[Xen-devel] Ad-hoc test instructions (was Re: [xen-unstable test] 123379: regressions - FAIL)

[Ian Jackson <ian.jackson@xxxxxxxxxx>]

Apropos of the irc conversation below, in particular my suggestion to
use mg-repro-setup.  Probably, an appropriate rune is

  ./mg-repro-setup -f123855 -Ejgross@xxxxxxxx 123855 
test-amd64-amd64-xl-qemuu-ovmf-amd64 guest-saverestore.2 
host=alloc:'{equiv-albana}'

(You will have wanted to
   git clone ~osstest/testing.git
   cd testing
to make yourself a working tree.  To run pieces of osstest, your
cwd should be the root of the osstest source tree.)

What mg-repro-setup will do is:

 * Make a new flight consisting of one job
     test-amd64-amd64-xl-qemuu-ovmf-amd64
   whose recipe and parameters are copied from the corresponding job
   in 123855 - and which will reuse builds from 123855 too.

 * Try to allocate a host according to '{equiv-albana}', which is an
   expression meaning "any host with the equiv-albana flag set", ie
   one of the (two) albanas.  This will have to wait for a slot,
   but as a command line user you get a highish priority.

 * Run the repro flight, including all the steps up to and including
   the one with testid "guest-saverestore.2".  This will wipe the
   albana machine allocated above and install on it the same
   versions of everything as used for the same job in 123855.

   (You can also ask mg-repro-setup to wipe and reinstall an existing
   host you have already allocated, or to reuse existing host and put
   the relevant Xen and kernel on it but without wiping it.  See the
   usage message.)

 * Email you with a report, comparing the results with 123855

After this, you will still have the relevant albana[01] machine
allocated and you may log into it etc. etc.  If you want to connect to
its serial port,
   ssh -vt serial2.test-lab sympathy -r albanaN
(from your own workstation)

You can power cycle it with
   mg-hosts power albanaN reboot

Also you may use the flight constructed by ./mg-repro-setup for your
own ad-hoc-tests:

  export OSSTEST_FLIGHT=<whatever number you were told by mg-repro-setup>
  export OSSTEST_JOB=test-amd64-amd64-xl-qemuu-ovmf-amd64

  ./ts-guest-saverestore host=albanaN debian

When you are done, you must release the host manually

  mg-allocate ^albananN

Please ask me on IRC if you have any questions.  There are also docs
in the osstest tree.  HTH.

Ian.


14:48 <juergen_gross> Okay. Already found some interesting samples, e.g. from 
                      4.9 tests. All found up to now on different hosts 
                      (silvana0, chardonnay0, huxelrebe1)
14:49 <juergen_gross> So the chances for some hardware specific bug are rather 
                      slim, other than x86 :-)
14:55 <juergen_gross> Diziet: are most tests now running with dom0 4.14 kernel?
14:55 <Diziet> Basically all the x86 tests that aren't tests of some Linux 
               version.
14:56 <juergen_gross> Okay, so it is no strange coincidents all failures have 
                      been with 4.14. :-) BTW: even 4.6 Xen has the same problem
14:58 <juergen_gross> Is it possible to run ts-guest-saverestore in a loop on a 
                      machine to have an idea how long it takes to reproduce it?
14:58 <Diziet> Yes.
14:58 <juergen_gross> I'll write a hypervisor patch to get more diagnostics then
14:58 <Diziet> The easiest way is probably to use mg-repro-setup to find a 
               host, set everything up, repro the test up to that point, and 
               then run ts-repeat-test by hand in a shell
14:59 <Diziet> You could indeed also install a new hypervisor after 
               mg-repro-setup has done its thing
15:00 <juergen_gross> I guess the best would be if you could send me mail with 
                      either the commands or a hint where to find the needed 
                      info. I should do those tests myself to learn using them, 
                      but I'd appreciate some help to start
15:00 <Diziet> Of course.
15:00 <juergen_gross> Thanks.
15:00 <Diziet> You mentioned having found some other occurrences.
15:00 <juergen_gross> Yes.
15:01 <Diziet> You might want to pick one that has a relatively simple guest 
               (and ideally not a Windows one)
15:01 <Diziet> Can you tell me the flight number and job name from that ?
15:01 <juergen_gross> Flight 123855 test-amd64-amd64-xl-qemuu-ovmf-amd64
15:02 <Diziet> t\z
15:02 <Diziet> ta I mean
...
15:09 <juergen_gross> Shows clearly the same problem
15:09 <Diziet> Indeed

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel