DESCRIPTION

JCBC-287: Make sure node is in bucket node list.

When the SDK is attached against a 1.8 cluster, it can be the
case that when the streaming node is failovered by hand (and not
because of a power plug), the REST API is still available. It will
show the bucket information and accept streaming connections, but
does not participate actively.

In effect, this leads to delayed config discovery and timeouts. To
remedy this situation, the ConfigurationProvider now not only checks
if a configuration for the bucket is available, but also if the
current node is present in the list. If it is not, the next entry
will be tried (so it will be skipped).

Change-Id: Iabb3214d5f823898091fba35ff0db2ee49ba2dc0


COMMENTS

author: Natasha Sullivan
date: 2013-04-18 09:04:54.401000000

Uploaded patch set 2.

-------------------------------------
author: Andreas Short
date: 2013-04-19 00:55:16.913000000

Patch Set 2: (1 inline comment)

This seems like a good approach, but it's a bit of an odd case.  

This relies on the node being not in the cluster list, but still aware of the cluster list.  If a node is failover'd, it should revert back to an unconfigured state I think, so we'd be better off checking for that.  That's a known problem for libcouchbase actually.

I think this is a good workaround for now and an improvement, but we may want to consider this a bit more fully.

Line:275, src/main/java/com/couchbase/client/vbucket/ConfigurationProviderHTTP.java -> I'm really not a fan of exclamation points in such log messages.  It espcially doesn't make sense if it's a warning.  Do you think we need to change this?

-------------------------------------
author: Andreas Short
date: 2013-04-19 00:59:31.832000000

Patch Set 2:

Actually, now that I think about this a bit more, when the problem presents itself is when the node has been re-added and it's coming back *in* to the cluster.

If the configuration received is not correct and it's a member of the cluster, that's a bigger problem and maybe this check is good for "sanity", but it'd be a problem if a node is a member of the cluster and sending out bad configurations.

-------------------------------------
author: Andreas Short
date: 2013-05-08 15:27:49.160000000

Patch Set 2: I would prefer that you didn't submit this

Just recording my objection to this.  We need to have a complete understanding of the cause before we add this change.

-------------------------------------
author: Natasha Sullivan
date: 2013-12-17 11:49:11.957000000

Abandoned

not needed anymore with cccp and other changes

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