Linux Sysadmin Notes

Oracle for Linux Admins

4 min read

Tap Listen. On Android keep Chrome in the foreground and volume up.

Oracle for Linux Admins

Concept

Oracle is a database plus an instance. The database is files on disk: control files, datafiles, online redo, temp, and (usually) archived redo. The instance is memory (the SGA) and background processes (ora_pmon_<SID>, LGWR, DBWn, SMON, and others) attached to those files.

From the Linux seat you own the host, the listener port, the Oracle OS user, ORACLE_HOME, disk, and the alert log. You do not need to be the DBA to tell “instance down” from “listener down” from “archive destination full”.

Why it matters

  • Apps say “database is down” when only the listener is down, or only one service name is unregistered
  • Killing -9 an Oracle process, or filling the archive disk, is how a slow host becomes a recovery
  • sqlplus / as sysdba works from the server and proves nothing about the network path the app uses
  • The alert log is the source of truth. systemctl status on a wrapper is not

Mental Model

App  --TCP 1521-->  listener (tnslsnr)  --hands off-->  server process
                                              |
                                              v
                              instance: SGA + ora_pmon_<SID>, LGWR, DBWn, ...
                                              |
                                              v
                              files: control, data, redo, archivelogs (FRA)

ORACLE_SID picks the instance. ORACLE_HOME picks the software tree. /etc/oratab maps SID:ORACLE_HOME:Y|N and is what oraenv reads.

A bequeath connection (sqlplus / as sysdba on the host, OS user in the dba group) bypasses the listener. A TNS connection does not.

Key Commands

# Which instances is this host supposed to run?
cat /etc/oratab
grep -v '^#' /etc/oratab
 
# Is the instance process up? (replace ORCL)
ps -ef | grep -E '[o]ra_pmon_'
ps -ef | grep -E '[t]nslsnr'
 
# Environment the DBA account expects
sudo -iu oracle
. /usr/local/bin/oraenv    # or . oraenv, then type the SID
echo "$ORACLE_HOME $ORACLE_SID"
 
# Local proof the instance is open. Not the app path.
sqlplus -s / as sysdba <<'SQL'
select instance_name, status from v$instance;
select open_mode from v$database;
select name, free_mb from v$asm_diskgroup;   -- only if ASM; ignore errors otherwise
SQL
 
# Listener
lsnrctl status
lsnrctl services
 
# Alert log (19c path shape; diag root can differ)
# $ORACLE_BASE/diag/rdbms/<dbname>/<SID>/trace/alert_<SID>.log
ls -lt "$ORACLE_BASE"/diag/rdbms/*/*/trace/alert_*.log
tail -n 80 "$ORACLE_BASE"/diag/rdbms/*/*/trace/alert_${ORACLE_SID}.log

Startup and shutdown belong to the DBA or a documented job. If you must, from the oracle account:

sqlplus / as sysdba
STARTUP;
SHUTDOWN IMMEDIATE;     -- clean, can wait on sessions
-- SHUTDOWN ABORT;      -- last resort; instance recovery on next start

Do not kill -9 PMON or LGWR to “restart Oracle”. That is an abort, and you will own the recovery.

Common Failure Modes & Symptoms

What you seeLikely meaningFirst checks
App cannot connect, ora_pmon existsListener down, wrong port, service not registeredlsnrctl status; `ss -lptn
ORA-12541: no listenerNothing on that host:portListener process, firewall, correct host
ORA-12514: listener does not know of serviceInstance up, service name not registeredlsnrctl services; SID vs service name
ORA-01034: ORACLE not availableInstance not startedora_pmon; alert log
ORA-00257: archiver errorArchive dest / FRA fulldf on FRA and archive mount; alert log
Host up, no ora_pmon after rebootoratab flag N, or dbstart not enabled/etc/oratab; how this host autostarts
Local sqlplus works, app does notBequeath vs listener, or firewall / SQL*NetTest with the app’s TNS entry
Disk full on / but datafiles are fineAudit, trace, or listener log on another filesystemdf -h every Oracle mount, not just data

Investigation Tips

  • Split the path: process up, listener up, service registered, port reachable from the app host, then SQL. Do them in that order.
  • Read the alert log before you restart. The last ORA- is usually the incident.
  • ORACLE_HOME and PATH differ per SID on a multi-home host. oraenv or the unit file, not your login profile.
  • Archive and FRA disks fill while data disks look fine. Check every mount in the alert log path and in v$recovery_file_dest if you can query.
  • Permissions: datafiles owned by oracle:oinstall. A restored tree as root will not open.
  • Patching the OS kernel or glibc can be fine; replacing ORACLE_HOME libraries by hand is not. Oracle has its own patch tool (OPatch / opatch).

Personal Lessons Learned

  • The first “Oracle is down” ticket was a listener. ora_pmon_ORCL was fine and lsnrctl status was not. Restarting the instance would have made a short outage into crash recovery.
  • sqlplus / as sysdba from the DB host passed, and the app still failed with ORA-12514. The service name in the app URL was not the SID. lsnrctl services was the check that mattered.
  • An archive destination on a 20 GB volume filled overnight. Datafiles had hundreds of GB free. df on the wrong mount is how you miss ORA-00257.

Graph View

Backlinks