LTP Automation ~~~~~~~~~~~~~~~ The LTP automation scripts are intended to completely automate the updating, running, and reporting of an LTP test run. The ltpupdate script simply produces a copy of the latest CVS code and packages it up as a tgz file and uploads it to the designated LTP host machine. The ltp_master script first kicks off the test (this could possibly be controlled by another script say after a new release for example). This script uploads the ltprun script (which does the actual test run) to each machine specified. Currently I use this mostly as a single shot run. I envision a time to come when I have enough machines to run a copy of LTP against a test farm of say the same distro on different machine configurations, or on every machine I have control over for example. The ltprun scripts does what one would usually do on a system to be tested, download the code, compile, execute, but it also handles the reporting process as well. It will upload the data to the LTP host to be collected later. The ltp_check script does the brunt of the results processing. It produces a results.out file which has a summary of the system under test, the success (or failure) of the test and if there are failures the output from each testcase. Here's the scenario (pardon the ASCII art); ---------- ------ ----------- |master| |test| |ltp_host| ---------- ------ ----------- | | | +-------------+---------------------+ I've divided it up into 3 machine types viz, "master", "test" and "ltp_host". For purposes of this documents lts consider we have one test machine. ltp_master script ~~~~~~~~~~~~~~~~~ The ltp_master script is the driver that runs LTP testsuite on a machine of choice, it takes a single argument ie the name of the victim machine ltp_master can be invoked in the following 3 ways: #1 ltp_master -f system_file - system_file is an ASCII file containing the list of victim machines. #2 ltp_master system.name.dom.com - space separated names of victim machines. #3 ltp_master - will prompt user for the names of the victim machines. I reccomend using scheme #2, this will help seperate different distros, platforms etc, for example, We might say have a file with RedHat machines, or all SuSE, or all machines of a certain platform (say zSeries). Regardless of how it's invoked ltp_master does the following on each victim machine. - FTP connect to the test system and uploads ltprun - Telnet to each system and kick off ltprun ( runs under nohup ) ltprun script ~~~~~~~~~~~~~ ltprun is where bulk of the automation of ltp is done. It does the following. - downloads the ltp test suite tarfile (ltp.tgz) from ltp_host. - uncompress the test suite and compile it. - runs the test suite. - tars up the results. - uploads results back to ltp_host. In case any of the get, build, run fails, the script will still upload the results back to the ltp_host machine, so you don't have to try to remember which runs were kicked off and correlate that with which runs were reported. ltp_check ~~~~~~~~~ ltp_check lives on the master machine, this script has to be run as root user and it saves the ltp test results in /root/ltp_results. ltp_check does the following; - FTPs to ltp_host and downloads all of the *-ltpoutput.tgz files - Creates a results directory for each machine (using the short hostname) and for each run (based on timestamping) - Uncompresses the gathered result files into the new directory - parses results gathered, saves it as results.out file in the same directory. The results are sorted by system name and then run time. Note: - Check through the scripts for appropriate variables that you may need to change in order to customize further, they should be at the beginning of the file. - This "fire and forget" might not work 100%. But if you find a test (like pth_str02) that hangs you can log in as root and do something like; killall pth_str02 a few times and the test should pick up where it left off. Please send comments or suggestions to: Jay Huie wjhuie@us.ibm.com Linux System Test Phone: 845-435-8164