LNet Unit Test Infrastructure (LUTF) Requirements

Goals

• Test LNet via the Dynamic LNet Configuration interface
• Provide an easy to use and expandable infrastructure to create and add test scripts
• Provide a method to easily parse results and understand the health of the LNet Code.

Background and strategic fit

Currently there exists very little testing for LNet. With the expected LNet features to be added in the future, it is important to be able to test LNet as thoroughly as possible, as well as be able to regression test existing functionality. The LUTF fulfills that need.

Overview

The LUTF provides a master/slave work flow. The LUTF Master is deployed on one node and the Agents are deployed on the test nodes. The master controls the agent by directing them to execute python tests scripts. The master manages the interactions between the different LUTF agents.

Each agent collects the result from the test script and transfers it to the master. The master parses and summarizes the results in a YAML file.

The LUTF provides infrastructure for the test scripts to pull information about the different nodes on which the agent runs. This way the script writer only concentrates on writing testing logic, rather than writing code to run the test scripts or collect results. It also provides an abstraction layer, so that the scripts don't need to have prior knowlege of the nodes the tests will run on.   

Automating the LUTF

To increase the chances of people using the LUTF, its interface must be user friendly. Here are the key aspects that must be streamlined

1. Deploying the LUTF on a system, which includes installing any dependences:
   1. The LUTF shall be included in Lustre's test rpm. So installing that RPM will automatically install the LUTF
   2. The LUTF dependencies. The LUTF shall provide a script which is simple to execute to install all dependencies
   3. The LUTF shall officially support CentOS only.
2. Starting the LUTF can be done in two ways, a fully automated way or a more manual way. For the manual way:
   1. The LUTF shall accept a YAML config on the master which identifies specific configuration parameters. Everything has a default value, so if no YAML config or incomplete configuration is provided then the defaults will be used.
      
      1. The master address
      2. The master port: the port the listens to agents attempting to connect
      3. The daemon flag to indicate if the master should be started in daemon mode.
      4. python script path: Is the path to find test scripts
   2. The LUTF shall accept a YAML config on the agent which identifies specific configuration parameters. Almost all parameters have a default value, except for master address, so if no YAML config or incomplete configuration is provided then the defaults will be used.
      1. The master address
      2. The master port
      3. The daemon flag
      4. The python script path
3. Starting the LUTF automatically, requires identifying the master, the agent and the test scripts to execute. The test scripts can be a directory or a specific test script.
   1. The LUTF shall accept a YAML script which identifies the IP addresses of the master, agents and the test scripts to execute
   2. The LUTF shall use that information to deploy the LUTF on the indicated nodes
      1. The LUTF shall assume that the Lustre test rpm has been installed on these nodes
   3. The LUTF shall start an instance on the master and the agents and execute the scripts

Examples of the LUTF usage

It is assumed that python 2.7.5 or greater is already installed on the target nodes.

Manual Method

# on master and agents
#> python lutf-deploy.py

# on master
#> python lutf-start.py master-config.yaml
#LUTF-PROMPT> ## Python prompt that can be used to control the agents and execute scripts

# on agent
#> python lutf-start.py agent-config.yaml

Automated Method

# on master
#> python lutf-launch.py lutf-config.yaml

Requirements

Interface Requirements

LUTF Master Requirements

LUTF Agent Requirements