Recently, I encountered several pitfalls and analyzed them one by one.
1. the hostname and address of the host name are inconsistent.
When another stress test environment was deployed and submitted to the test group for stress testing, an error was reported when my colleague modified the pg_0000.conf file to reload the configuration file. (The localhost. localdomain server cannot be found)
When I saw this error, I thought about it carefully. I performed operations on the deployment document, and both/etc/hosts and/etc/sysconfig/network were modified.
Ping the host name is normal, gpssh mutual trust is returned when the hadoop-test1, hadoop-test2, hadoop-test3.
It is normal to check the host name settings again.
When Psql logs on to the database and checks gp_segment_configuration, it finds that the hostname and address are inconsistent.
Baidu checked the circle and confirmed that the address name was changed and the host name was not changed because it was not restarted after/etc/hosts and/etc/sysconfig/network were modified.
At first, I tried to modify gp_segment_configuration roughly and found an error.
update gp_segment_configuration set hostname=‘hadoop-test1‘ where address=‘hadoop-test1‘;
ERROR: permission denied: "gp_segment_configuration" is a system catalog
The allow_system_table_mod parameter is found in the method of searching and modifying the system table.
set allow_system_table_mods=‘dml‘; update gp_segment_configuration set hostname=‘hadoop-test1‘ where address=‘hadoop-test1‘;
Run gpstop-u again to find that the configuration file is loaded normally.
Conclusion: After modifying the host name configuration file, try to restart the machine (or other methods) to make the host name take effect. Avoid the inconsistency between the address and host name, and cause subsequent configuration loading and other commands to fail to be executed.
Reference: 78853737
The configuration file cannot be loaded because the greenplum hostname and address are inconsistent.