This mysql_update service retries every 30 seconds, but what usually
happens is that it starts at same time as mariadb, make a first attempt
at connect which fails because mariadb is not ready, sleep for 30
seconds, reties and succeed. Which is not efficient because we don't
need to wait for so long.
Instead of sleeping for 30 seconds between each retries, start with a
short delay and increase each time, but not more than 30 seconds.

We want to wait for mysql_upgrade pseudo-service to have created users
and installed groonga, not just that something listen on the port.

Instead of writing the instructions ourselves, source mroonga provided
install script.

shared support is now integrated in slapos.cfg

This fixes commit f9d5bde3.

We also often have MAKEFLAGS with -l option. For b2, this option sets
a time limit in seconds for each job, which led to random failures.

As found by jm@nexedi.com :

>>> with open('/dev/null') as f:
...   subprocess.check_call(('echo', 'foo'), stdout=f)

echo: write error: Bad file descriptor

To easily check regressions in profiles and check that fonts in PDF are what we expect.

See merge request !715

See merge request !718

Check what fonts are used when converting to PDF, using HTML as input.

Checking that something listen on the port is not enough to know when
instance is ready, because apache listen on the port very early. Also,
this does not detect problems on backend.

This fixes a regression in commit a7a36453
("Clean up some zc.recipe.egg sections installing the same scripts/interpreter").

This means the user don't have any instance at this point.

See merge request !717

See commit 1e514399.

The fixes a conflict with 'runner-exporter' due bugs in cns.recipe.symlink

Anyway, a much better idea would be to add ~/software_release/bin to PATH.

This reverts commit 4f4354c3.
It duplicated commit d13b42b7.

It was in conflict with resilient-request-*-pseudo-replicating-*-public-key.

  This means the user don't have any instance at this point.

Even when no referencing section is instanciated.

When full ERP5 / Wendelin software release compilation is done this requires more than 2GB of RAM due to groonga compilation constraints (cc1 gets OOM killed). Thus increase memory for DeploymentTests only.

[duplicity] has at least 2 issues:
- use os.chdir
- creation of paths outside of location

Since commit ebc1636d ("instanciate a
single admin node, by default in the first subpartition"), there was
only 1 admin node (and 1 neoctl wrapper) and this was inconvenient if
the partition hosting it lost.

Only 1 admin node does monitoring.