已关闭
[Bug-Report|缺陷反馈]: ubx上面混合使用rdma/sdma/udma失败，样例执行失败 #536
wangmengmengwang65667创建于 22 天前关闭于 1 天前
22 天前 将 suqwe 设为负责人
Vector
22 天前 评论：
22 天前 评论：
经分析，是因为当前udma使用的rootinfo未适配UBX机型，后续将进行修复
22 天前 添加了label：bug-report
Vector
21 天前 评论：
21 天前 评论：
当前UDMA默认从路径 /etc/hccl_rootinfo.json 读取机器拓扑信息。若该路径下不存在对应文件，则由shmem自动生成rootinfo。然而，当前自动生成的逻辑尚未适配UBX机型，在CLOS组网场景下，netlayer字段被错误地设置为 TOPO_FILE_DESC，从而无法正确识别网络拓扑信息，最终引发初始化失败。
20 天前 关联了pull request：[WIP]fix: route mixed UBX topo-file-desc clos links
8 天前 关联了pull request：[release/v1.6.0] fix: route mixed UBX topo-file-desc links (#740)
1 天前 添加了label：wait-feedback
suqwe
1 天前 评论：
1 天前 评论：
当前issue失败的原因是ubx 的rootinfo文件无法区分出clos和fullmesh端口。 导致之前的方案依赖rootinfo判断fullmesh还是clos会出现失败。
解决方案：当前在这种场景，会通过topo文件进行区分是否是clos还是fullmesh。从而避免判断失误，导致建链失败问题。
目前pr已经合入，请进行回归验证。’
1 天前 issue状态由 进行中 改变为 已完成
1 天前 关闭了 issue
1 天前 添加了label：resolved
Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.
Describe the current behavior / 问题描述 (Mandatory / 必填)
在rdma样例的基础上，构造udma/sdma场景，执行样例失败
Environment / 环境信息 (Mandatory / 必填)
1825/ubx环境
Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)
在rdma样例的基础上，构造udma/sdma场景，执行样例失败
Describe the expected behavior / 预期结果 (Mandatory / 必填)
样例执行成功
Related log / screenshot / 日志 / 截图 (Mandatory / 必填)
2026-08-27 11:48:01.436308 info 1335702 pid[1334939][SHM SHMEM acc_tcp worker.cpp:193] Worker AccWrk:0 error Aa ab,_ 第23项,共 ×个
2026-08-27 11:48:01.607152 info 1335703 pid[1334939][SHM SHMEM acc_tcp_worker.cpp:193] Worker AccWrk:1 49 项
2026-08-27 11:48:01.935760 info 1335701 pid[1334939] [SHM SHMEM acc_tcp_link delay_cleanup.h:129] AccDelay Cleanup thread thread exiting
8T ΡITeDОT uaаM3аq NUTI ysаw 3DauTp u waw x1SW :TdwT /adwa XISW [vz:ddo /1dwa JaqsTSau waw x1SW WBWHS WHS] [6£6VE£T]pTd 6£6VEET Snqap V6V9£6 TO:8V:TI LZ-80-97O7
and 1
2026-08-27 11:48:01.106733 info 1334941 pid[1334941] [SHM SHMEM topo reader.cpp:1037] Clos route selected, myRank 3, targetRank 0, routeIdx 0 netLayer 3, netIn
stanceId "cluster”, planeId "planeO”, portCount 1, localEidIndex 8, remoteEidIndex 8
2026-08-27 11:48:01.322112 error T - 3au 18 xapuT 0I3 JoJ paTTeJ apeaJOquTodpugwwo>H [ZVST:ddo JaSeuew quodsueu3TewpnTabTAap WBWHS WHS] [TV6VEET]pTd [V6VEET
2026-08-27 11:48:01.322115 error T аI3 U4TM 0 xpT аqnou g uaad JoJ paTTeJ quTodpuganeau) [Z87T:ddo JaSeuew quodsueuq ewpn aoT^ap WBWHS WHS] [TV6VEET]pTd TV6VEET
ndex 8
2026-08-27 11:48:01.322121 error 1334941 pid[1334941] [SHM_SHMEM device_udma_transport_manager.cpp:320] PrepareOpenDevice failed.
2026-08-27 11:48:01.322122 error t- :1au (7 :ad/1 (paTTeJ JaSeuew quodsueuq uado [0g:ddo uaSeuew quodsueun aqtsodwo) WBNHS WHS] [TV6VEET]pTd TV6VEET
.ssabons s[auueup wwoDy /ouqsad [6T/T:ddo JaSeuew qJodsueu ewpn apTAaP W3WHS WHS] [TV6V£ET]pTd TV6VEET OJuT VZTZZE TO:8V:TT L7-80-97OZ的
.ssabons /uowaw JagsT8aJun [67/T:ddo JaSeuew qJodsueug ewpnTaoT^ap WBNHS WHS] [TV6VEET]pTd TV6VEET OJUT SZTZZE TO:87:II LZ-80-97OZ
razyleuty quodsueun ewps £ [ZTE:ddo uaSeuew quodsueuqTewps aoTAap WBNHS WHS] [TV6VEET]pTd TV6VEET OJUT 8ZEZSE TO:81:TT LZ-80-97OZ
.ssabons sTauueyp wwoby /ou1sad [6T/T:ddo JaSeuew quodsueu3 ewpn aつTAap WBNHS WMHS] [TV6VEET]pTd TV6VEET OJUT OEEZSE TO:81:II LZ-80-97O7的
2026-08-27 11:48:01.352331 info 1334941 pid[1334941] [SHM_SHMEM device_udma_transport_manager.cpp:1749] Unregister memory success.
2026-08-27 11:48:01.352333 error 1334941 pid[1334941] [SHM_SHMEM mem_entity_default.cpp:907] Failed to open device, ret: -4
2026-08-27 11:48:01.352335 error 1334941 pid[1334941][SHM_SHMEM mem_entity_default.cpp:70] init transport manager failed
2026-08-27 11:48:01.352338 error 1334941 pid[1334941] [SHM_SHMEM mem_entity_entry.cpp:216] initialize entity failed: -4
2026-08-27 11:48:01.352340 info 1334941 pid[1334941][SHM SHMEM mem entity default.cpp:33] Deconstruct MemEntity begin, try to release resource.
Special notes for this issue/备注 (Optional / 选填)