Help wanted: Linux-HA Test driver
alanr at bell-labs.com
Sun Nov 14 22:14:35 MST 1999
We have a very strong need for a test driver for Linux-HA. I thought I would
ask the list to see if someone else could be persuaded to write it because I
really don't have time to do it myself.
We really need a test driver or exerciser or whatever-you-want-to-call-it for
Linux-HA. As time goes on, this need will become more acute.
What I had in mind was a script which was capable of looking at config files,
and then examining the environment to see where resources were being served, and
comparing it to the state of the cluster and say "yes, that was right", or "no,
that was wrong".
For example, it might do an rsh to each machine and then ask each (through the
resource scripts) a "/etc/ha.d/resource.d/IPaddr 220.127.116.11 status" to
determine if that IP resource was up or not. It would do that for each resource
on each machine in the cluster after each "transition", to make sure each
resource was served somewhere, and that each was served in only one place.
These capabilities would be coupled to a random test case generator which would
take nodes down and up at random, and then track the results to ensure
For the purposes of this testing it would be OK to use rsh or ssh to run
commands remotely, as a method of stopping heartbeat and restarting it, and of
running the status scripts on remote machines. Examination of logs is probably
This script would also have to be able to track the timing in which it did
things, so that attempts to reproduce problems it uncovers would be possible.
This script would help at least three ways:
Improve the quality of the stuff going into CVS
Improve the quality of the stuff that gets released
Decrease the time I have to spend doing it
Simplify the porting job for Matt Soffen, and possibly others
Simplify verifying Linux-HA on new distributions and releases of Linux
Those of you who have suffered through the recent rash of silly bugs, can
*easily* see the need for these things.
The result should be something that others can hack on, and add test scenarios
I believe this is an VERY important job, and, if done right, can be very cool
Who's going to volunteer?
-- Alan Robertson
alanr at bell-labs.com
More information about the Linux-HA