It probably means there is a regression somewhere and we should try to improve
the situation, as it slows down deployments

as StringIO only accepts unicode, but socket.recv() returns a string

In case of error the drop of free space percentage is shown.

The most important information is if the disk won't become full in some time,
by default 30 days. This check uses collector db history to calculate speed
of filling the disk and predicts when it will become full.

Instead of outdated approach with threshold-file just support threshold,
provided in float in GB, defaulting to 5% of disk size if found in the
database and fallback to 2.0 GB, to raise promise error.

With nexedi/slapos.core!125,
this is not the case anymore by default.

This commit does not nothing when using older slapproxy.

/reviewed-on nexedi/slapos.toolbox!62

/reviewed-on nexedi/slapos.toolbox!63

/reviewed-on !51

/reviewed-on !59

promise-timeout provides global timeout for the promise, so use it as
base for calculating default timeout.

/reviewed-on !58

Enhancements:

 * implement ignore-code
 * cover http_code
 * cover timeout
 * implement verify
 * correctly cover ca-cert-file
 * use SLAPOS_TEST_IPV4

Improvements:

 * flake8ize test_check_url_available and check_url_available
 * cover more of check_url_available

It will be easier to improve it soon.

Additionally print exact exception message.

/reviewed-on !56

support broken symlink and don't care about modified files that wouldn't be transferred anyway.

/cc @jm for python review :)
/cc @luke @Nicolas to be sure it's what we want

/reviewed-on nexedi/slapos.toolbox!55

/reviewed-on nexedi/slapos.toolbox!34

source and exclude can be in artbitrary order, so sort them in order to have
stable calls.

This simplifies assertions in tests and is no-op for the code.

It is important for DevOps to understand which files were modified during
synchronisation procedure in order to track issues fast.