What about launching a couple of VM's using kvm/qemu? (this is where I'm
going at the moment)

I'd also like to see a wiki, or some other documentation that attempts to
capture some of the multi-node failure cases, and some sort of error injection
test harness, so I can start thinking about how to recreate the failure with
a couple of VM's 'in the cloud'.

On Sun, Sep 16, 2012 at 10:11:47AM +0100, Simon Wilkinson wrote:
> There's been a lot said on the list lately on the subject of testing, but 
> very little technical information. I thought that it might be worth 
> summarising the current state of the OpenAFS test suite, and to provide some 
> suggestions for how people could help make it better. Writing tests is a 
> pretty simple way of contributing to the whole OpenAFS effort.
> 
> The problem we've had, historically, is that there's a lot of configuration 
> information hard coded into OpenAFS servers. It was hard to start them up as 
> a user other than root, hard to run them on a non-standard port, and hard to 
> provide them with configuration files from places other than the default 
> location. This is now improving - YFS has paid for, and contributed, changes 
> to the vlserver and the ptserver which allow them to be run as any user, and 
> to use non-standard database and configuration file locations. This means 
> that it is now possible to exercise these servers through the automated suite.
> 
> Also from YFS, there is a load of support code to make running, starting and 
> stopping these servers easier. This all lives in tests/common
> 
> All of this means that it should be easy to start improving our test coverage 
> of both of these servers. If anyone has a few spare moments and wants to 
> improve our testing, adding a few more tests to either the vlserver and 
> ptserver suites would be a great place to start.
> 
> Longer term, I hope to modify the fileserver and volserver so that we can run 
> these from the test suite as well. This is harder, because they have a large 
> number of preconfigured paths and assumptions that makes it tricky to launch 
> them from an arbitrary developers account and tree. However, once this is 
> done, we'll be able to launch a complete AFS cell as part of make check.
> 
> Of course, none of this will cover the cases that Jeffrey and David have been 
> talking about. Many of the bugs that we encounter in production with AFS are 
> incredibly complex - they rely on timing or load issues that just can't be 
> simulated in a single machine environment. Other tests require particular 
> versions of clients and servers, or on servers being modified to exhibit 
> particular buggy behaviour.
> 
> That said, improving make check has value, even if it just frees up resources 
> to concentrate on these trickier bugs. Even basic functionality tests in make 
> check help us ship better releases.
> 
> If you are interested in writing some tests, feel free to ask me for more 
> information!
> 
> Cheers,
> 
> Simon
> 
> _______________________________________________
> OpenAFS-devel mailing list
> [email protected]
> https://lists.openafs.org/mailman/listinfo/openafs-devel
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel

Reply via email to