undefined symbol errors for libtransformer_server.so

[kaierlong]

gcc/g++: 9.3.0 cmake: 3.20.2 protobuf: 3.13.0 hdf5: 1.12.0 lightseq: master nvidia driver: 470.82.01 cuda: 11.4, V11.4.152 GPU: Tesla T4 OS: Ubuntu 20.04 x86_64 Docker image: nvcr.io/nvidia/tritonserver:21.07-py3

I build the libtransformer_server.so refered to docs/inference/build.md which will be used for nvidia tritonserver. But there are two undefined symbol errors.

And the error info as below:

ldd -r libtransformer_server.so

	linux-vdso.so.1 (0x00007ffcf093f000)
	libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fbd2aa74000)
	libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fbd2aa6e000)
	librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007fbd2aa63000)
	libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fbd2a881000)
	libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fbd2a732000)
	libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fbd2a715000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fbd2a523000)
	/lib64/ld-linux-x86-64.so.2 (0x00007fbd43bd3000)
undefined symbol: _ZN6nvidia15inferenceserver19GetDataTypeByteSizeENS0_8DataTypeE	(./libtransformer_server.so)
undefined symbol: _ZN6nvidia15inferenceserver21GetCudaStreamPriorityENS0_37ModelOptimizationPolicy_ModelPriorityE	(./libtransformer_server.so)

c++filt _ZN6nvidia15inferenceserver19GetDataTypeByteSizeENS0_8DataTypeE
nvidia::inferenceserver::GetDataTypeByteSize(nvidia::inferenceserver::DataType)

c++filt _ZN6nvidia15inferenceserver21GetCudaStreamPriorityENS0_37ModelOptimizationPolicy_ModelPriorityE
nvidia::inferenceserver::GetCudaStreamPriority(nvidia::inferenceserver::ModelOptimizationPolicy_ModelPriority)

any help?

[kaierlong]

I build the so on docker container, and docker image is nvcr.io/nvidia/tritonserver:21.07-py3.

[Taka152]

Currently, our server code is only tested on nvcr.io/nvidia/tensorrtserver:19.05-py3, newer version triton server may not be compatible, and you need to update the server code to support the latest triton server.

[kaierlong]

@Taka152 Do you have a plan to support triton inference server 2.x?

[neopro12]

We will support it within a month

[kaierlong]

@neopro12 OK. Thank you for reply.

[kaierlong]

@neopro12 How about the development of supporting triton 2.X going? If you already finished, I will hava a try and test.

[dave-rtzr]

Same here. Any updates?

[Taka152]

WIP

On Sat, Mar 5, 2022 at 4:36 AM dave-rtzr @.***> wrote:

Same here. Any updates?

— Reply to this email directly, view it on GitHub https://github.com/bytedance/lightseq/issues/259#issuecomment-1059513128, or unsubscribe https://github.com/notifications/unsubscribe-auth/AELIZAN4UL56NMWTCI32HITU6JX45ANCNFSM5MGJ74ZA . You are receiving this because you were mentioned.Message ID: @.***>

[lilyzlt]

I have error too, tritonserver r21.10 @Taka152

[lilyzlt]

@Taka152 that seems is Protobuf problem,(I aright keep same version) cmake version 3.20.6 Protobuf version 3.20.1 I fixed this issuse, Protobuf version problem

[lilyzlt]

@Taka152 更新一下，使用USE_TRITONBACKEND ON生成对应libtriton_lightseq.so文件进行load模型或者rdd -l都可以正常； 使用server/ 下的so文件仍然报undefined symbol: _ZN6nvidia15inferenceserver19GetDataTypeByteSizeENS0_8DataTypeE 错误；