ADVise

Dependencies

Usage

Install ADVise as follows:

pip install git+https://github.com/stackhpc/ADVise
mkdir working-dir && cd working-dir

Mungetout utility

To Download the introspection data (or use kayobe overcloud introspection data save instead):

m2-collect

It can be useful to limit the number of nodes for debugging purposes:

m2-collect --limit 4 -vv

To extract the introspection data and process it ready for ADVise input:

m2-extract introspection-data/*.json

This will have created the directories: extra-hardware, extra-hardware-json and extra-hardware-filtered. The contents of these files is as follows:

extra-hardware: input for ADVise
extra-hardware-json: unmodified extra-hardware data
extra-hardware-filtered: extra-hardware data stripped of all unique IDs. This can be used with the diff tool to look for differences between nodes. You can identify a group of similar servers using ADVise. Select one node from this group and one outlier and do a diff between them. You will have to grep for the system id in the extra-hardware data. The file names are consistent across all of the directories.

ADVise

advise-process -I ipmi -p 'path-to-output_dir/data/extra-hardware/*.eval' -o 'path-to-output_dir'

The ADVise tool will output a selection of results found under output_dir/results these include:

.html files to display network visualisations of any hardware differences.

The folder Paired_Comparisons which contains information on the shared and differing fields found between the systems. This is a reflection of the network visualisation webpage, with more detail as to what the differences are.

_summary, a listing of how the systems can be grouped into sets of identical hardware.
_performance, the results of analysing the benchmarking data gathered.
_perf_summary, a subset of the performance metrics, just showing any potentially anomalous data such as where variance is too high, or individual nodes have been found to over/underperform.

advise-visualise -output_dir 'path-to-output_dir'

The ADVise tool will also launch an interactive Dash webpage, which displays the network visualisations, tables with information on the differing hardware attributes, the performance metrics as a range of box-plots, and specifies which individual nodes may be anomalous via box-plot outliers. This can be accessed at localhost:8050.

Note

ADVise revives the package cardiff from https://github.com/redhat-cip/hardware/.

Author: Erwan Velu

ADVise integrates the package mungetout from https://github.com/stackhpc/mungetout.

Author: Will Szumski

This project has been set up using PyScaffold 2.5.11. For details and usage information on PyScaffold see http://pyscaffold.readthedocs.org/.