[Swan-dev] disabling services - ssh?

[Andrew Cagney]

I'm guessing neither of you use multiple groups of test domains, or if
you do, there's always a group of domains called east, west, ...
SSH only works for the original east, west, ... domains.  Having it
running in any other configuration is pointless - the host can't reach
the test domain's interfaces.

I'll change things to disable SSH when no guest is accessible from the
host.  I'd also encourage you both to point new developers to use
kvmsh and not ssh - its a pita to set up and doesn't work in more
complex environments.

On Wed, 12 Aug 2020 at 01:59, Antony Antony <antony at phenome.org> wrote:
>
> I started with this e-mail before Paul's request to stop this attemept.
> I guess for this round sshd will survive in default install!
>
> However, I post the e-mail to explictly document usecase of ssh for the
> future.
>
> On Mon, Aug 10, 2020 at 10:16:53PM -0400, Paul Wouters wrote:
> > On Aug 10, 2020, at 21:19, Andrew Cagney <andrew.cagney at gmail.com> wrote:
> > >
> > > I've been trimming the services installed/running from the KVM tests
> > > (zapping plymouth saved 30 minutes).  I've so far cut things down to:
> > > https://libreswan.org/wiki/Test_Suite_-_Performance#Boot_the_VMs
> > >
> > > One on my hit list is SSHD.  I don't use it.
>
> I can imagine disabling ssh would save few minutes from in a test run. No
> ssh would be ok for testing.libreswan.org which is just a testrunner host.
> However, other use cases it does not sound ideal.
> When I am interacting with the guest ssh is way more responsive than serial.
> Also fits work flow better than kvmsh.
>
> > I use it a lot. It works much better than serial ports which I always need
> > to stty sane/reset
>
> My workflow depend on ssh. And would be sad to see it disabled.
> If it is disabled, I would make an attempt to enable ssh in some optional
> way. Testing should support work flows too not just optize for speed!
>
> on test vm ssh is useful for opening multiple sessions. One session running
> tcpdump, another one runnign gdb ...  and ping. serial is not comparable to
> ssh when ssh is possible.
> as I recollect tcpdump would not write over 9pfs and have to be written to
> /tmp and copy over ssh.
>
> ./testing/utils/kvmsh.py does not support multiple sessions to guest.
>
> I ssh to swandefault(now and then) to install new kernel, just copy over
> vmlinuz file to /boot/...  or specific changes to compile install new
> iproute2. And then create the test vms.
>
> In short please leave the ssh alone! And celiberate getting down to < 2.5
> hours from 13 hours! while the number of tests increased from when it took
> 13 hours.
>
> -antony
>
> PS: I am all for adding optional hook to disable ssh, may be in
> f32-transmorgify. An optional hook to copy .bashrc, .vimrc.. from host,
> aspecified location would be useful.