pc.ram key is used for the initial memory, and it can't be removed.

This promise generalizes the check_port_listening promise to include
AF_UNIX sockets in addition to AF_INET and AF_INET6 host/port sockets.

The socket family and address is deduced from the set of arguments:
- `host` & `port`: AF_INET or AF_INET6
- `pathname`: filesystem pathname address for AF_UNIX
- `abstract`: '\0' is prefixed to build an abstract AF_UNIX address

All other combinations of arguments are forbidden.

The promise creates a socket accordingly and attempts to connect to the
provided address. The promise fails if the connection fails.

The check_port_listening promise has been removed.

See merge request nexedi/slapos.toolbox!97

This reverts commit 4a992aaa.

runpromise takes arguments from command line, but also read some from
a config file, so indeed it is mandatory to know partition_folder's
value, but it is not mandatory to pass it as an argument from the
command line

So instead of a traceback we get a better error message
written by argparse. Also , it will print a better help
message

On python3 several views were not functional, because subprocess output was
retrieved as bytes, but manipulated as a string.

This adds a few tests for views, not complete because it would be too much time
consuming for now.

See merge request nexedi/slapos.toolbox!94

Simple and predictable tool to resolve massively domain names to IPs.

See merge request nexedi/slapos.toolbox!93

See merge request !90

from
https://werkzeug.palletsprojects.com/en/0.14.x/contrib/fixers/#werkzeug.contrib.fixers.ProxyFix
to
https://werkzeug.palletsprojects.com/en/1.0.x/middleware/proxy_fix/#werkzeug.middleware.proxy_fix.ProxyFix

the behavior seems the same, default config in both cases is to trust one
level of proxy.

Monitoring is about states, not exact values, so report the states, as
otherwise it results with bloating databases, very long accesses to results
and long synchronisations.

Instead of many entries like:

http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: ERROR IP IP replied in 5.02s which is longer than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: ERROR IP IP replied in 5.03s which is longer than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: ERROR IP IP replied in 5.04s which is longer than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: ERROR IP IP replied in 5.01s which is longer than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in 0.08s which is shorter than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in 1.37s which is shorter than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in 0.09s which is shorter than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in 1.37s which is shorter than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in 0.09s which is shorter than maximum 5.00s

It's minimised to:

http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: ERROR IP IP replied in more time than maximum 5.00s
http://url/ : http_query: OK status code 302 on IPs IP elapsed_time: OK IP IP replied in less time than maximum 5.00s

This affects:

 * bot_status
 * http_query

Obviously some information are removed from the monitor, but anyway they are
just text entries and not parsable. So other approaches shall be used to make
graphs from those values, which is not covered by pure promise status.

Arguments passed to apachedex are complex, they contain quotes, backslashes,
spaces and they are propagated through several levels of buildout profiles
and scripts.
Simplify all this by relying on argparse ability of passing arguments from
a file. Users of this tool are now expected to generate a config file and
specify the path of this config file as an arguments.

This fixes division by 0.

See merge request !86

Use shlex from standard library which splits arguments properly instead of
implementing our own buggy version.

fixup for ecf144ee

current_project variable is mandatory

"test_networkbench: increase resolution time tolerance" addresses [this](https://erp5.nexedi.net/test_result_module/20200911-8C18D791/14):

```
======================================================================
FAIL: test_dnsbench_fail (slapos.test.test_networkbench.TestDNSBench)
Test dns failure resolution
----------------------------------------------------------------------
Traceback (most recent call last):
File "/srv/slapgrid/slappart20/srv/testnode/cum/inst/test0-0/parts/slapos.toolbox/slapos/test/test_networkbench.py", line 66, in test_dnsbench_fail
self.assertLess(info[3], 1)
AssertionError: 2.0736539363861084 not less than 1
```

"cpu load plugin: fix test and minor cleanups" addresses [this](https://erp5.nexedi.net/test_result_module/20200911-C444DB47/13):

```
======================================================================
ERROR: test_check_cpu_load_run (slapos.test.promise.plugin.test_check_server_cpu_load.TestCheckServerCPULoad)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/srv/slapgrid/slappart9/srv/testnode/bvi/inst/test0-0/parts/slapos.toolbox/slapos/test/promise/plugin/test_check_server_cpu_load.py", line 48, in test_check_cpu_load_run
    self.launcher.run()
  File "/srv/slapgrid/slappart9/srv/testnode/bvi/soft/65dbbd3d11daa99cf78c9566a0a4798e/parts/slapos.core-repository/slapos/grid/promise/__init__.py", line 884, in run
    failed_promise_output))
PromiseError: Promise 'server-cpu-load-promise.py' failed with output: Error: Promise timed out after 0.5 seconds
```
Some more fixes are in nexedi/slapos!809

See merge request nexedi/slapos.toolbox!85

Sometimes it's not enough, as we can guess from the tests failures.

See merge request nexedi/slapos.toolbox!84

We don't really care how long DNS resolution took here, since we see
cases where on test machines a name resolution takes more than one second,
increase the minimal time to something higher to prevent such failures.

Also remove not used commented out code.

This test was failing randomly because it uses a timeout of 0.5 seconds, which
is sometimes not enough to run commands.

Increase the timeout to something large enough so that we don't have timeout
failures when test machine is busy.

Also simplify the whole thing by using python instead of parsing the output
of processes.