[PD] Fix KV cache corruption on abort by notifying ongoing prefill

[ShangmingCai]

Motivation

A lightweight alternative to #24580. This should address almost all race conditions caused by long ITL aborts.

Note: this PR cannot address the case when an ongoing prefill has already called a transfer sync that would take a very long time (maximum value -> mooncake timeout: 30s), but this case has an extremely low probability of happening (extremely hard to reproduce). We could introduce deferred release of decode-side KVCache to fix this as well, but this could harm the performance of the pre-allocation process of decode (considering this aborted req is a large ITL req that will occupy many many KVCache slots).

Modifications

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): ❌ Run #27023869436
Latest PR Test (Extra): ❌ Run #27023869006

[ShangmingCai]

/rerun-group disaggregation

[github-actions]

Results for /rerun-group disaggregation:

🚀 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 (1 test): ✅ View workflow run

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

⛔ registered/disaggregation/test_disaggregation_aarch64.py: Unknown runner_config 4-gpu-gb200 in test/registered/disaggregation/test_disaggregation_aarch64.py — not in scripts/ci/runner_configs.yml.

Known runner_configs: 1-gpu-large, 1-gpu-small, 2-gpu-large, 4-gpu-b200, 4-gpu-h100, 8-gpu-b200, 8-gpu-h20, 8-gpu-h200, deepep-4-gpu-b200, deepep-4-gpu-h100, deepep-8-gpu-h200

[gemini-code-assist]

Code Review

This pull request introduces a deferred KV cache release mechanism for decode-initiated aborts to prevent RDMA corruption from in-flight prefill writes. It implements an abort notification flow using ABORT and ABORT_ACK messages, along with a configurable grace period. The review feedback highlights a critical bug where an incorrect attribute name (decode_initiated_abort instead of abort_initiated) disables the deferred release mechanism. Additionally, the reviewer noted potential background thread crashes due to unvalidated message parsing and a potential memory leak from late-arriving ABORT_ACK messages.

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.

[ShangmingCai]

/rerun-group disaggregation

[github-actions]

Results for /rerun-group disaggregation:

🚀 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 (1 test): ✅ View workflow run

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

⛔ registered/disaggregation/test_disaggregation_aarch64.py: Unknown runner_config 4-gpu-gb200 in test/registered/disaggregation/test_disaggregation_aarch64.py — not in scripts/ci/runner_configs.yml.

Known runner_configs: 1-gpu-large, 1-gpu-small, 2-gpu-large, 4-gpu-b200, 4-gpu-h100, 8-gpu-b200, 8-gpu-h20, 8-gpu-h200, deepep-4-gpu-b200, deepep-4-gpu-h100, deepep-8-gpu-h200