The most important change is that it does not discard readable cells too
quickly anymore. A partition can now have multiple FEEDING cells, to avoid
going below the wanted level of replication.

The new algorithm is also better at minimizing the amount replication.

The MySQL implementation is written to work around the issue reported at
  https://jira.mariadb.org/browse/MDEV-12867

Since it's not worth anymore to keep track of the last connection activity
(which, btw, ignored TCP ACKs, i.e. timeouts could theorically be triggered
before all the data were actually sent), the semantics of closeClient has also
changed. Before this commit, the 1-minute timeout was reset whenever there was
activity (connection still used as server). Now, it happens exactly 100 seconds
after the connection is not used anymore as client.

The next line (MTClientConnection) already logs new connections
and the storage node is necessarily in RUNNING state.

Now, the primary master is the running master with None displayed in the
last column. Before, it could be the id timestamp of when it was secondary,
which was obsolete information.

This fixes up commit 23b6a66a,
which reimplements election.

  poll raised, retrying
  Traceback (most recent call last):
    ...
    File "neo/client/handlers/master.py", line 41, in notPrimaryMaster
      super(PrimaryNotificationsHandler, self).notPrimaryMaster(*args)
    File "neo/lib/handler.py", line 157, in notPrimaryMaster
      assert primary != self.app.server
    File "neo/client/app.py", line 109, in __getattr__
      return self.__getattribute__(attr)
  AttributeError: 'Application' object has no attribute 'server'

This fixes the following crash:

  Traceback (most recent call last):
    ...
    File "neo/master/handlers/identification.py", line 94, in requestIdentification
      uuid = app.getNewUUID(uuid, address, node_type)
    File "neo/master/app.py", line 449, in getNewUUID
      assert uuid != self.uuid
  AssertionError

This really fixes the bug described in
commit 40bac312,
which could probably be reverted. It only reduced the probability of failure.

What happened is that the second conflict on 'a' for t3 what first reported by
an answer to first store with:
- a base serial at which a=0
- a conflict serial at which a=7
However, the cached data is not 8 anymore but 12, since a second store already
occurred after the first conflict (reported by the other storage node).

When this conflict was resolved before receiving the conflict for second store,
it gave:

  resolve(old=0, saved=7, new=12) -> 19

instead of:

  resolve(old=4, saved=7, new=12) -> 15

(if we still had the data of the first store, we could also do
  resolve(old=0, saved=7, new=8)
 but that would be inefficient from a memory point of view)

The bug was difficult to reproduce. testNotifyReplicated had to be run many
many times before that race conditions trigger it. The test was changed to
enforce some of them, and the above scenario now happens almost always.

- atomic write to disk to avoid corruption
- update when the address changes (not only when a node is removed/added)

This fixes 2 issues:
- Because neoctl connects to admin nodes without requesting identification,
  the protocol version was not checked, which could even be dangerous
  (think of a user asking for information, but the packet sent by neoctl
  could be decoded as a packet to alter data, like Truncate).
- In case of mismatched protocol version, the error was not loggued on the
  node that initiated the connection.

Compatibility is handled as follows:
- For an old node receiving data from a new node, the 2 high order bytes of the
  packet id, which is always 0 for the first packet, is decoded as the packet
  code. Packet 0 has never existed, which results in PacketMalformedError.
- For a new node receiving data from an old node, the id of the first packet,
  which is always 0, is decoded as the version, which results in a version
  mismatch error.

This new protocol also guarantees that there's no conflict with SSL.

For simplification, the packet length does not count the header anymore.

When using network byte order ('!'), the size of struct items is independant of
the platform. They have never changed from one version of Python to another.

The election is not a separate process anymore.
It happens during the RECOVERING phase, and there's no use of timeouts anymore.

Each master node keeps a timestamp of when it started to play the primary role,
and the node with the smallest timestamp is elected. The election stops when
the cluster is started: as long as it is operational, the primary master can't
be deposed.

An election must happen whenever the cluster is not operational anymore, to
handle the case of a network cut between a primary master and all other nodes:
then another master node (secondary) takes over and when the initial primary
master is back, it loses against the new primary master if the cluster is
already started.

In order to do that correctly, this commit contains several other changes:

When connecting to a primary master, a full node list always follows the
identification. For storage nodes, this means that they now know all nodes
during the RECOVERING phase.

The initial full node list now always contains a node tuple for:
- the server-side node (i.e. the primary master): on a master, this is
  done by always having a node describing itself in its node manager.
- the client-side node, to make sure it gets a id timestamp:
  now an admin node also receives a node for itself.

This keeps the connection fully functional when a handler raises an exception.

Commits like 7eb7cf1b
("Minimize the amount of work during tpc_finish") dropped what was done in
commit 07b48079
("Ignore some requests, based on connection state") to protect request handlers
when they respond.

This commit fixes this in a generic way.

The initial intention was to rely on stable sorting when several events have
the same key. For this to happen, sorting must not continue the comparison with
the second item of events.

This could lead to data corruption (conflict resolution with wrong base):

  FAIL: testNotifyReplicated (neo.tests.threaded.test.Test)
  ----------------------------------------------------------------------
  Traceback (most recent call last):
    File "neo/tests/threaded/__init__.py", line 1093, in wrapper
      return wrapped(self, cluster, *args, **kw)
    File "neo/tests/threaded/test.py", line 2019, in testNotifyReplicated
      self.assertEqual([15, 11, 13, 16], [r[x].value for x in 'abcd'])
    File "neo/tests/__init__.py", line 187, in assertEqual
      return super(NeoTestBase, self).assertEqual(first, second, msg=msg)
  failureException: Lists differ: [15, 11, 13, 16] != [19, 11, 13, 16]

  First differing element 0:
  15
  19

  - [15, 11, 13, 16]
  ?   ^

  + [19, 11, 13, 16]
  ?   ^