Problem regenerating gitlab docker images

[Andrew Bartlett]

On Wed, 2019-10-30 at 15:39 -0700, Christof Schmitt via samba-technical 
wrote:
> Hi,
> 
> while fixing the build problem with the system heimdal libraries, i
> am
> also trying to establish a gitlab test for this build. This requires
> adding the heimdal-multidev package to the Ubuntu images.
> Regenerating
> the images now fails for CentOS7 and CentOS8, which should not
> changed.
> I already fixed one problem for CentOS7 to avoid the conflict between
> nettle and compat-nettle.
> 
> Now CentOS 7 fails with:
> https://gitlab.com/samba-team/devel/samba/-/jobs/337893367
> + docker build -t samba-ci-centos7 --build-arg
> 'SHA1SUM=9dae5d8732853f0025bcf9528e94935f0e7ebada'
> bootstrap/generated-dists/centos7
> Cannot connect to the Docker daemon at tcp://docker:2375. Is the
> docker daemon running?
> ERROR: Job failed: exit code 1

Garming did this on my suggestion:

https://gitlab.com/samba-team/samba/merge_requests/803/diffs?commit_id=a6b38e4ed0f409b88ba9df765951afaab9195047

The issue is that, in order to progress work while the Gitlab.com
runners are busy, we also declare 'shared' on our own runners. 

However we haven't worked out how to do the docker-in-docker thing
there, and probably don't want to (due the security risk).

The alternative is to stop our runners declaring 'shared' and make our
jobs that can run on either just require 'docker'. 

This is probaby the better approach.  It may mean CI for older branches
take longer when gitlab.com is busy, but master is where all the action
is.  For our private CI (for security embargoed stuff) we will still
declare 'shared'.

In the meantime, just try again later, our own runners have a slower
tick cycle to try and loose the race if they can.

Andrew Bartlett

-- 
Andrew Bartlett
https://samba.org/~abartlet/
Authentication Developer, Samba Team         https://samba.org
Samba Development and Support, Catalyst IT   
https://catalyst.net.nz/services/samba