[cctry]

Changes

1. Adaptive yield: an optimistic prefill no longer gives up at every chunk boundary but it yields only when a bootstrap-finished request is waiting for the batch.
2. Reframe retry to attempt: counting total optimistic forward passes instead of requeue. Previously retry limit is 1 means the req has two forward opportunities before bootstrap finished.
3. Park instead of retry: a request whose prefill finishes before bootstrap completes no longer releases and requeue. It parks in the inflight queue waiting for transfer.

Fixes along the way.

• Chunks computed while pending now store their input logprobs.
• reset_for_retract now clears temp_input_token_ids_logprobs_val/idx.
• Scheduler prefill log now reports #optimistic-req per batch.

---

CI States

Latest PR Test (Base): ❌ Run #29210525648
Latest PR Test (Extra): ❌ Run #29210525451

[gemini-code-assist]

Code Review

This pull request refactors the optimistic prefill retry mechanism in the disaggregation scheduler, renaming retries to attempts and tracking attempt counts directly on the request object. The code review highlights two critical issues: first, an assert statement wrapping a function with side effects (finalize_bootstrap) that could be compiled out in optimized environments, along with a missing state update for pending_bootstrap; second, a double-increment bug on prefill_attempt_count that prematurely limits the number of allowed prefill attempts.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[cctry]

/tag-run-ci-label

[cctry]

/rerun-group disagg

[github-actions]

⛔ disagg: Unknown test group disagg.

Known groups: 4-gpu-models, 8-gpu-models, amd, ascend, attention, backends, bench_fn, chunked_prefill, constrained_decoding, core, cp, cpu, cuda_graph, dcp, debug_utils, disaggregation, dllm, dp_attn, dp_engine, ep, eplb, eval, function_call, gb300, hicache, input_embedding, jit, kernels, kv_canary, layers, lora, mem_cache, mla, mlx, mock_model, model_loading, models, models_e2e, moe, musa, observability, openai_server, ops, page_major, perf, pp, prefill_only, profiling, quant, radix_cache, reasoning, rl, rotary, sampling, scheduler, scripted_runtime, sessions, spec, stress, tokenizer, unit, utils, vlm, xpu

[cctry]

/rerun-group disaggregation

[github-actions]

Results for /rerun-group disaggregation:

🚀 4-gpu-gb300 (1 test): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_aarch64.py

🚀 2-gpu-h100 (3 tests): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_basic.py
cd test/ && python3 registered/disaggregation/test_disaggregation_decode_offload.py
cd test/ && python3 registered/disaggregation/test_disaggregation_optimistic_prefill.py

🚀 8-gpu-h20 (4 tests): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_decode_radix_cache.py
cd test/ && python3 registered/disaggregation/test_disaggregation_different_tp.py
cd test/ && python3 registered/disaggregation/test_disaggregation_dp_attention.py
cd test/ && python3 registered/disaggregation/test_disaggregation_pp.py

🚀 8-gpu-h200 (2 tests): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_dsv4.py
cd test/ && python3 registered/disaggregation/test_disaggregation_hisparse.py

🚀 8-gpu-h200 (1 test): ❌ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_hybrid_attention.py

🚀 1-gpu-5090 (2 tests): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_disaggregation_xpu.py
cd test/ && python3 registered/disaggregation/test_specv2_kvcache_offloading.py

🚀 4-gpu-h100 (1 test): ✅ View workflow run

cd test/ && python3 registered/disaggregation/test_epd_disaggregation.py

[cctry]

https://github.com/sgl-project/sglang/actions/runs/29210654951/job/86699353345
Only failed test is known to be flaky and tracked at #30946