[
https://issues.apache.org/jira/browse/TRAFODION-2651?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16168640#comment-16168640
]
ASF GitHub Bot commented on TRAFODION-2651:
-------------------------------------------
GitHub user zcorrea opened a pull request:
https://github.com/apache/incubator-trafodion/pull/1233
[TRAFODION-2651] Fixed monitor-to-monitor communication by adding
reconnect logic to better handle various communication glitches which
previously
resulted in false node down situations.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/zcorrea/incubator-trafodion TRAFODION-2651
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/incubator-trafodion/pull/1233.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #1233
----
commit 5e09a1e3a41d81394a0beee7eb67f7475a344daf
Author: Zalo Correa <[email protected]>
Date: 2017-09-15T22:39:06Z
[TRAFODION-2651] Fixed monitor-to-monitor communication by adding reconnect
logic to better handle various communication glitches which previously
resulted in false node down situations.
----
> The monitor to monitor process communication cannot handle a network reset
> ---------------------------------------------------------------------------
>
> Key: TRAFODION-2651
> URL: https://issues.apache.org/jira/browse/TRAFODION-2651
> Project: Apache Trafodion
> Issue Type: Bug
> Components: foundation
> Affects Versions: 2.2-incubating
> Reporter: Gonzalo E Correa
> Assignee: Gonzalo E Correa
> Fix For: 2.3-incubating
>
>
> The monitor to monitor socket communication does not have reconnect logic to
> handle a network reset or transient network errors.
> Analysis:
> • During a ~20 second network reset window, no errors are detected by
> open sockets
> o Open sockets are dead, but there is no indication from the TCP/IP stack
> that socket is in an error condition
> • Once the network is restored, a CONNECTIONLOSS is reported by the
> Zookeeper Client Library.
> o However, reconnect logic reestablishes connection with quorum.
> • At EPOLL expiration time, EPOLL logic report “Not heard from peer=n”
> and treats peer as Node Down.
> o The node down logic deletes corresponding znode,
> CZClient::WatchNodeDelete()
> o All monitor processes continually check for expired znodes for each
> node in the cluster, including their own znode
> An expired znode is handled as a down node
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
