SGLang + Mooncake 启动报错
收藏回复举报
SGLang + Mooncake 启动报错
t('forum.solved') 已解决
新人帖
发表于2026-03-30 10:35:32
0 查看

【环境信息】

root@ott2-1-00001:/# uname -a

Linux ott2-1-00001 5.10.0-182.0.0.95.r2220_157.hce2.aarch64 #1 SMP Wed Jul 30 11:31:47 UTC 2025 aarch64 aarch64 aarch64 GNU/Linux

root@ott2-1-00001:/# cat /sys/devices/virtual/dmi/id/product_name

Atlas 900 A3 SuperPoD Compute Node

root@ott2-1-00001:/# cat /usr/local/Ascend/ascend-toolkit/latest/aarch64-linux/ascend_toolkit_install.info

package_name=Ascend-cann-toolkit

version=8.5.0

innerversion=V100R001C25SPC001B232

compatible_version=[V100R001C15],[V100R001C18],[V100R001C19],[V100R001C20],[V100R001C21],[V100R001C23]

arch=aarch64

os=linux

path=/usr/local/Ascend/cann-8.5.0

root@ott2-1-00001:/#

【Mooncake 启动成功】

W0330 01:31:09.372989  1869 transfer_engine_py.cpp:77] Using default malloc/free for protocol: rdma

I0330 01:31:09.373056  1869 transfer_engine_impl.cpp:597] Metrics reporting is disabled (set MC_TE_METRIC=1 to enable)

I0330 01:31:09.373068  1869 transfer_engine_impl.cpp:105] Transfer Engine parseHostNameWithPort. server_name: 192.168.0.88 port: 12001

I0330 01:31:09.373092  1869 transfer_engine_impl.cpp:172] Transfer Engine RPC using P2P handshake, listening on 192.168.0.88:16640

root@ott2-1-00001:/# mooncake_master \

  --enable_http_metadata_server=1 \

  --http_metadata_server_port=8081 \

  --rpc_port=8082 \

  --eviction_high_watermark_ratio=0.95

WARNING: Logging before InitGoogleLogging() is written to STDERR

I0330 01:12:43.056917   419 master.cpp:819] Master service started on port 8082, max_threads=4, enable_metric_reporting=1, metrics_port=9003, default_kv_lease_ttl=5000, default_kv_soft_pin_ttl=1800000, allow_evict_soft_pinned_objects=1, eviction_ratio=0.05, eviction_high_watermark_ratio=0.95, enable_ha=0, enable_offload=0, ha_backend_type=etcd, ha_backend_connstring=, etcd_endpoints=, client_ttl=10, rpc_thread_num=4, rpc_port=8082, rpc_address=0.0.0.0, rpc_conn_timeout_seconds=0, rpc_enable_tcp_no_delay=1, rpc protocol=tcp, cluster_id=mooncake_cluster, root_fs_dir=, global_file_segment_size=9223372036854775807, memory_allocator=offset, enable_http_metadata_server=1, http_metadata_server_port=8081, http_metadata_server_host=0.0.0.0, put_start_discard_timeout_sec=30, put_start_release_timeout_sec=600, max_total_finished_tasks=10000, max_total_pending_tasks=10000, max_total_processing_tasks=10000, pending_task_timeout_sec=300, processing_task_timeout_sec=300, enable_snapshot=0, enable_snapshot_restore=0, snapshot_interval_seconds=600, snapshot_backup_dir=, snapshot_object_store_type=, snapshot_catalog_store_type=, snapshot_retention_count=2, max_retry_attempts=10, enable_cxl=0, cxl_path=/dev/dax0.0, cxl_size=8589934592

I0330 01:12:43.056970   419 master.cpp:739] Starting C++ HTTP metadata server on 0.0.0.0:8081

I0330 01:12:43.057273   419 http_metadata_server.cpp:109] HTTP metadata server started on 0.0.0.0:8081

I0330 01:12:43.057286   419 master.cpp:748] C++ HTTP metadata server started successfully

I0330 01:12:44.072237   431 master_service.cpp:353] Task cleanup thread started

【问题1:SGLang 启动报错:Failed to install Ascend transport】

root@ott2-1-00001:/# export MC_LOG_LEVEL=INFO

export MOONCAKE_TE_META_DATA_SERVER="P2PHANDSHAKE"

export MOONCAKE_PROTOCOL="ascend"

sglang serve --model-path /home/model/qwen3-4b/ \

  --enable-hierarchical-cache \

  --hicache-storage-backend mooncake

...

I0330 02:30:08.763191  3485 ascend_direct_transport.cpp:83] install AscendDirectTransport for: 192.168.0.88:16089

E0330 02:30:08.763227  3485 ascend_direct_transport.cpp:203] Call acl failed, detail: [PID: 3485] 2026-03-30-02:30:08.767.175 Invalid_Argument(EE1001): The argument is invalid.Reason: rtGetDevMsg execution failed, the context is a null pointer.

        Solution: 1.Check the input parameter range of the function. 2.Check the function invocation relationship.

        TraceBack (most recent call last):

        ctx is NULL![FUNC:GetDevErrMsg][FILE:api_impl.cc][LINE:6333]

        The argument is invalid.Reason: rtGetDevMsg execution failed, the context is a null pointer.

E0330 02:30:08.767397  3485 ascend_direct_transport.cpp:92] AscendDirectTransport: cannot allocate local segment, ret: -1

E0330 02:30:08.767405  3485 transfer_engine_impl.cpp:199] Failed to install Ascend transport

[2026-03-30 02:30:08] Mooncake Transfer Engine initialization failed.

...

本帖最后由 匿名用户 于 2026/04/02 15:07:48 编辑

我要发帖子