{"old": "/home/jenkins/agent/workspace/multiarch/src-openeuler/aarch64/LMCache-Ascend/old_rpms/", "new": "/home/jenkins/agent/workspace/multiarch/src-openeuler/aarch64/LMCache-Ascend/new_rpms/", "compare_result": "pass", "compare_details": {"same": {"same_details": {"old": ["LMCache-Ascend-0.4.4-5.oe2609.aarch64.rpm"], "new": ["LMCache-Ascend-0.4.4-6.oe2609.aarch64.rpm"]}, "same_num": 1}, "diff": {"diff_details": {"LMCache-Ascend": {"name": {"old": "LMCache-Ascend-0.4.4-5.oe2609.aarch64.rpm", "new": "LMCache-Ascend-0.4.4-6.oe2609.aarch64.rpm"}, "RPM Level": "level4", "rpm requires": {}, "rpm provides": {}, "rpm files": {"more": ["/usr/share/LMCache-Ascend/examples/disagg_prefill/1p1d/configs", "/usr/share/LMCache-Ascend", "/usr/share/LMCache-Ascend/examples/disagg_prefill/disagg_proxy_server.py", "/usr/share/LMCache-Ascend/examples/disagg_prefill/1p1d/README.md", "/usr/share/LMCache-Ascend/examples/disagg_prefill/1p1d/configs/lmcache-decoder-config.yaml", "/usr/share/LMCache-Ascend/examples/disagg_prefill/disagg_proxy_request.py", "/usr/share/LMCache-Ascend/examples/disagg_prefill", "/usr/share/LMCache-Ascend/examples", "/usr/share/LMCache-Ascend/examples/disagg_prefill/1p1d", "/usr/share/LMCache-Ascend/examples/disagg_prefill/1p1d/configs/lmcache-prefiller-config.yaml"]}, "rpm abi": {}, "rpm cmd": {}, "rpm lib": {}, "rpm symbol": {}}}, "diff_num": 0}, "less": {"less_details": {}, "less_num": 0}, "more": {"more_details": {}, "more_num": 0}}, "pr_link": "https://gitcode.com/src-openeuler/LMCache-Ascend/pull/8", "pr_changelog": "* Sun Sep 20 2026 jing_qifeng <jing_qifeng@163.com> - 0.4.4-6\n- Ship the PD-disaggregation proxy example carried by the backport:\n  disagg_proxy_server.py, disagg_proxy_request.py and the 1p1d configs\n  are installed under /usr/share/LMCache-Ascend/examples/disagg_prefill/\n  (runnable code kept out of %_docdir so nodocs container installs still\n  receive it).\n- Apply the P2P controller patch shipped inside Source0\n  (docker/lmcache-controller.diff) to the lmcache sources, so the\n  controller resolves P2P lookups on the requesting worker\n  (find_worker_key / find_kv_with_worker_info target_worker_id);\n  required for P2P KV sharing with tensor parallelism > 1.  The lmcache\n  0.4.4 sdist is fetched from the mirror with 'pip download\n  --no-binary' at build time instead of being vendored, and the patched\n  tree is wheeled with --no-build-isolation so the sdist's\n  torch==2.10.0 build requirement is not pulled into an isolated build\n  env.\n- Backport the PD disagg proxy edge-case fixes pushed on the\n  feature/pd-kvpool-adaptation branch after the original backport\n  (97088fd, c14139a) as Patch3: accept token ID prompts for completions\n  and reject invalid JSON request bodies with HTTP 400; the shipped\n  example and its tests now match the branch HEAD."}