Administering Command Center Agents
The Command Center metrics collection agent,
ccagent, runs on segment hosts and receives real-time metrics emitted by the metrics collection database extension. Each segment host has one
ccagent process. The metrics collection extension connects to
ccagent using Unix Domain Sockets (UDS) to transfer metrics from Greenplum Database. Starting Greenplum Command Center with the
gpcc start command starts the Command Center agent on each segment host. Stopping Command Center with
gpcc stop ends these processes. The Command Center backend monitors these agents and restarts them when necessary.
Installing the gpperfmon database configures Greenplum Database to run agents on the master host and each segment host. The agents are started automatically when the database system starts up. The Greenplum Database postmaster process monitors the agents and restarts them when necessary.
gpmmonagent runs on the Greenplum Database master host. It collects initial query information from the master. Every 15 seconds, by default, it prompts the
gpsmonagents to send their accumulated data. The
gpmmonprocess saves the metrics data and logs to CSV text files that feed the external tables in the gpperfmon database.
gpsmonagent runs on each Greenplum Database segment host. It listens on a UDP port for metrics emitted by Greenplum Database, gathers additional metrics from the operating system, and forwards data to the
gpmmonagent when requested.
This topic describes basic agent administration tasks, including adding hosts and viewing agent log files.
The metrics collection agent runs queries on Greenplum Database on behalf of Command Center to perform activities such as retrieving information to display in the Command Center UI, saving state in the gpperfmon and postgres databases, inserting alert event records, and harvesting query history for the gpmetrics history tables. The agent runs these queries using the gpmon database role.
If you do not want all of this query activitiy to be logged in the
pg_log log file, you can set logging options for the gpmon role using the
ALTER ROLE command. For example, these commands set the gpmon role to log only DDL statements (
DROP) and to log only fatal and panic errors.
ALTER ROLE gpmon SET log_statement TO DDL; ALTER ROLE gpmon SET log_min_messages to FATAL;
When you add or replace Greenplum Database hosts, you must reinstall the Greenplum Command Center software to ensure the software is installed on the new hosts. It is not necessary to uninstall Command Center before reinstalling. Stop Command Center and restart it to start agents on the new hosts.
gpsmon agents on new hosts are detected automatically by the master agent,
gpmmon. Whenever the
gp_enable_gpperfmon server configuration parameter is enabled on the master, the
gpmmon agent automatically detects, starts, and begins harvesting data from new segment agents.
Log messages for the Command Center metrics collector agents are saved in the
logs directory of the Command Center installation directory.
Log messages for the gpperfmon master agent are written to the following file by default:
To change the gpperfmon log file location, edit the
log_location parameter in
On the segment hosts, gpperfmon agent log messages are written to a
gpsmon.log file in the segment instance’s data directory. For a host with multiple segments, the agent log file is located in the data directory of the first segment, as listed in the
gp_configuration table by dbid. If the segment agent is unable to log into this directory, it will log messages to the home directory of the user running Command Center (typically gpadmin).
At higher logging levels, the size of the gpperfmon log files may grow dramatically. To prevent the log files from growing to excessive size, you can add an optional log rollover parameter to
gpperfmon.conf. The value of this parameter is measured in bytes. For example:
max_log_size = 10485760
With this setting, the log files will grow to 10MB before the system rolls over the log file. The timestamp is added to the log file name when it is rolled over. Administrators must periodically clean out old log files that are no longer needed.