Re: [Merge] lp:~michael-abbott/epics-base/dynamic-array into lp:epics-base

[Ralph Lange <Ralph.Lange@bessy.de>]

On Thu 24 Jun 2010 11:15:36 Andrew Johnson wrote:
> On Thursday 24 June 2010 08:31:54 Ralph Lange wrote:
>   
> > Even more definitely it's not an easy task, as acctst involves things
> > like pulling and reinserting the network plug in the IOC to check tcp
> > circuit timeout/cleanup etc..., stuff that is not easily automated.
> >     
>
> .. and note that pulling the network plug is *not* the same as killing the IOC 
> process on Linux, since the latter results in the OS cleaning up the socket 
> for you whereas pulling the plug means that all packets stop.  It might be 
> possible to suspend the process to simulate it, but I'd want someone to 
> compare a couple of network traces to make sure it behaves exactly the same 
> way.
>   

For a serious result, acctst hast to be run on a reasonable set of 
different architecture/OS combinations for client and server.
That set should probably include the major host platforms (Linux 
32/64bit, Windows, MacOS, Solaris), and a set of different typical 
target architectures (68k, PPC, ARM) running vxWorks (5 and 6) and RTEMS.
If we have a working (at least semi-) automated test suite, we could and 
should split this over multiple institutes. The more combinations we get 
covered, the better. There should be an automated way of submitting test 
results to a central result matrix. I guess there are frameworks around 
that allow this kind of workflow.

Personally I always pull the plug for VME targets and suspend/continue a 
softIOC for the Linux target. I'm not sure that the effect is really 
100% the same, but at least for acctst it looks similar enough to pass 
all tests. :-)
In a test rack you could power cycle a mini switch to reliably shut down 
the network.

Ralph