#!/usr/bin/env bash
# =============================================================
# rman-backup.sh
#
# Comprehensive RMAN backup script for Oracle databases.
#
# Schedule:
#   - Saturday night: Level 0 (full) incremental backup
#   - All other nights: Level 1 incremental backup
#
# Policies:
#   - Archive logs backed up 2 times before deletion
#   - 15-day recovery window for database files and archive logs
#   - Compressed backupsets with parallelism 2
#   - Controlfile and SPFILE backed up every run
#
# Maintenance:
#   - Crosscheck and remove expired/obsolete backups each run
#
# Usage:
#   chmod +x rman-backup.sh
#   # Add to crontab for 11 PM nightly:
#   # 0 23 * * * /opt/oracle/scripts/rman-backup.sh >> /var/log/oracle/rman-cron.log 2>&1
#
# Exit codes:
#   0   backup completed successfully
#   1   RMAN returned errors
#   2   environment setup error
# =============================================================

set -uo pipefail

# ----- Configuration (override via env) -----
ORACLE_SID="${ORACLE_SID:-FREE}"
ORACLE_HOME="${ORACLE_HOME:-/opt/oracle/product/23ai/dbhomeFree}"
BACKUP_BASE="${BACKUP_BASE:-/opt/oracle/backup/${ORACLE_SID}}"
LOG_DIR="${LOG_DIR:-${BACKUP_BASE}/logs}"
NLS_DATE_FORMAT="YYYY-MM-DD HH24:MI:SS"
PARALLELISM="${PARALLELISM:-2}"
RETENTION_DAYS="${RETENTION_DAYS:-15}"
ARCHIVELOG_BACKUP_COPIES="${ARCHIVELOG_BACKUP_COPIES:-2}"

export ORACLE_SID ORACLE_HOME NLS_DATE_FORMAT
export PATH="$ORACLE_HOME/bin:$PATH"

# ----- Setup -----
mkdir -p "$BACKUP_BASE"/{database,archivelog,controlfile,spfile,logs}

RUN_DATE=$(date +%Y%m%d)
RUN_TS=$(date +%Y%m%d_%H%M%S)
LOG_FILE="$LOG_DIR/rman_backup_${RUN_TS}.log"

# Determine backup level by day of week (6 = Saturday)
DOW=$(date +%u)
if [[ $DOW -eq 6 ]]; then
  LEVEL=0
  LEVEL_DESC="FULL_L0"
else
  LEVEL=1
  LEVEL_DESC="INCR_L1"
fi

echo "============================================="
echo "RMAN Backup — $ORACLE_SID"
echo "Date:  $(date '+%Y-%m-%d %H:%M:%S')"
echo "Level: $LEVEL ($LEVEL_DESC)"
echo "Log:   $LOG_FILE"
echo "============================================="

# ----- Validate environment -----
if [[ ! -x "$ORACLE_HOME/bin/rman" ]]; then
  echo "ERROR: rman not found at $ORACLE_HOME/bin/rman" >&2
  exit 2
fi

if ! pgrep -f "ora_pmon_${ORACLE_SID}" > /dev/null 2>&1; then
  echo "ERROR: database instance $ORACLE_SID is not running" >&2
  exit 2
fi

# ----- Run RMAN -----
rman target / log="$LOG_FILE" <<RMAN

# ---- Configure policies ----
CONFIGURE RETENTION POLICY TO RECOVERY WINDOW OF $RETENTION_DAYS DAYS;
CONFIGURE BACKUP OPTIMIZATION ON;
CONFIGURE CONTROLFILE AUTOBACKUP ON;
CONFIGURE CONTROLFILE AUTOBACKUP FORMAT FOR DEVICE TYPE DISK TO '${BACKUP_BASE}/controlfile/cf_autobackup_%F';
CONFIGURE ARCHIVELOG DELETION POLICY TO BACKED UP ${ARCHIVELOG_BACKUP_COPIES} TIMES TO DISK;
CONFIGURE DEVICE TYPE DISK PARALLELISM $PARALLELISM BACKUP TYPE TO COMPRESSED BACKUPSET;
CONFIGURE SNAPSHOT CONTROLFILE NAME TO '${BACKUP_BASE}/controlfile/snap_${ORACLE_SID}.f';

# ---- Database backup ----
RUN {
  ALLOCATE CHANNEL d1 DEVICE TYPE DISK;
  ALLOCATE CHANNEL d2 DEVICE TYPE DISK;

  BACKUP INCREMENTAL LEVEL $LEVEL
    TAG '${LEVEL_DESC}_${RUN_DATE}'
    FORMAT '${BACKUP_BASE}/database/${ORACLE_SID}_%T_%s_%p'
    DATABASE;

  # ---- Archive log backup + cleanup ----
  # DELETE ALL INPUT respects the configured deletion policy,
  # so archive logs are only removed from disk after being
  # backed up the required number of times (default 2).
  BACKUP
    TAG 'ARCHLOG_${RUN_DATE}'
    FORMAT '${BACKUP_BASE}/archivelog/${ORACLE_SID}_archlog_%T_%s_%p'
    ARCHIVELOG ALL DELETE ALL INPUT;

  # ---- Controlfile and SPFILE ----
  BACKUP CURRENT CONTROLFILE
    TAG 'CTL_${RUN_DATE}'
    FORMAT '${BACKUP_BASE}/controlfile/${ORACLE_SID}_ctl_%T_%s_%p';

  BACKUP SPFILE
    TAG 'SPF_${RUN_DATE}'
    FORMAT '${BACKUP_BASE}/spfile/${ORACLE_SID}_spf_%T_%s_%p';

  RELEASE CHANNEL d1;
  RELEASE CHANNEL d2;
}

# ---- Maintenance ----
CROSSCHECK BACKUP;
CROSSCHECK ARCHIVELOG ALL;
DELETE NOPROMPT OBSOLETE;
DELETE NOPROMPT EXPIRED BACKUP;
DELETE NOPROMPT EXPIRED ARCHIVELOG ALL;

# ---- Reporting ----
LIST BACKUP SUMMARY;
REPORT NEED BACKUP DAYS $RETENTION_DAYS;

EXIT;
RMAN

RMAN_RC=$?

# ----- Post-RMAN -----
echo ""
echo "============================================="
if [[ $RMAN_RC -eq 0 ]]; then
  # Check for RMAN errors in the log even if exit code was 0
  if grep -qi "RMAN-" "$LOG_FILE" 2>/dev/null; then
    echo "WARNING: RMAN messages found in log. Review $LOG_FILE"
    exit 1
  fi
  echo "RMAN completed successfully."
else
  echo "RMAN exited with code $RMAN_RC. Review $LOG_FILE"
  exit 1
fi

echo "Backup level: $LEVEL_DESC"
echo "Log file:     $LOG_FILE"
echo "Backup base:  $BACKUP_BASE"
echo "============================================="

# ----- Cleanup old log files (keep 30 days) -----
find "$LOG_DIR" -name "rman_backup_*.log" -mtime +30 -delete 2>/dev/null

exit 0