Oracle service failover后自动切回原来的node

After database service resources in CRS failed over to another node in a RAC Cluster, they were terminated and automatically fail back to the original node incorrectly.

 

CAUSE

The "fail back" was actually a relocation of the service resource, initiated by user process(UiServer):

----------------------------------------
crsd_148.trc:2017-12-19 12:24:32.701587 :UiServer:1239340800: {1:54916:896} Container [ Name: UI_RELOCATE
crsd_148.trc:2017-12-19 12:24:32.701630 :UiServer:1239340800: {1:54916:896} Sending to PE. ctx= 0x7f830003a9f0, ClientPID=96087 ===========>
crsd_148.trc:2017-12-19 12:24:32.707319 :UiServer:1239340800: {1:54916:896} Response: c4|5!ORDERk7|MESSAGEt79|CRS-2673: Attempting to stop 'ora.orcl.orclsrv.svc' on 'node1'k7|MSGTY
PEt1|3k5|OBJIDt33|ora.orcl.orclsrv.svc 1 1k4|WAITt1|0
crsd_148.trc:2017-12-19 12:24:32.777894 :UiServer:1239340800: {1:54916:896} Response: c4|5!ORDERk7|MESSAGEt78|CRS-2677: Stop of 'ora.orcl.orclsrv.svc' on 'node1' succeededk7|MSGTYP
Et1|3k5|OBJIDt33|ora.orcl.orclsrv.svc 1 1k4|WAITt1|0
crsd_148.trc:2017-12-19 12:24:32.780142 : AGFW:1251948288: {1:54916:896} Agfw Proxy Server received the message: RESOURCE_START[ora.orcl.orclsrv.svc 1 1] ID 4098:8698047
crsd_148.trc:2017-12-19 12:24:32.780238 : AGFW:1251948288: {1:54916:896} Creating the resource: ora.orcl.orclsrv.svc 1 1
crsd_148.trc:2017-12-19 12:24:32.780339 : AGFW:1251948288: {1:54916:896} Initializing the resource ora.orcl.orclsrv.svc 1 1 for type ora.service.type
crsd_148.trc:2017-12-19 12:24:32.780380 : AGFW:1251948288: {1:54916:896} SR: acl = owner:oracle:rwx,pgrp:dba:r--,other::r--,group:dba:r-x,user:oracle:r-x
crsd_148.trc:2017-12-19 12:24:32.780665 : AGFW:1251948288: {1:54916:896} Agfw Proxy Server sending message: RESOURCE_ADD[ora.orcl.orclsrv.svc 1 1] ID 4356:2828 to the agent /u01/app/12.1.0.2/grid/bin/oraagent_oracle

The relocate was initiated by a clientPID 96087, using a "JAVA" program:

2017-12-19 12:24:32.701587 :UiServer:1239340800: {1:54916:896} Container [ Name: UI_RELOCATE
API_HDR_VER:
TextMessage[3]
CLIENT:
TextMessage[]
CLIENT_NAME:
TextMessage[java]==============>
CLIENT_PID:
TextMessage[96087] ================>
CLIENT_PRIMARY_GROUP:
TextMessage[dba]

Further analysis showed that this was caused due to the presence of a custom FAN (Fast Application Notification) script defined inside the $GRID_HOME/racg/usrco directory.  The following example of the custom script shows that it forcibly relocates a service back to the preferred node after it has failed over to an available node:

Custom script:
# if service is not running, then start it
#
echo " service stopped, starting" >> "$LOGFILE"
$ORACLE_HOME/bin/srvctl start service -d "$DATABASE" -s "$service" >> "$LOGFILE"
else
#
# Service is running, but is it running on preferred instance?
#
RUNNING=( 'echo "$SRVSTATUS" | sed -rne "s/.* ([a-zA-Z0-9]+)/\1/p" | tr "," "\n"' )
echo "${RUNNING[@]} = ${PREFERRED[@]}"
if ! in_array "$INSTANCE" "${RUNNING[@]}" ; then
echo " not running on preferred $INSTANCE" >> "$LOGFILE"
#
# Find the first non-preferred running instance
#
CURRENT=""
for inst in "${RUNNING[@]}"; do
if ! in_array "$inst" "${PREFERRED[@]}" ; then
CURRENT="$inst"
break
fi
done
#
# Relocate
#
if [[ -n "$CURRENT" ]]; then
echo " relocate $CURRENT -> $INSTANCE" >> "$LOGFILE"
$ORACLE_HOME/bin/srvctl relocate service -d "$DATABASE" -s "$service" -i "$CURRENT" -t "$INSTANCE" >> "$LOGFILE" <<<<<<<<<<<<<<<<<<<<<<<<
fi
else
#
# Service is already running on preferred instance, no need to do anything
#
echo " running on preferred $INSTANCE" >> "$LOGFILE"
fi
fi
fi
fi
done

SOLUTION

1.  If the custom script is a requirement in the environment, ignore the relocation.

2.  If the custom script is not a requirement or not wanted, verify and remove any user defined scripts in the '$GRID_HOME/racg/usrco' directory that were created for the services fail back.