ファイル
rocm-systems/projects/rdc/python_binding
Bill(Shuzhou) Liu e16a8bcaf5 Identify GPUs using PCI device identifier in RDC Prometheus plugin
Add a new option --enable_pci_id to Prometheus plugin, which will map
the GPU index to the PCI Device Identifier.

Change-Id: I38a2a7e4841975da095391002397d4515ffb8e0d


[ROCm/rdc commit: 23ab2c0671]
2022-05-05 09:16:05 -04:00
..
2020-08-17 14:09:37 -05:00
2020-12-01 10:56:36 -05:00
2020-11-10 14:26:49 -05:00
2020-11-10 14:26:49 -05:00
2020-08-17 14:09:37 -05:00
2022-04-27 14:38:48 -04:00

Quick start

If you do not have the RDC installed, please specify the RDC library path using:

$ export LD_LIBRARY_PATH=<rdc_libs_path>

Then you can run RdcReader in python_binding folder:

$ python RdcReader.py

Prometheus plugin

Install the prometheus_client:

$ pip install prometheus_client

Start the rdcd with auth and then run plugin to connect to it:

$ python rdc_prometheus.py

Check the options of the plugin:

$ python rdc_prometheus.py --help

Verify the plugin is running:

$ curl localhost:5000

In the managment computer, install the Prometheus from https://github.com/prometheus/prometheus

Modify the file prometheus_targets.json to add the compute nodes running the plugin. Start the Prometheus

$ prometheus --config.file=<full path of the rdc_prometheus_example.yml>

Browse to localhost:9090 in the managment computer for metrics from RDC.