DESCRIPTION

JVMCBC-102: Gracefully handle non-success CCCP responses.

Motivation
----------
When CCCP is not available, the loader needs to fall back to
HTTP to eventually get a good configuration. This happens
under the following 2 cases:

 - Only a memcached bucket is opened
 - Couchbase Server 2.2 or earlier is used

If this is true and one or more nodes of the cluster are not
responding because they are down, a pathological state in the
loaders arise which stops them from bootstrapping properly
even if they could.

Modifications
-------------
The root cause of the problem is that the mergeDelayError
does not work if an error is raised and the other doesn't
respond quickly (or never at all). The one specific error
raised actually indicates that we can move on with the next
step, rather than failing completely.

This change fixes the situation by "tunneling" this specific
exception through the mergeDelayError, making sure the code
can progress through it and then converting it back into
an error.

It's not super elegant, but it works well for the issue
to solve. Don't put this into Rx tutorial slides.

Result
------
More resiliency during partial bootstrap failure under
http-only config loading scenarios.

Change-Id: I6eec3d03d21146897b9f527e959a384e18eb34ae


COMMENTS

author: Natasha Sullivan
date: 2015-11-04 17:04:05.540000000

Uploaded patch set 3.

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