a later successful prepared attach and leak write state
Rejected: Chunked-only guard | the invalid assumption is in the generic CP write ack state machine
Confidence: high
Scope-risk: narrow
Directive: Do not drain CP write acks unless every ack id is registered in pending_host_backups or ongoing_write_through
Tested: Remote red-green test_cp_hicache_metadata.py::TestHiRadixCacheCPBackup::test_writing_check_defers_unattached_prepared_ack
Tested: Remote PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_hicache_metadata.py test/registered/unit/mem_cache/test_cp_hicache_load_back_owner_lanes.py (116 passed)
Tested: python -m py_compile python/sglang/srt/mem_cache/hiradix_cache.py test/registered/unit/mem_cache/test_cp_hicache_metadata.py
Not-tested: Full chunked-prefill ETE replay after this commit
Co-authored-by: OmX <omx@oh-my-codex.dev>
Chunked prefill can revisit a sub-page CP HiCache tail while a per-layer backup is still in flight. The old insert path split the radix node first and only then tried to prune the stale tail, so an unprunable pending backup raised after tree mutation and propagated to the scheduler.
This makes split/prune atomic from the radix-tree perspective: drain completed write acks only on the conflict path, preflight pending/unprunable state before split, and return a deferred insert result when the backup is still in flight. Unfinished requests keep their KV ownership for transfer/release instead of rematching or freeing pages under a stale tree state.
Constraint: CP HiCache backup metadata is node/page owned and cannot be repartitioned while per-layer D2H is pending
Rejected: Split the pending backup node | would require repartitioning in-flight backup metadata and host reservations
Rejected: Delete the stale tail unconditionally | risks freeing device/host pages still owned by pending backup
Confidence: high
Scope-risk: moderate
Directive: Do not move stale-tail prune after split without a preflight; pending backup split conflicts must remain non-mutating
Tested: remote g0034 container py_compile for mem_cache files
Tested: remote g0034 PYTHONPATH=python pytest -q test/registered/unit/mem_cache/test_cp_hicache_metadata.py test/registered/unit/mem_cache/test_cp_hicache_load_back_owner_lanes.py => 115 passed
Tested: local git diff --check
Not-tested: full chunked prefill ETE after service restart
Co-authored-by: OmX <omx@oh-my-codex.dev>
The LPF direct-kernel work exposed an allocation-dependent performance boundary: layer-page-first reduces descriptors only when host pages have contiguous extents. Capture the benchmark evidence and owner-lane allocator interpretation so future changes do not rediscover the same constraint.
Constraint: Current CP compute-owner allocation selects pages by modulo owner lane, e.g. (page_id - 1) % cp_size.
Rejected: Document LPF as an unconditional production improvement | random and owner-lane same-layout benchmarks are neutral without compact allocation.
Rejected: Treat owner_lane benchmark as full allocator replay | it models the modulo-lane constraint, while free/release history can add fragmentation.
Confidence: medium
Scope-risk: narrow
Directive: Tie any production LPF switch to compact/extent-aware host allocation or gather/staging support.
Tested: Local doc marker check for C115/C116/C117.
Tested: Remote tai-kernel CUDA pytest for the referenced kernel/test changes -> 45 passed in 2.52s.
Not-tested: Full SGLang ETE after changing host allocation policy; no such policy change is included.
NSA FP8 CP shared-KV reuse must operate on packed page-slot rows, not bf16 compact rows. The change keeps current-only and partial-current reuse inside the page-aligned materialization contract, fails fast for non-page-aligned CP split inputs, and prevents FP8 FlashMLA-KV prefill from reaching incompatible in-seq CP metadata.
Constraint: NSA FP8 persistent MLA KV rows are packed 656-byte records and CP shared KV cache management is page-granular.\nConstraint: FlashMLA-KV prefill metadata is not CP-local after NSA in-seq splitting.\nRejected: Silently splice bf16 current rows into FP8 materialized cache | corrupts the packed cache layout.\nRejected: Keep FP8 FlashMLA-KV auto prefill under NSA CP | reaches num_splits shape errors after q-row splitting.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not re-enable FP8 FlashMLA-KV prefill for NSA in-seq CP until metadata is rebuilt after CP splitting or made CP-local.\nTested: Local git diff --check and py_compile for touched SGLang files.\nTested: Remote g0034 related unit sweep recorded in docs: test_nsa_cp_utils.py, test_cp_shared_kv_layout.py, test_cp_shared_kv_runtime.py, test_cp_hicache_metadata.py passed.\nNot-tested: Full FP8 ETE startup and performance run after this commit.
Launching CP shared-KV prefetch from MLA prepare or the indexer can make
next-layer prefix work overlap current-layer MQA/materialization instead of
the attention window. Centralize the launch in the NSA backend after
current-layer materialization and before attention, and leave the early
indexer hook inert so the call site cannot regress silently.
The accompanying notes capture the draft-as-forward-layer follow-up plan and
the latest OOM diagnosis: the observed 178945-token failure matches the CP
in-seq MQA logits allocation, so the follow-up fix is q-dimension chunking in
_get_topk_ragged_with_cp(), not a max-prefetch-size gate.
Constraint: CP shared-KV prefetch must avoid overlapping current-layer MQA/materialization and must not add silent fallback behavior.
Rejected: Limit maximum prefetch size | hides the CP logits peak and can reduce cache/prefetch effectiveness.
Rejected: Keep prepare/indexer launch sites | they place next-layer collectives in the wrong overlap window.
Confidence: medium
Scope-risk: moderate
Directive: Do not reintroduce early CP prefetch launch without checking Nsight overlap and CP MQA memory peaks.
Tested: Local git diff --check and py_compile for touched Python files.
Tested: Remote container py_compile plus targeted pytest: 3 passed, 5 warnings.
Not-tested: Full ETE under production traffic after this commit.
Not-tested: CP in-seq MQA logits chunking; documented as follow-up.
Synchronous CP shared KV full-hit and partial-current paths can now use the tai-kernel CUDA IPC slot-dense materialize path instead of first copying local owner pages and then running a dense CP all-reduce. This keeps the async prefetch pipeline unchanged while routing the safer synchronous runtime path through descriptor-driven owner-page reads.
The runtime builds descriptors from the existing page-aligned slot contract, caches peer pointer tables for long-lived KV/index buffers, and falls back with explicit warnings if the tai-kernel IPC capability is missing. Unit coverage locks offset handle exchange, descriptor construction, and both MLA/index full and partial-current calls.
Constraint: Async prefetch currently has unresolved scheduling contention with GEMM/MoE and current-layer KV collectives, so this commit intentionally does not wire IPC into prefetch.
Constraint: tai-kernel must provide offset-aware CUDA IPC symbols from af9fb67.
Rejected: Replace prefetch all-reduce in this slice | Nsight shows prefetch timing and communicator contention need a separate scheduling design.
Rejected: Fail-fast on missing tai IPC immediately | remote ETE still needs to validate production deployment capability before removing the warning fallback.
Confidence: medium
Scope-risk: moderate
Directive: Keep prefetch and synchronous materialize decisions separate until prefetch has a low-SM/copy-engine schedule.
Related: tai-kernel af9fb67
Tested: git diff --check; remote runtime/unit evidence recorded in docs/advanced_features/nsa_prefill_cp_page_aligned_cache_contract.md.
Not-tested: Fresh GLM5 ETE after this commit; async prefetch IPC path.
Co-authored-by: OmX <omx@oh-my-codex.dev>
The materialize transport direction changed after measurement: direct CUDA IPC peer-page materialize was initially slow because of an under-parallel launch policy, not because slot-dense materialize was inherently too expensive. Capture the corrected evidence and the remaining production constraints before SGLang runtime wiring starts.
Constraint: Current near-term goal accepts SM-consuming kernels for low latency/high throughput; low-SM prefetch friendliness is deferred.
Rejected: Treat the first slow IPC result as a design blocker | tuned launch parameters beat dense all-reduce across the measured 4k-120k prefix range.
Rejected: Switch consumers to owner-concat immediately | slot-dense fused materialize is competitive enough to preserve the current consumer contract for the next integration step.
Confidence: medium
Scope-risk: narrow
Directive: Keep this document updated with every benchmark/result correction to avoid re-litigating stale conclusions.
Tested: Documentation update based on remote g0034 tai-kernel CUDA tests and cp_shared_kv_ipc_sm_tuned_20260531_181926 benchmark log
Not-tested: SGLang runtime ETE serving with IPC materialize enabled
Add a production-style CP shared KV collective benchmark and keep the page-aligned cache contract ledger current with the collective, unpack, multi-extend prefix, P2P, and CUDA IPC findings. The benchmark makes the final consumer layout explicit so dense all-reduce and owner-packed all-gather are compared after producing the same logical-dense result.\n\nConstraint: Prefixes may be assembled from multiple radix/cache extents, so single-run zigzag ordering is only a gated fast path.\nRejected: Treat rank-major all-gather output as the final product | consumers require logical-dense ordering.\nRejected: Switch index materialization based on noisy microbenchmarks | current evidence is not strong enough for production.\nConfidence: medium\nScope-risk: narrow\nDirective: Do not replace generic logical-dense materialization with zigzag ordered gather unless a run descriptor proves the prefix shape.\nTested: python -m py_compile benchmark/hicache/bench_cp_shared_kv_production_collective.py\nTested: Remote production-style benchmark logs under /mnt/beegfs/cjy/log/cp_shared_kv_collective_bench_*_20260531_*.log\nNot-tested: Production SGLang ETE with a CUDA IPC/P2P materialization path; this commit only adds benchmark/documentation surfaces.
Page-aligned CP shared KV can pad out_cache_loc beyond valid current rows, so current reuse now gates MLA composition on the valid extend rows and permits draft partial-current reuse once the TAI sparse-page capability check passes. The TAI current-slot path self-tests sparse pages before use and falls back to the torch reference when the installed kernel is stale.
Eviction success and no-op diagnostics were also moved from INFO to DEBUG so owner-lane and host-admission churn does not flood production logs; true write failures remain WARNING.
Constraint: CP shared KV uses page-aligned physical reservations where valid suffix rows can be shorter than padded out_cache_loc.
Constraint: Production failure/fallback logs must remain visible, but hot successful eviction paths should not emit INFO per victim/rank.
Rejected: Keep draft partial-current reuse disabled | would preserve avoidable full materialization on draft cache-hit suffixes.
Rejected: Trust the TAI current-slot kernel unconditionally | stale kernels can corrupt sparse current-page composition.
Confidence: medium
Scope-risk: moderate
Directive: Do not reintroduce INFO logging in eviction hot paths without rate limiting and runtime evidence.
Tested: local py_compile for touched Python files
Tested: local git diff --check
Tested: remote container py_compile for touched Python files
Tested: remote PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_hicache_metadata.py::TestHiCacheEvictLoggingLevels::test_evict_hot_path_success_logs_are_debug_only test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py -> 82 passed, 5 warnings, 2 subtests passed
Not-tested: full ETE traffic after this commit; draft partial-current accept length still needs user-driven runtime validation
The index path now mirrors the target MLA partial-current contract: prefetched or synchronously materialized prefix pages are composed with valid current index K/scale rows in slot-dense page buffers. Current-only batches keep the compact current-index fast path, while partial cache-hit batches share one composed dense index buffer across the in-seq prev/next topk pair.\n\nThe prefetch consume path remaps through the slot page inverse instead of treating the slot-dense buffer as physical-pool capacity, and current-index quantization uses valid extend rows so padded out_cache_loc does not disable reuse.\n\nConstraint: CP shared KV remains page-slot based; padding rows must stay invisible to attention/index semantics\nConstraint: Draft/EAGLE partial-current reuse remains guarded by should_reuse_current_extend_kv\nRejected: Replace prefix all-reduce with all-gather | NCCL all-gather still uses SM and would require an additional compose/scatter step\nConfidence: medium\nScope-risk: moderate\nDirective: Do not reintroduce current-only gating for index reuse; partial target cache hits must compose prefix + valid current rows\nTested: Local py_compile for touched Python files\nTested: g0034 sglang-glm5-dev-2 PYTHONPATH=python python -m pytest -q test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py -> 77 passed, 5 warnings, 2 subtests passed\nNot-tested: Full ETE traffic with latest commit; CUDA perf impact of index partial-current prefetch under production load
