Update Release Candidate



See merge request !53

@rafael says frontend infrastructure is not yet ready to handle slave
requests for frontends which come with lab.nexedi.com frontend SR URLs.

So revert apache-frontend-URL related part of 86deb9c6 (*: slapos.git
moved to https://lab.nexedi.com/nexedi/slapos.git) back for now.

/cc @jp
/reviewed-by @rafael
/reviewed-on !52

86deb9c6 (*: slapos.git moved to https://lab.nexedi.com/nexedi/slapos.git)
went over whole tree and updated slapos.git URL but when changing
non-top-software.cfg-like files forgot to update their md5sum in
software releases.

Fixes started to appear - e.g. 8b1496c3 (kvm: fix md5sum, thanks
@alain.takoudjou), but generally the whole tree remains more-or-less
broken.

Fix it. (not tested - manually checked files and updated md5sums)

/reported-by @rafael

This updates links to slapos.git in the tree to point to new location.

We do so whole-tree except one place in stack/monitor/ :

    [download-monitor-static]
    recipe = hexagonit.recipe.download
    url = http://git.erp5.org/gitweb/slapos.git/snapshot/930be99041ea26b7b1186830e5eb56ef0acc1bdf.tar.gz
    ...

(see d8800c0b "monitor: Download statics files from snapshot")

The reason we do not update that link yet, is that 930be99041ea26b7b1186830e5eb56ef0acc1bdf
is a tree object and gitlab does not allow to dowload tree object as archives (yet ?)

So for now that link stays unconverted - and we'll think how to do with it.

Fix image display on jupyter

This fix is for bug 
https://nexedi.erp5.net/bug_module/20160212-76EAAC/view?ignore_layout:int=1
where the image base64 string was only showing. The error in this case was that the `mime_type` of content code_result was not being updated by the kernel before display.

Reported by : @Tyagov 
Please review: @Tyagov @kirr 

For testing, you can use this jupyter instance with updated kernel running : 
https://[2001:67c:1254:e:8b::335d]:8888/tree

There are also some example notebooks in the aforementioned jupyter.

See merge request !47

This fix is for bug https://nexedi.erp5.net/bug_module/20160212-76EAAC/view?ignore_layout:int=1
where the image base64 string was only shoeing. The error in this case was that the mime_type
of content code_result was not being updated by the kernel before display.

Reported by : @ivan

To pick up kirr/gitlab-ce@32fa1180

/cc @kazuhiko, @jerome
/reviewed-by TrustMe

6a4e45c5 (gitlab: Sync sidekiq service launch with upstream) claimed to
sync sidekiq startup with upstream, but missed one upstream change in
between 8.2 and 8.4:

    https://gitlab.com/gitlab-org/omnibus-gitlab/commit/2eb7634f

My bad - I should have reviewed git log for

    https://gitlab.com/gitlab-org/omnibus-gitlab/blob/master/files/gitlab-cookbooks/gitlab/templates/default/sv-sidekiq-run.erb

and I did not noticed that change because I did not and just compared
queues list visually.

Pick it up.

/cc @kazuhiko, @jerome
/reviewed-by TrustMe (I hope)

This brings slapos gitlab port up to date with current stable gitlab
version 8.4, and teaches Unicorn startup script to wait a bit for
PostgreSQL to start and be ready, before unicorn startup proceeds.

/cc @kazuhiko
/somewhat-reviewed-by @jerome
/reviewed-on nexedi/slapos!44

  This changes were squashed into a single commit and it includes:

    - install slapos_configurator by default
    - Allow define ssl authentication for a certain zope family
    - Allow define custom zope path per zope family

  Those changes are originally written by Alain Takoudjou and rewrited by Rafael Monnerat

To pick up kirr/gitlab-ce@a5965b54.

  The local changes are not yet generic enough to be shared among stack, so
  Introduce local changes here better

  There is no strong reason to use git.erp5.org to test code anymore

  slapos.package is not an egg anymore, it's implementation was replaced by
    the usage of ansible with playbooks.

As it was outlined in 5a744de7 (gitlab: Compile assets on instantiation
and make sure DB is properly setup/migrated before unicorn runs) we are
performing DB initialization and migration in pre-action as part of
unicorn startup script. But that currently has one drawback:

    if all services start at the same time - e.g. both PostgreSQL and
    Unicorn - and that is a common scenario when SR is compiled and
    instantiated / started, PostgreSQL is usually not yet ready to
    process queries from Unicorn startup script, and first-time Unicorn
    startup fails.

Until now this problem was workarounded by manually starting unicorn
second time - after some time postgresql is started and ready. But why
do it manually, if we can do the same logic automatically. So fix it:

    in Unicorn startup script wait a bit (up to 5 seconds) for
    PostgreSQL to become ready.

/cc @kazuhiko, @jerome

This is a SR to build NayuOS binary images.
More info on https://www.nayuos.com and in the README.

Merging this recipe from a standalone repository was a decision made together with @kirr and @vpelletier.

component/openblas: put `build-common-options` in one line

Include fix for openblas

See merge request !45

Even though we do not use LFS objects, builds and build artifacts,
gitlab backups corresponding directories and recreates them upon
restore.

Current state is: the directories are living under Rails root - that is
in our gitlab-work/ inside instance, and since gitlab-work is recreated
almost on every re-instantiation if we'll have some content in that
directories - it will be lost.

Don't allow even a chance for that to happen and move what is designed
to hold data to under var/ .

NOTE we create shared/lfs-objects/ and shared/artifacts/ because if we
    do not have this dirs, gitlab backup fails. Upstream code just
    assumes the directories are there.

/cc @kazuhiko, @jerome

Upstream added `pages` and `elasticsearch` queues to sidekiq

    https://gitlab.com/gitlab-org/omnibus-gitlab/commit/be6844ff
    https://gitlab.com/gitlab-org/omnibus-gitlab/commit/801d6ed7

but since those queues are used for EE-only services we do not enable
them.

/cc @kazuhiko, @jerome

As gitlab-workhorse is now serving everything, it proxies unknown-to-it
requests to unicorn, and that returns redirect to /users/sign_in for
/non-existent, not 403.

Adjust the promise and just verify whether we can get /static.css

NOTE nginx promise now effectively depends on gitlab-workhorse being up
    and running - because nginx is now configured to be only an SSL
    terminator and all requests processing is done by gitlab-workhorse and
    services below it.

    If in the future we'll need a pure nginx-is-up promise, we can
    introduce something to nginx config and verify that particular http
    endpoint.

/cc @kazuhiko, @jerome

add `-documentRoot ...` argument to gitlab-workhorse start as upstream
started to do:

    https://gitlab.com/gitlab-org/omnibus-gitlab/commit/d769f751

( gitlab-workhorse now needs to know where document root is as it now
  serves all content - e.g. static files too

  this option was introduced here:

    https://gitlab.com/gitlab-org/gitlab-workhorse/commit/1a7009e4  )

/cc @kazuhiko, @jerome

I manually reviewed

    files/gitlab-config-template/gitlab.rb.template
    files/gitlab-cookbooks/gitlab/attributes/default.rb

from omnibus-gitlab 8.4.2+ce.0-3-g68d5ee8 and picked up changes:

- unicorn OOM killer memory limits were raised up (because gitlab is now
  more memory hungry according to talks on gitlab.com)

  https://gitlab.com/gitlab-org/omnibus-gitlab/commit/f8ee4a19

- there is no default limit for nginx client_max_body_size - it just
  goes to gitlab-workhorse as is at any size

  https://gitlab.com/gitlab-org/omnibus-gitlab/commit/78862837

/cc @kazuhiko, @jerome