p17860254
/
categraf
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
categraf/inputs/nvidia_smi
History
yushuangyu b004dc4c61 1. add kafka metrics collector 
2. change Sample funcs from package inputs to types to avoid cycle imports
3. add runtimex.Stack to print stack detail when panic
 		2022-07-03 22:01:41 +08:00
..
README.md add TOMCAT README 2022-06-13 12:18:09 +08:00
builder.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
csv.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
fields.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
nvidia_smi.go 1. add kafka metrics collector 2022-07-03 22:01:41 +08:00
parser.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
scrape.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
types.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00
util.go rename plugin net_response and nvidia_smi 2022-05-14 16:35:10 +08:00

README.md

nvidia_smi

该采集插件的原理，就是读取 nvidia-smi 的内容输出，转换为监控数据上报。是把 nvidia_gpu_exporter 的代码给集成过来了。

Configuration

配置文件在 conf/input.nvidia_smi/nvidia_smi.toml

# # collect interval
# interval = 15

# 下面这个配置是最重要的配置，如果要采集 nvidia-smi 的信息，就打开下面的配置，
# 给出 nvidia-smi 命令的路径，最好是给绝对路径
# 相当于让 Categraf 执行本机的 nvidia-smi 命令，获取本机 GPU 的状态信息
# exec local command
# nvidia_smi_command = "nvidia-smi"

# 如果想远程方式采集远端机器的 GPU 状态信息，可以使用 ssh 命令，登录远端机器
# 在远端机器执行 nvidia-smi 的命令输出，通常 Categraf 是部署在每个物理机上的
# 所以，ssh 这种方式，理论上用不到
# exec remote command
# nvidia_smi_command = "ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null SSH_USER@SSH_HOST nvidia-smi"

# Comma-separated list of the query fields.
# You can find out possible fields by running `nvidia-smi --help-query-gpus`.
# The value `AUTO` will automatically detect the fields to query.
query_field_names = "AUTO"

TODO

GPU 卡已经关注哪些监控指标，缺少监控大盘JSON和告警规则JSON，欢迎大家 PR