See merge request nexedi/slapos!1027

This was not implemented and is not needed because we can put a default caddy
shared instance in front of an instance of software/nginx-push-stream to have
a working server side event solution.

For a simple software with only one software type like this one, using a
recipe like switch-softwaretype is not necessary.

URITemplate is more suitable because it also describe how we expect the clients
to pass parameters

inotifyx is not used by this software

softwaretype recipe is deprecated

These variables were never modified, so it's easier to just "hardcode" them
where they are used.

with a self sign certificate for now

config generated using https://ssl-config.mozilla.org/#server=nginx&version=1.19.2&config=modern&openssl=1.1.1k&hsts=false&ocsp=false&guideline=5.6

When EventSource from browser, it's not possible to pass basic authentication.
Some polyfills such as https://www.npmjs.com/package/event-source-polyfill
allow to pass custom headers, but it seems easier to rely on native EventSource.

We did not really need basic authentication in the first place, because in our
use cases we don't send confidential information over SSE. Also, we can use
hard to guess channel IDs, so basic authentication does not add much.

Buildout and SlapOS convention is to use "-" not "_"

cf https://slapos.nexedi.com/slapos-Tutorial.Create.Software.Release#:~:text=Buildout%20Profile%20Conventions

This also fixes a md5sum that got out of sync

This part was wrongly inheriting from nginx, but it should inherit from
nginx-common instead, so that it is not shared, because as it installs from a
git clone we can not share it.

It would be possible to make it shared by downloading a git archive and using
a dummy slapos.recipe.cmmi recipe for nginx-push-stream-module parts, so that
it is also shared, but since nothing depends on nginx and it is small, that
does not seem to be worth it.

This reverts commit f34aeeab.

this had the side effect of installing `promise-check-free-disk-space` which
is still used - at least by caddy tests.

In the test output, there are lots of tracebacks like this one:

    "./software/caddy-frontend/test/test.py", line 534, in do_GET
        for header in config.pop('X-Drop-Header', '').split():
    AttributeError: 'NoneType' object has no attribute 'split'

this does not seem to impact the test result, but it makes the result hard to
read.

Now that slapos-node-software service does not use --all we have to use another
approach for this test which re-installs an already installed software.

Use the slapos-node-software-all services which was made for similar use cases.

This was changed as part of b049c057 (upgrade to slapos.toolbox version
0.123, 2021-07-19), but this is not a slapos.toolbox promise plugin, it's
still an old style promise.

when running on test nodes some promise fail because testnode itself has
monitoring and the promise connects to testnode's own monitoring.

This reveals a real problem, but we need to fix testnode first to be able
to use this.

With gcc 6.3.0 we see build failures:

    error: no matching function for call to 'tesseract::STRING::STRING()'

According to https://github.com/tesseract-ocr/tesseract/issues/3210#issuecomment-774599019
minimal gcc version is version 7.

This probably started to break since d6dd75df (component/ghostscript:
version up 9.54.0, 2021-05-25) and 5f616dd7 (component/tesseract: version
up 4.1.1, 2021-05-20).

This gcc is needed for both tesseract and ghostscript, because ghostscript
build again its embedded copy of tesseract.

In b049c057 (upgrade to slapos.toolbox version 0.123, 2021-07-19), we
translated check-secure = 1 to config-http-code = 401, but check-secure = 1
also accepted responses with status code 200.

Thoses promises where checking servers without authentication enabled, they
reply with 200, so update the promises accordingly to expect 200.

This fixes a username/password confusion from c536eb7c (stack/monitor:
change frontend url promise to use username/password, 2021-07-26)

This change has no effect on stand-alone setups (which is what is
automatically deployed).
On setups configured to replicate from another database, it allows them to
takeover the role of the primary server, which allows rebuilding a
redundant setup. On such setup, the cost of this option is the extra I/O
load of binlogs being written to (in addition to the relay log and
database), and the corresponding disk space use.

Commit 7d1ea024 (version up: zc.buildout 2.7.1+slapos013) switched buildout to
not use egg wheels, even for wheels that are source-only without any binaries.
As the result some eggs stopped to install because they setup_require other
eggs and neither buildout handles setup_requires automatically, nor there is
explicit setup_require declaration in corresponding components.

An example of such egg is python-dateutil: it setup_requires setuptools-scm,
and without switching python-dateutil to zc.recipe.egg:custom with
setup_requires the build fails.

The way 7d1ea024 used to workaround this is to mark all such eggs with :whl
suffix marking them that "it is ok to use wheel for this particular egg". The
complete list of eggs for which such markings was done in 7d1ea024 is:

    ipykernel
    plantuml
    mock
    PasteScript
    backports.functools-lru-cache
    cliff
    stevedore
    CacheControl
    pytest-runner
    jsonschema
    lockfile
    python-dateutil

However this list is not really complete because after 7d1ea024 the build
started to fail for e.g. zodbtools test:

        Traceback (most recent call last):
          File "<string>", line 10, in <module>
          File "/srv/slapgrid/slappart16/srv/testnode/dfp/soft/c670f74c8de08964f63c4b9f7d48c3d1/eggs/setuptools-44.1.1-py2.7.egg/setuptools/command/easy_install.py", line 2321, in main
          ...
          File "/srv/slapgrid/slappart16/srv/testnode/dfp/soft/c670f74c8de08964f63c4b9f7d48c3d1/eggs/setuptools-44.1.1-py2.7.egg/pkg_resources/__init__.py", line 786, in resolve
            raise DistributionNotFound(req, requirers)
        pkg_resources.DistributionNotFound: The 'setuptools-scm' distribution was not found and is required by the application
        An error occurred when trying to install pytest 4.6.11. Look above this message for any errors that were output by easy_install.
        While:
          Installing zodbtools-python.
          Base installation request: 'zodbtools[test]'
            Requirement of zodbtools[test]: mock
            Requirement of zodbtools[test]: pytz
            Requirement of zodbtools[test]: freezegun
            Requirement of zodbtools[test]: pytest
            Requirement of zodbtools[test]: dateparser
            Requirement of zodbtools[test]: six
            Requirement of zodbtools[test]: pygolang>=0.0.0.dev6
            Requirement of zodbtools[test]: zope.interface
            Requirement of zodbtools[test]: zodburi
            Requirement of zodbtools[test]: ZODB
            Requirement of mock: funcsigs>=1
            Requirement of mock: six
            Requirement of freezegun: python-dateutil>=2.7
          Getting distribution for 'pytest==4.6.11'.
        Error: Couldn't install: pytest 4.6.11

and similarly for pygolang and nxdtest tests.

I would be more comfortable if the fix would be to teach buildout to either
handle setup_requires automatically, or to automatically install wheels if the
wheel is source-only. However, given the way - that 7d1ea024 started to use
to handle such situations - let's fix this via explicitly marking pytest, and
other related eggs that setup_require something and are source-only, as :whl.

NOTE we did not notice the problem earlier because on testnodes, after the
switch to new buildout, pytest egg was already installed and incremental
rebuild did not need to reinstall pytest egg. So most of pygolang and zodbtools
test results remained in "ok" state, and only zodbtools-kirr, that was recently
reconfigured or changed testnode, indicated this issue. That's why the problem
remained unnoticed for a long time.

/reviewed-by @jerome
/reviewed-on nexedi/slapos!1026

See merge request nexedi/slapos!1022

This introduces some breaking changes to the check_url_available
promise:

- http_code has been renamed to http-code
- check-secure = 1 has been deprecated; http-code = 401 should be used
  instead

See merge request nexedi/slapos!1011

also apply the change in software/theia/test/dummy/software.cfg for consistency

  This is a partial port, just to have more tests that
  weren't afected by the patches for keep apache as entry-point
  of the backend.

See merge request nexedi/slapos!1009

This reverts commit cf99b892