With the mpm limits, the webrunner fails often to instanciate because of the error in promise below :

2020-04-28 07:45:23 slapos[4284] INFO Checking promise monitor-httpd-listening-on-tcp.py...
2020-04-28 07:45:28 monitor-httpd-listening-on-tcp.py[4284] ERROR ERROR: HTTPSConnectionPool(host='2001:67c:1254:e:8e:66f4:0:4e9c', port=8386): Read timed out. (read timeout=5)

Checking the open connection with ss, I could find many lines like :

tcp    CLOSE-WAIT  518     0        [2001:67c:1254:e:8e:66f4:0:4e9c]:8386                           [2001:67c:1254:e:8e:66f4:0:4e9c]:49540

So apache was too slow to answer, or busy doing something. Connections
to it were getting queued, making promise fail, making another "slapos
node instance" required, creating more frozen tcp connections.

For this change, I took inspiration from the webrunner apache config,
which doesn't define these parameters at all neither.

I did a fast performance check, with this commit, when apache is started
it consumes ~40Mo. I didn't check before this commit, but I doubt it
would be much less than that.

log accesses (the logrotate conf already exists for them), and
remove the pattern "<IfDefine Port>" which was used when this
apache conf was extended by the software's apache extending
the monitor stack

Profiles for nexedi/erp5!948

Testnode expose the log folder for test suite, so that they can expose more logs and snapshots.

Because these can be big, we expose a new "keep-log-days" option so that we can se shorter delay for test nodes running software release tests.

Another important visible change is that testnode now automatically restart when configuration is changed.

See merge request !731

Allow to configure the number of days to keep logs

Use NameWidth=* so that long filenames are never truncated

Use SuppressColumnSorting to disable the ablity to sort so that we can
wget -r all the logs for local analysis

Since 2de4c80f generic_mysql does not
sleep so long between retries

This recipe now uses strings for ip

See merge request !741

See merge request !720

See merge request !739

TODO: There are still direct references to specific version of Python in:
  - slapos component
  - bazel templates
  - nextcloud & theia SR

if last_backup was an int, the script failed with this error:

result undefined, global datetime = <module 'datetime' from '/srv/slapgrid/
slappart6...6d33cee74/lib/python2.7/lib-dynload/datetime.so'>,
datetime.datetime = <type 'datetime.datetime'>,
datetime.datetime.fromtimestamp = <built-in method fromtimestamp of type
object>, last_backup = '1587724103'
<type 'exceptions.TypeError'>: a float is required
      args = ('a float is required',)
      message = 'a float is required'

This was broken in 8725ab1a.

fix the way of getting parameters in the new recipe.

 * fix section pbs in instance-pull-backup.cfg.in after change of
argument names.
 * fix the order of the 2 callbacks as the order is very important

See nexedi/slapos!739

Also use better argument names. The function is used in pbs recipe and
was wrongly removed in 569439d5 commit.

See nexedi/slapos!739

The '.lock' suffix is automatically added by lockfile. See
https://opendev.org/openstack/pylockfile/src/tag/0.12.2/lockfile/__init__.py#L218

See nexedi/slapos!739

extra_requires was added in slapos.core, which simplifies this part.

It is broken since commit 71ed0078
("pdftk removed from erp5").