12c/18c GIMR or -MGMTDB


[Update]
MGMTDB is no longer needed beginning with 19c. It's a useless, resource-wasting database anyway. But in 12c and 18c, you must at least keep mgmt DB and listener enabled in order to avoid patching errors, even if you keep them down in normal operation in order to give all the power to your user-facing database.
srvctl status mgmtdb
srvctl status mgmtlsnr
srvctl stop mgmtdb
srvctl stop mgmtlsnr
In 19c, if you accidentally installed it and wish to uninstall it, do steps 1 and 2 in
How to Move/Recreate GI Management Repository (GIMR / MGMTDB) to Different Shared Storage (Diskgroup, CFS or NFS etc) (Doc ID 1589394.1)
that is:
# <GI_HOME>/bin/crsctl stop res ora.crf -init
# <GI_HOME>/bin/crsctl modify res ora.crf -attr ENABLED=0 -init
$ <GI_HOME>/bin/dbca -silent -deleteDatabase -sourceDB -MGMTDB



* Official documents

FAQ: 12c Grid Infrastructure Management Repository (GIMR) (Doc ID 1568402.1)
	https://support.oracle.com/epmos/faces/DocumentDisplay?id=1568402.1
Oracle 12c Grid Infrastructure Management Repository (GIMR) – Everything You Wanted To Know
	https://www.youtube.com/watch?v=gSOW-zhjxSA


* Directory or file name beginning with -
To change directory into a directory named with - as the first character, or run any command passing such a name as an argument, remember to type -- before the argument, e.g.
cd -- -MGMTDB
vi -- -MGMTDB_m001_17522.trc


* Command to dump performance data

oclumon dumpnodeview -h
e.g.
oclumon dumpnodeview -last "00:10:00" -i 30


* Connect to -MGMTDB locally

export ORACLE_SID=\-MGMTDB
or
if /etc/oratab has "-MGMTDB:/u01/app/grid:N":
. oraenv
-MGMTDB
sqlplus / as sysdba
alter session set container=GIMR_DSCREP_10; --if needed


* Connect remotely (warning: unsupported)

Server side:
create user ... identified by ...;
grant ... to ...;

Client side:
tnsnames.ora:
mgmtdbtest =
  (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = dbservervip)(PORT = 1525))
    (CONNECT_DATA =
      (service_name=gimr_dscrep_10)
    )
  )


* Reduce resource usage

MGMTDB is useless and takes away 1 GB memory by default. But we can't shut it down for supportability reason. Change environment to it with oraenv, disable its use of HugePages, reduce SGA (by resetting the params), and bounce the instance:
$ . oraenv
-MGMTDB
$ sqlplus / as sysdba
SQL> alter system set use_large_pages=false scope=spfile;
SQL> alter system reset sga_max_size; <-- probably not set in spfile, it's OK
SQL> alter system reset sga_target;
SQL> alter system set db_cache_advice=off;
$ srvctl stop mgmtdb -force
$ srvctl start mgmtdb
After it's restarted, the instance may be on a different node.
Disabling HugePages in MGMTDB makes it easier to calculate HugePages allocation. The database only runs on one node. Without setting use_large_pages to false, accomodating its HugePages requirement on that single node but not the others would either complicate HugePages setup or waste memory depending on whether you configure the same HugePages on all nodes.


* Some interesting tables in MGMT

Table CHA.CHA_META_DATA:
Cluster health data is stored in tablespace SYSMGMTDATADB (also in PCMRADMIN.GIMR_CONFIG_TBL), configured retention time 72 hours, refresh_ms 5000.
Table CHM.CHMOS_ACTIVE_CONFIG_INT_TBL: Basic OS info including HugePages, system file descriptor limit. Not sure what NUMDISKS refers to.
CHM has multiple *INT_TBL tables with historical data going back to 3 days ago.
CHM.CHMOS_ASMINST_DB_INT_TBL is actually about ASM's clients, the RDBMS's including _mgmtdb itself.
CHM.CHMOS_CPU_INT_TBL: The *percent columns may need to be divided by 100? (e.g. 806 means 8.06%)
CHM.CHMOS_DEVICE_INT_TBL: historical stats about disk devices /dev/dm-* and /dev/sd* (including partitions)
CHM.CHMOS_FILESYSTEM_INT_TBL: /u01 file system usage at different times
CHM.CHMOS_NIC_INT_TBL: NIC stats
CHM.CHMOS_PROCESS_INT_TBL: process stats
CHM.CHMOS_PROC_AG_INT_TBL: instance CPU, memory, processes, etc. usage; not sure what PROC and AG mean
CHM.CHMOS_SYSTEM_PERIODIC_INT_TBL: potentially useful; to be deciphered
CHM.CHMOS_SYSTEM_SAMPLE_INT_TBL: overall system stats
Change _INT_TBL to _DATA for most tables to query views, which may be easier to read.
PCMRADMIN.GIMR_PDB_PARAMS_TBL_INT: audit retention time, configured size limits of various components




Ref:
https://www.freelists.org/post/oracle-l/Real-use-of-12c-MGMT-database,4