See merge request nexedi/slapos!1388

See merge request nexedi/slapos!1352

also use same version as stack/slapos.cfg for other eggs.

The official version of manuel has now the fix for which we needed a dev
version.

Pick the latest python2 compatible version.

Rotate every 100MB over enb.xlog, enb.xlog.1, ... enb.xlog.9

Add a service that runs `xamari xlog stats[samples,rf]/<period>s` in
enb/gnb instance. For now the resulting logfile enb.xlog/gnb.xlog is
placed in monitor public folder to be served by monitor http server.

Optionally the service also includes `x.drb_stats/<period>s`, which
implies 100Hz polling. This is enabled by default.

Marking that publish failed is important, but allow the operator of the system
understand what is the real problem (aka - do not swallow exception
information) is also very important.

In nexedi/slapos@0cf70a6e we already migrated the NEO URI concatenation to the
recent format used by WCFS. Unfortunately we missed separating the address
(IPv6 + port) of the master nodes from each other. Without adding the
separator ',' between them, WCFS doesn't work with NEO clusters which have
multiple master nodes, because NEO/go can't split the different master node
addresses.

The access log is useful for usual analysis, but CDN has to provide deep
metrics regarding each request, thus implement the frontend log, which follows
more or less HAProxy's httplog
http://docs.haproxy.org/2.6/configuration.html#option%20httplog and can be
correlated with a timestamp to access log.

The URI string is dropped from the frontend log, as it's already supported by
the access log. Thanks to this it's safe to use !!! (triple exclamation mark)
to split the log line, as those characters are not expected to be found in
usual metrics provided by HAProxy.

See merge request nexedi/slapos!1383

Expose and document HTTP3 related cluster switches.

Switch to Recurls with HTTP3 enabled curl for testing.

%HU in case of HTTP/2 results with full URL being logged, while just the path
is expected in the log.

Even if the change comes from http://git.haproxy.org/?p=haproxy.git;a=commit;h=30ee1efe676e8264af16bab833c621d60a72a4d7
it has been detected since full HTTP/2 aware test client has been used.

This class mimics requests, but as an engine to access sites it uses curl
command, which is expected to be provided with a lot of features, including
full HTTP3 support.

It's fetched from the upstream repository with well known working revision.

It's very important to have readable and contextable information on failures,
in order to track root cause of the problem.