Troubleshooting gpfdist issues
If the test described in Testing your Pivotal Greenplum Configuration and Environment fails, you should carry out the following checks:
Did gpfdist start on the correct port or protocol?
To check whether gpfdist is listening on the correct port or protocol, enter the following command:
For Windows:
$ netstat -a -n | find "<<gpfdist-port>>"
For Linux:
$ netstat -a -n | grep "<<gpfdist-port>>"
The following is the output that you should get if gpfdist started on port 8080 on Windows:
$ netstat -a -n | find "8080"
TCP 0.0.0.0:8080 0.0.0.0:0 LISTENING
TCP [::]:8080 [::]:0 LISTENING
This indicates that gpfdist is listening on any network interface on both IPv4 (0.0.0.0:8080) and IPv6 ([::]:8080).
If only the IPv6 line is shown in most cases there is a local networking configuration problem. Pivotal Greenplum's recommendation in this case is to disable IPv6 on the local computer (see the Microsoft knowledge base article on how to carry this out at http://support.microsoft.com/kb/929852).
Can Pivotal Greenplum reach gpfdist?
For gpfdist to work, all Pivotal Greenplum database segments must be able to communicate with the local machine (the ETL machine as described by Pivotal Greenplum) using the HTTP protocol.
While gpfdist is running, run the following command from each one of the Pivotal Greenplum database segment nodes to ensure that all segments can access gpfdist:
$ wget http?//<<Pivotal Greenplum-host>>?<<Pivotal Greenplum-port>>/gpfdist/status
This should return a status page with the following or similar content:
read_bytes 0
total_bytes 0
total_sessions 0
When carrying out the network checks from the Pivotal Greenplum internal master node, there is usually a file called seg_host that contains a list of the Pivotal Greenplum database segment node names. If this file exists, you can check access from all segment nodes using a single Pivotal Greenplum gpssh command:
$ gpssh -f seg_host
=> wget http?//<<Pivotal Greenplum-host>>?<<Pivotal Greenplum-port>>/gpfdist/status
If this check fails, a network or system manager must change the network or system configuration so that the check succeeds.