drill-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sorabh Hamirwasia (JIRA)" <j...@apache.org>
Subject [jira] [Created] (DRILL-5098) Improving fault tolerance for connection between client and foreman node.
Date Sat, 03 Dec 2016 02:05:58 GMT
Sorabh Hamirwasia created DRILL-5098:

             Summary: Improving fault tolerance for connection between client and foreman
                 Key: DRILL-5098
                 URL: https://issues.apache.org/jira/browse/DRILL-5098
             Project: Apache Drill
          Issue Type: Improvement
          Components: Client - JDBC
            Reporter: Sorabh Hamirwasia
            Assignee: Sorabh Hamirwasia
             Fix For: 1.10

With DRILL-5015 we allowed support for specifying multiple Drillbits in connection string
and randomly choosing one out of it. Over time some of the Drillbits specified in the connection
string may die and the client can fail to connect to Foreman node if random selection happens
to be of dead Drillbit.
Even if ZooKeeper is used for selecting a random Drillbit from the registered one there is
a small window when client selects one Drillbit and then that Drillbit went down. The client
will fail to connect to this Drillbit and error out. 

Instead if we try multiple Drillbits (configurable tries count through connection string)
then the probability of hitting this error window will reduce in both the cases improving
fault tolerance. During further investigation it was also found that if there is Authentication
failure then we throw that error as generic RpcException. We need to improve that as well
to capture this case explicitly since in case of Auth failure we don't want to try multiple

This message was sent by Atlassian JIRA

View raw message