linker設定rpath的選項範例如下

1

gcc -Wl,-rpath=/opt/nvidia/deepstream/deepstream-6.4/lib test.cpp

RPATH有一個特殊識別符號$ORIGIN，表示執行檔的所在目錄

1

gcc -Wl,-rpath='$ORIGIN/../lib' test.cpp

查看函式庫的rpath

1

readelf -d libnvds_rest_server.so |head -20

或是

1

objdump -x libnvds_rest_server.so |grep 'R.*PATH'

CMake使用兩個變數來控制RPATH：INSTALL_RPATH和BUILD_RPATH，install步驟的rpath是INSTALL_RPATH

1
2

SET_TARGET_PROPERTIES(nvds_rest_server
    PROPERTIES INSTALL_RPATH "/opt/nvidia/deepstream/deepstream-6.4/lib")

cmake的另一個設定RPATH的方法
https://discourse.cmake.org/t/library-rpath-resolution/887/7
SET(CMAKE_SKIP_BUILD_RPATH FALSE)
SET(CMAKE_BUILD_WITH_INSTALL_RPATH TRUE)
SET(CMAKE_INSTALL_RPATH “${ORIGIN}”)

參考:
https://www.cnblogs.com/chanyuantiandao/p/16553199.html

查詢硬碟空間

1

du -sh ./*

有時候硬碟空間可能因為資料夾沒有mount，所以雖然占用硬碟空間但是du無法顯示出來，可以使用下面的指令來查看:

1

shopt -s dotglob && sudo mount / /mnt -o bind && du -sh /mnt/* && umount /mnt

參考:
https://www.reddit.com/r/linuxquestions/comments/1ed61ri/root_directory_full_despite_du_adding_up_to_55_of/

https://www.onejar99.com/linux-command-du/

關閉程式前發送EOS事件給pipeline的方法

1
2
3
4
5
6
7

// 發送EOS訊號讓filesink可以正常結束
gst_element_send_event(appCtx[i]->pipeline.pipeline, gst_event_new_eos());
msg = gst_bus_timed_pop_filtered(GST_ELEMENT_BUS (appCtx[i]->pipeline.pipeline), 
GST_CLOCK_TIME_NONE, 
(GstMessageType) (GST_MESSAGE_ERROR | GST_MESSAGE_EOS));//gst_bus_timed_pop_filtered 是synchronously取得訊號，如果沒有訊號會一直等待並把執行續卡住
gst_element_set_state (appCtx[i]->pipeline.pipeline, GST_STATE_NULL);
g_main_loop_quit (main_loop);

https://gstreamer.freedesktop.org/documentation/gstreamer/gstelement.html#gst_element_get_state

https://gstreamer-devel.narkive.com/g1omJFAj/gst-devel-graceful-exit-shutdown-of-a-pipeline#post3

https://stackoverflow.com/questions/38170733/gstreamer-not-flushing-to-the-filesink

範例

https://gitlab.freedesktop.org/freedesktop/snippets/-/snippets/1760

https://gist.github.com/CaptainOnly/f88cbb6ac5fdeae80200fbb3f7be695c

用nmtui設定

顯示目前路由表

1

ip route

用ip route add設定

1

ip route add 10.179.102.0/24 via 10.179.20.254 dev enp75s0f1

https://docs.redhat.com/en/documentation/red_hat_enterprise_linux/8/html/configuring_and_managing_networking/configuring-static-routes_configuring-and-managing-networking#proc_configuring-a-static-route-by-using-nmtui_configuring-static-routes

用nmtui設定

https://docs.redhat.com/en/documentation/red_hat_enterprise_linux/8/html/configuring_and_managing_networking/configuring-static-routes_configuring-and-managing-networking#proc_configuring-a-static-route-by-using-nmtui_configuring-static-routes

如果硬碟沒有分割區，可以用以下指令擴大硬碟區域:
https://unix.stackexchange.com/questions/643026/how-to-resize-a-disk-without-partition

1
2

# resize2fs /dev/[DISK_ID][PARTITION_NUMBER]
resize2fs /dev/sdc

https://tech.sars.tw/posts/gcp-linux-enlarge-disk-size/

https://www.tecmint.com/parted-command-create-linux-partitions/

https://godleon.github.io/blog/Linux/Linux-extend-lvm-from-unused-space/

LVM說明
https://linux.vbird.org/linux_basic_train/centos8/unit14.php

打開ssh登錄日誌

1

nano /etc/ssh/sshd_config

如果硬碟沒有分割區，可以用以下指令擴大硬碟區域:
https://unix.stackexchange.com/questions/643026/how-to-resize-a-disk-without-partition

1
2

# resize2fs /dev/[DISK_ID][PARTITION_NUMBER]
resize2fs /dev/sdc

https://tech.sars.tw/posts/gcp-linux-enlarge-disk-size/

https://www.tecmint.com/parted-command-create-linux-partitions/

https://godleon.github.io/blog/Linux/Linux-extend-lvm-from-unused-space/

LVM說明
https://linux.vbird.org/linux_basic_train/centos8/unit14.php

可以用symble link的方式，將服務檔放到/etc/systemd/system/目錄下，讓systemd管理服務。

更改service檔後，需執行以下指令，讓systemd重新載入設定檔:

1
2
3

sudo systemctl daemon-reload
sudo systemctl enable appname
sudo systemctl start appname

移除服務:

1
2
3
4
5
6
7

systemctl stop [servicename]
systemctl disable [servicename]
rm /your/service/locations/[servicename]
# and symlinks that might be related
rm /your/service/locations/[servicename] 
systemctl daemon-reload
systemctl reset-failed

查看服務整的log

1

journalctl -u appname

參考:
製作fastapi服務
設定檔重載
symbolic link
移除服務

載入Region_TRT plugin

必須要載入libnvinfer_plugin，可以用python的trt.init_libnvinfer_plugins載入，

1
2

TRT_LOGGER = trt.Logger(trt.Logger.ERROR)
trt.init_libnvinfer_plugins(TRT_LOGGER,"")

https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Plugin/IPluginRegistry.html#tensorrt.init_libnvinfer_plugins

印出目前已經載入的plugin

PLUGIN_CREATORS = trt.get_plugin_registry().plugin_creator_list
for plugin_creator in PLUGIN_CREATORS:
print(plugin_creator.name)

參考並修改:https://developer.nvidia.com/zh-cn/blog/tensorrt-custom-layer-cn/

tensorrt Region layer 說明書

https://docs.nvidia.com/deeplearning/tensorrt/api/c_api/structnvinfer1_1_1plugin_1_1_region_parameters.html#ad2c6bba4f07221add9b6d9abf0a4e312

inference範例參考:
https://leimao.github.io/blog/TensorRT-Python-Inference/