In the StatusController the statistics for the
members on python3 is a byte string which cannot
be jsonified.

This changes so it loops through the statistics
for the member and ensures the key is a string
and not a byte string.

This reverts commit 1daba130be03f8bcc2654677ad9b69db70bb74ce.

Update the docs etc. to use rados directly.

Under newer MySQL versions check constraints are supported and
must have unique names.

Drop and re-create the check constraints on the resource and
resource_history tables to ensure that they are uniquely named
for upgrades in existing deployments.

Ensure that check constraints are created with unique names in
new deployments.

Debug causes Travis to drop the job because
there is to much output.

This reverts commit df3ed0a4a2772ca6224cb4a9c3093a7f1d728c6f.

There is a issue when executing pifpaf directly
from tox, something with psutil/subprocess interaction
that makes it not find the run-upgrade-tests.sh script.

These are treated as error, suppress for now
and we can go back and fix docs later.

This is the current hosting platform now that the name has expired.

- force upgrade of six
- remove README.rst cleanup as it's been done permanently (and i didn't
realise this line existed when i did it)
- add skip_existing[1]. each ENV runs the deploy so travis tries to
upload multiple times and fails on each ENV except for the first

[1] https://docs.travis-ci.com/user/deployment/pypi/#upload-artifacts-only-once

- don't concat the values when getting unique times across series
- create a common sum function instead of making a sum timeseries and
ripping out just the values.
- fix test

Fix miscellanous encoding issue when running under Python 3:

 - Encoding of member_id prior to passing into tooz.
 - Decoding of member ID's during response processing
   for status API calls.

it's failing.

this was previously used to log the aggregation performance (points/time).
this logging doesn't exist anymore since we moved to bulk saving so
remove it.

seems long_description needs to render cleanly to upload to pypi:
https://github.com/pypa/warehouse/issues/5890

related: #1025

infinity and nan are not part of json standard. the default python
json serialiser will dump `inf` and `nan` (unquoted) and load the
them.

we've historically returned `nan` and supported ignoring it by specifying
`fill=dropna`.

i don't believe `"inf"` or `"nan"` is significantly better and since
i don't want to break existing behaviour of nan, i think it's best
to just treat infinity as we do nan for now. anything more opinionated
probably requires more discussion.

related: #1023

Reduce the number of required reviewers

fix gendoc

With the low activity going on, it might be hard to get two reviews.

gendoc throws: `AttributeError: 'Sphinx' object has no attribute 'info'`

use logging provided by sphinx.util (not sure why it's using sphinx logger
but meh)

don't fail everything if one group missing metric

Old aggregate API was allowing rate:XXX format for rateofchange
aggregation.

This change restores it since we use the new internal API.

Closes: #1016

it's possible that not all groups will contain a specific metric.
rather than fail, return the groups that do have the matching metric.
only if all groups don't have matching metrics, should we fail.

Fixes: #1013

catch and carry on.

Closes: #996

the REST output seems to be clashing with the formatting of previous
element. remove the formatting and move the REST output above.

The error detail looks like:

  {
    'cause': 'Invalid input',
    'reason': 'boulet',
    'detail': ['field'],
  }

Note:

Voluptuous prints `r` before data path for
str(voluptuous.Invalid) in Python2. To simplify
tests in Python2 and Python3, the error reasons
are checked by regex.

This extension does not work anymore and is not maintained upstream.

Fixes #953