Linux Sysadmin Notes

Oracle Storage and the FRA

4 min read

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

Oracle Storage and the FRA

Concept

Oracle storage on Linux is either files on a filesystem (datafiles, tempfiles, online redo, control files) or disks in ASM. The fast recovery area (FRA) is a quota-capped location for archivelogs, backups, and flashback logs. It is the volume that fills when the database is “fine” and commits are about to stop.

Linux df sees filesystem files. It does not see ASM diskgroup free space unless you query Oracle or asmcmd.

Why it matters

  • Datafile autoextend will eat a filesystem until df hits 100%, then writes fail
  • FRA full is ORA-00257 / ORA-19815, which looks like an app outage and is a space problem
  • Online redo is not a backup. Deleting it to free space corrupts the instance
  • A snapshot of a running Oracle filesystem without the database in backup mode or a crash-consistent storage snapshot is not a restore plan

Mental Model

Control files     where the database thinks its files are
Datafiles         tablespaces (SYSTEM, SYSAUX, UNDO, app, TEMP)
Online redo       small groups, circular, required to open
Archivelogs       copies of filled redo, required for PITR
FRA               budget for archivelogs + RMAN pieces + flashback

TEMP growing is sort and hash spill, not user data. UNDO growing is open transactions. Neither is fixed by deleting rows in the app without a commit.

ASM diskgroups (+DATA, +FRA) hide the Linux path. df on / will not warn you.

Key Commands

# Filesystem layout the host actually has
findmnt -A | grep -iE 'oracle|oradata|fra|u01'
df -hT /u01 /u02 /opt/oracle 2>/dev/null
# big files — do not run du on an ASM path
du -xh -d 1 /u01 2>/dev/null | sort -h
 
# From SQL, as a DBA, filesystem or ASM
sqlplus -s / as sysdba <<'SQL'
set lines 200
col name format a60
select name, round(total_mb/1024,1) total_gb, round(free_mb/1024,1) free_gb
from v$asm_diskgroup;   -- errors if this instance is not ASM; that is fine
 
select tablespace_name, round(used_percent,1) pct
from dba_tablespace_usage_metrics
order by used_percent desc;
 
select name from v$datafile;
select name from v$tempfile;
select member from v$logfile;
 
show parameter db_recovery_file_dest
select name, round(space_limit/1024/1024/1024,1) limit_gb,
       round(space_used/1024/1024/1024,1) used_gb
from v$recovery_file_dest;
SQL
 
# ASM free space without SQL
. oraenv   # SID often +ASM
asmcmd lsdg

RMAN is how archivelogs and backup pieces should leave the FRA:

rman target /
LIST ARCHIVELOG ALL;
DELETE NOPROMPT ARCHIVELOG ALL COMPLETED BEFORE 'SYSDATE-2';
-- only if that matches the site retention, and backups exist

Do not invent that retention during an incident. Read the runbook. If there is no runbook, free space by moving trace files and listener logs, and call the DBA before deleting anything under the FRA.

Common Failure Modes & Symptoms

SymptomLikely causeFirst checks
ORA-00257 / commits stuckFRA or archive dest fullv$recovery_file_dest; df on that path
ORA-01653 cannot extendTablespace or filesystem cannot growusage metrics; df of that datafile’s mount
ORA-01652 temp cannot extendSort spill, temp tablespaceTEMP free; who is sorting
df fine, Oracle says fullASM diskgroup, or FRA quota below filesystem freeasmcmd lsdg; space_limit
Snapshot restore will not openFiles inconsistent, or control file from another timeDo not “fix” with random rm; DBA recovery
Permissions after restoreFiles owned by rootchown oracle:oinstall only on the restored tree you mean

Investigation Tips

  • List every mount Oracle uses before you declare disk healthy. Control file, data, redo, FRA, and diag/ are often four filesystems.
  • FRA space_limit can be smaller than the filesystem. Oracle stops at the quota and leaves df looking comfortable.
  • Autoextend on a datafile will consume the filesystem with no further prompt. A 90% tablespace with autoextend is a disk ticket, not a SQL ticket.
  • TEMP at 100% is pressure, not corruption. Resize or kill the sort after you know the session. Do not delete the tempfile while the instance is up.
  • Backup window I/O is real. An RMAN backup to the same disks as datafiles will show up in iostat and in app latency. See O and Latency.
  • Filesystem snapshots of a running instance are not a substitute for RMAN unless storage and the DBA have agreed the snapshot is crash-consistent and tested.

Personal Lessons Learned

  • FRA quota was 50 GB on a 200 GB volume. df said 40% used. Oracle stopped archiving. The column that mattered was space_used versus space_limit, not df.
  • I almost removed an online redo member because the directory looked like logs. Redo members are a few fixed files the instance cannot open without. Archivelogs are the copies. Learn the names before you clean.
  • A storage snapshot restored the datafiles and not the matching control file. The instance would not mount. That was a recovery problem, not a filesystem permission problem.

Graph View

Backlinks