gppkg remove and install are failing
Article ID: 419368
Updated On:
Products
Issue/Introduction
Removing or adding a gppkg to Greenplum cluster fails with the messages below:
[gpadmin@emea-cdw bin]$ gppkg -r MetricsCollector
20251124:10:57:40:862614 gppkg:emea-cdw:gpadmin-[INFO]:-Starting gppkg with args: -r MetricsCollector
20251124:10:57:40:862614 gppkg:emea-cdw:gpadmin-[INFO]:-Uninstalling package MetricsCollector-6.15.0_gp_6.30.0-rhel8-x86_64.gppkg
20251124:10:57:40:862614 gppkg:emea-cdw:gpadmin-[INFO]:-Validating rpm uninstallation cmdStr='rpm --test -e MetricsCollector-6.15.0-6.30.0 --dbpath /usr/local/greenplum-db-6.30.0/share/packages/database'
20251124:10:57:41:862614 gppkg:emea-cdw:gpadmin-[CRITICAL]:-gppkg failed. (Reason='') exiting...
[gpadmin@emea-cdw bin]$ gppkg -i /usr/local/greenplum-cc-6.15.0/gppkg/MetricsCollector-6.15.0_gp_6.30.0-rhel8-x86_64.gppkg
20251124:10:57:55:862648 gppkg:emea-cdw:gpadmin-[INFO]:-Starting gppkg with args: -i /usr/local/greenplum-cc-6.15.0/gppkg/MetricsCollector-6.15.0_gp_6.30.0-rhel8-x86_64.gppkg
20251124:10:57:55:862648 gppkg:emea-cdw:gpadmin-[INFO]:-Installing package MetricsCollector-6.15.0_gp_6.30.0-rhel8-x86_64.gppkg
20251124:10:57:55:862648 gppkg:emea-cdw:gpadmin-[INFO]:-Validating rpm installation cmdStr='rpm --test -i /usr/local/greenplum-db-6.30.0/.tmp/MetricsCollector-6.15.0-6.30.0.x86_64.rpm --dbpath /usr/local/greenplum-db-6.30.0/share/packages/database --prefix /usr/local/greenplum-db-6.30.0'
20251124:10:57:55:862648 gppkg:emea-cdw:gpadmin-[CRITICAL]:-gppkg failed. (Reason='MetricsCollector-6.15.0_gp_6.30.0-rhel8-x86_64.gppkg is already installed.') exiting...
Cause
There are 2 known causes for this error:
- The gp_segment_configuration table has 2 different "hostname" for the same host.
- The coordinator and standby coordinator are running on hosts that are also running segments.
Resolution
Solution 1
Check gp_segment_configuration table, the "hostname" for each host should be correct.
The "hostname" column should be the result of the command "hostname" when run on the host. For example:
# Run "hostname" on all hosts
$ gpssh -f /tmp/hostfile
=> hostname
[ cdw] emea-cdw
[sdw1] emea-sdw1
[sdw2] emea-sdw2
=>
Then check the "hostname" entries in gp_segment_configuration
gpadmin=# select dbid,content,hostname from gp_segment_configuration order by content,dbid;
dbid | content | hostname
------+---------+-----------
1 | -1 | emea-cdw
10 | -1 | sdw1
2 | 0 | emea-sdw1
6 | 0 | emea-sdw2
3 | 1 | emea-sdw1
7 | 1 | emea-sdw2
4 | 2 | emea-sdw2
8 | 2 | emea-sdw1
5 | 3 | emea-sdw2
9 | 3 | emea-sdw1
(10 rows)
In the above example dbid=10 has the incorrect hostname "sdw1". It should be "emea-sdw1"
To correct this run the following:
gpadmin=# BEGIN;
BEGIN
gpadmin=# set allow_system_table_mods=true;
SET
gpadmin=# update gp_segment_configuration set hostname='<HOSTNAME>' where dbid=<DBID>; -- replace <HOSTNAME> and <DBID> with appropriate values.
UPDATE 1
gpadmin=# -- re-run the above SELECT to check all is OK and then COMMIT the changes, or ROLLBACK if a mistake was made.
Solution 2
If the coordinator has segments running on it AND the standby coordinator host also has segments running on it and the Greenplum version is below 6.31.2, then raise a ticket with Broadcom support and refer to this article.
The code fix for this issue is released in Greenplum 6.31.2 and above.
Feedback
