Configure Data Guard Broker and Fast-Start Failover for Oracle Database 26ai on Google Cloud NetApp Volumes

06/24/2026 Contributors

Configure Oracle Data Guard Broker and Fast-Start Failover with a dedicated Observer to enable automatic role transitions for Oracle Database 26ai on Google Cloud NetApp Volumes.

Tier-specific: This procedure applies only to the Prod HA (Data Guard + FSFO) tier.

This procedure covers enabling the broker on both databases, creating the Data Guard configuration, enabling FSFO with MaxAvailability protection mode, installing Oracle Instant Client on the Observer host, starting the Observer as a systemd service with wallet-based credentials, and testing switchover and failover. After ENABLE CONFIGURATION, manage transport and roles through DGMGRL (not ad-hoc LOG_ARCHIVE_DEST_* SQL).

Step 1: Enable Data Guard Broker

Enable the Data Guard Broker on both database hosts and create the broker configuration that links the primary and standby databases under unified management.

Set dg_broker_start=TRUE on the primary and standby database hosts:

sudo -u oracle bash -c '
. ~/.bash_profile
sqlplus / as sysdba <<SQL
ALTER SYSTEM SET dg_broker_start=TRUE SCOPE=BOTH;
EXIT
SQL'

On the primary, connect to DGMGRL with OS authentication and create the broker configuration:

On the Observer host only, use dgmgrl /@orcl after the auto-login wallet exists. Do not put passwords on the dgmgrl command line.

sudo -u oracle bash -c '
export ORACLE_HOME=/u01/app/oracle/product/26ai/db_1
export ORACLE_SID=orcl
export PATH=$ORACLE_HOME/bin:$PATH
dgmgrl /
'

DGMGRL> CREATE CONFIGURATION 'orcl_dg' AS
        PRIMARY DATABASE IS 'orcl' CONNECT IDENTIFIER IS orcl;
DGMGRL> ADD DATABASE 'orcls' AS CONNECT IDENTIFIER IS orcls;
DGMGRL> ENABLE CONFIGURATION;
DGMGRL> SHOW CONFIGURATION;
-- Expect: Configuration Status: SUCCESS, both members SUCCESS.

Validate the configuration — fix any WARNING or non-NULL ERROR before Step 3: Configure FSFO properties and enable:
```
DGMGRL> VALIDATE DATABASE 'orcls';
DGMGRL> SHOW CONFIGURATION VERBOSE;
```

Step 2: Confirm flashback for FSFO

Confirm that flashback database is enabled on both hosts. Flashback is required for FSFO auto-reinstate, which allows the former primary to automatically rejoin the configuration as a standby after a failover.

Confirm flashback_on is YES on both database hosts:

sudo -u oracle bash -c '
. ~/.bash_profile
sqlplus -s / as sysdba <<<"SELECT flashback_on FROM v\$database;"
'
# Expected on both hosts: YES

On the primary only, if flashback retention is not already set:

sudo -u oracle bash -c '
. ~/.bash_profile
export ORACLE_SID=orcl
sqlplus / as sysdba <<SQL
ALTER SYSTEM SET db_flashback_retention_target=1440 SCOPE=BOTH;
EXIT
SQL'

Step 3: Configure and enable FSFO

Set SYNC redo transport, configure MaxAvailability protection mode, define FSFO targets on each database, and enable Fast-Start Failover.

Set the redo transport mode to SYNC on both databases and raise the protection mode to MaxAvailability:

DGMGRL> EDIT DATABASE 'orcl'  SET PROPERTY LogXptMode='SYNC';
DGMGRL> EDIT DATABASE 'orcls' SET PROPERTY LogXptMode='SYNC';
DGMGRL> EDIT CONFIGURATION SET PROTECTION MODE AS MaxAvailability;

Set the FSFO targets so each database names the other as its failover target, then configure the threshold and auto-reinstate behavior:

-- Each side names the other
DGMGRL> EDIT DATABASE 'orcl'  SET PROPERTY FastStartFailoverTarget = 'orcls';
DGMGRL> EDIT DATABASE 'orcls' SET PROPERTY FastStartFailoverTarget = 'orcl';

-- 30 s is the default; lower for faster RTO but more sensitive to network blips
DGMGRL> EDIT CONFIGURATION SET PROPERTY FastStartFailoverThreshold = 30;
DGMGRL> EDIT CONFIGURATION SET PROPERTY FastStartFailoverAutoReinstate = TRUE;

Enable Fast-Start Failover and confirm the configuration:

DGMGRL> ENABLE FAST_START FAILOVER;
DGMGRL> SHOW FAST_START FAILOVER;
-- Expected: Threshold 30 seconds, Target orcls, Observer not yet registered.

Step 4: Install Instant Client on Observer

Install Oracle Instant Client on the dedicated Observer VM (oradg-obs), create a dedicated oracle OS user, and configure the Oracle Net environment so the Observer can connect to both database members on TCP/1521.

Install Oracle Instant Client packages on the Observer host (oradg-obs):

# Use -el8 / -el9 if the Observer is on an older OL/RHEL release
sudo dnf install -y oracle-instantclient-release-el10
sudo dnf install -y oracle-instantclient-basic \
                    oracle-instantclient-sqlplus \
                    oracle-instantclient-tools

Create a dedicated oracle OS user that will own the wallet and the systemd unit:
```
sudo useradd -u 54321 -m oracle
sudo passwd -l oracle
```

Configure the Oracle Net environment and create tnsnames.ora with entries for both database hosts:

sudo mkdir -p /etc/oracle/network/admin
sudo chown -R oracle:oracle /etc/oracle

sudo -u oracle tee /home/oracle/.bash_profile >/dev/null <<'EOF'
export ORACLE_HOME=/usr/lib/oracle/26/client64
export LD_LIBRARY_PATH=$ORACLE_HOME/lib
export PATH=$ORACLE_HOME/bin:$PATH
export TNS_ADMIN=/etc/oracle/network/admin
EOF

# tnsnames.ora — must reach both DB hosts on TCP/1521
sudo tee /etc/oracle/network/admin/tnsnames.ora >/dev/null <<'EOF'
orcl =
  (DESCRIPTION =
  (ADDRESS = (PROTOCOL = TCP)(HOST = oracdb1)(PORT = 1521))
                (CONNECT_DATA = (SERVER = DEDICATED)(SERVICE_NAME = orcl)))
orcls =
  (DESCRIPTION =
  (ADDRESS = (PROTOCOL = TCP)(HOST = oracdb2)(PORT = 1521))
                (CONNECT_DATA = (SERVER = DEDICATED)(SERVICE_NAME = orcls)))
EOF
sudo chown oracle:oracle /etc/oracle/network/admin/tnsnames.ora

Step 5: Run Observer as a systemd service

Create an auto-login wallet with credentials for both database members, then configure and start the Observer as a systemd service so it survives reboots and automatically reconnects to the configuration.

Store credentials for a dedicated Data Guard administrative account (for example, SYSDG) in the wallet rather than SYS. Credentials must never appear on a dgmgrl command line, where they are visible to ps and journalctl; always connect using /@<tns_alias> on the Observer.

Create the encrypted wallet and populate credentials for both database members:

sudo -iu oracle bash <<'BASH'
mkdir -p $TNS_ADMIN/wallet
mkstore -wrl $TNS_ADMIN/wallet -create     # prompts for a wallet password — store in your secrets manager
mkstore -wrl $TNS_ADMIN/wallet -createCredential orcl  sys ChangeMe!1
mkstore -wrl $TNS_ADMIN/wallet -createCredential orcls sys ChangeMe!1
BASH

sudo tee /etc/oracle/network/admin/sqlnet.ora >/dev/null <<'EOF'
WALLET_LOCATION = (SOURCE = (METHOD = FILE) (METHOD_DATA = (DIRECTORY = /etc/oracle/network/admin/wallet)))
SQLNET.WALLET_OVERRIDE = TRUE
EOF
sudo chown oracle:oracle /etc/oracle/network/admin/sqlnet.ora
sudo chmod -R 0700 /etc/oracle/network/admin/wallet

sudo -iu oracle ls -l /etc/oracle/network/admin/wallet
# Expected: cwallet.sso and ewallet.p12

sudo -iu oracle bash <<'BASH'
sqlplus -L "/@orcl as sysdba" <<'SQL'
SELECT database_role FROM v$database;
EXIT
SQL
BASH

sudo -iu oracle bash <<'BASH'
sqlplus -L "/@orcls as sysdba" <<'SQL'
SELECT database_role FROM v$database;
EXIT
SQL
BASH

sudo -iu oracle dgmgrl /@orcl  'SHOW CONFIGURATION;'
sudo -iu oracle dgmgrl /@orcls 'SHOW CONFIGURATION;'

Generate the auto-login wallet (cwallet.sso) so the Observer systemd service can start without a password prompt. If cwallet.sso is missing after running mkstore, use orapki from the Instant Client tools package or a database home to create it, then re-add the stored credentials:
```
sudo -iu oracle orapki wallet create \
  -wallet /etc/oracle/network/admin/wallet \
  -auto_login
sudo -iu oracle ls -l /etc/oracle/network/admin/wallet
# Expected: cwallet.sso and ewallet.p12
```

Create the systemd unit, enable the service, and verify the Observer is connected:

sudo tee /etc/systemd/system/dgmgrl-observer.service >/dev/null <<'EOF'
[Unit]
Description=Oracle Data Guard Fast-Start Failover Observer
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=oracle
Group=oracle
Environment=ORACLE_HOME=/usr/lib/oracle/26/client64
Environment=LD_LIBRARY_PATH=/usr/lib/oracle/26/client64/lib
Environment=TNS_ADMIN=/etc/oracle/network/admin
Environment=PATH=/usr/lib/oracle/26/client64/bin:/usr/bin:/bin
ExecStart=/usr/lib/oracle/26/client64/bin/dgmgrl -silent /@orcl "START OBSERVER FILE IS '/var/lib/oracle/dgmgrl-observer.dat'"
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target
EOF

sudo install -d -o oracle -g oracle -m 0755 /var/lib/oracle
sudo install -o oracle -g oracle -m 0640 /dev/null /var/log/dgmgrl-observer.log

sudo tee /etc/logrotate.d/dgmgrl-observer >/dev/null <<'EOF'
/var/log/dgmgrl-observer.log {
    weekly
    rotate 8
    compress delaycompress missingok notifempty
    create 0640 oracle oracle
    copytruncate
}
EOF

sudo systemctl daemon-reload && sudo systemctl enable --now dgmgrl-observer.service
sudo systemctl status dgmgrl-observer.service

Observer must read CONNECTED from the primary (a DISCONNECTED Observer silently suspends FSFO):

DGMGRL> SHOW FAST_START FAILOVER;
DGMGRL> SHOW CONFIGURATION;       -- Configuration Status: SUCCESS, FSFO: ENABLED

Step 6: Test FSFO

Validate the Data Guard configuration with VALIDATE DATABASE, then perform a planned switchover and, in a test window, an unplanned VM-reset failover to confirm FSFO is working end to end.

Test a planned switchover and restore the original topology:

DGMGRL> VALIDATE DATABASE 'orcls';
DGMGRL> SWITCHOVER TO 'orcls';
DGMGRL> SHOW CONFIGURATION;
DGMGRL> SWITCHOVER TO 'orcl';        -- restore topology

Test an unplanned failover using a VM reset in a controlled test window:

Use a VM Reset (crash-style test); a normal Stop may not trigger FSFO. Tail /var/log/dgmgrl-observer.log on oradg-obs to monitor failover progress; restore topology when done.

What's next?

Oracle Data Guard Broker, Fast-Start Failover, and Observer configuration is now in place for this deployment.