DESCRIPTION

JVMCBC-417: Shift the carrier node refresh list on each attempt.

Motivation
----------
Previously the node lists were shuffled once for refresh polling
and always for a refresh attempt, always in "randomized" ordering.
This does not guarantee consistent polling of all the confi nodes
in the cluster.

This changeset introduces a more reliable and consistent polling
mechanism, mainly motivated by the new "fast failover"
requirements.

Modifications
-------------
This changeset adds a global volatile counter that gets used to
consistently shift the given node lists forward, both for
tainted polling and proactive refresh.

The shifting makes sure that eventually every node will get its
turn instead of randomliy shuffling or even just once for a full
refresh polling cycle.

Result
------
Better poll cycle mechanics during rebalance and proactive config
refresh, preparing for "fast failover".

Change-Id: I5aa0d0da031e15c099924588df76493d5f540a5e
Reviewed-on: http://review.couchbase.org/76934
Tested-by: Natasha Sullivan xxx@xxx.xxx
Reviewed-by: Marques Ellis xxx@xxx.xxx


COMMENTS

author: Natasha Sullivan
date: 2017-04-26 16:38:59.025000000

Change has been successfully cherry-picked as 0741bc7c4cd8efbb68bb176ca890317597b15b4a by Natasha Sullivan

-------------------------------------
