Thursday, July 14, 2016

Removing a Failed Node From the Cluster

This post list steps for removing a failed node from a cluster. The steps differs from steps in the previous node deletion posts (11gR1,11gR2 and 12c) such that one node has suffered a catastrophic failure and is not available for any kind of command or script executions. Therefore all the activities involved in removing the failed node are executed from a surviving node.
The environment used in this case is a two node RAC with role separation(11.2.0.4). Under normal operation it has the following resources and status. (formatted status)
Resource Name                       Type                      Target             State              Host
-------------                       ------                    -------            --------           ----------
ora.CLUSTER_DG.dg                   ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.CLUSTER_DG.dg                   ora.diskgroup.type        ONLINE             ONLINE             rhel6m2
ora.DATA.dg                         ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.DATA.dg                         ora.diskgroup.type        ONLINE             ONLINE             rhel6m2
ora.FLASH.dg                        ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.FLASH.dg                        ora.diskgroup.type        ONLINE             ONLINE             rhel6m2
ora.MYLISTENER.lsnr                 ora.listener.type         ONLINE             ONLINE             rhel6m1
ora.MYLISTENER.lsnr                 ora.listener.type         ONLINE             ONLINE             rhel6m2
ora.MYLISTENER_SCAN1.lsnr           ora.scan_listener.type    ONLINE             ONLINE             rhel6m2
ora.asm                             ora.asm.type              ONLINE             ONLINE             rhel6m1
ora.asm                             ora.asm.type              ONLINE             ONLINE             rhel6m2
ora.cvu                             ora.cvu.type              ONLINE             ONLINE             rhel6m2
ora.gsd                             ora.gsd.type              OFFLINE            OFFLINE
ora.gsd                             ora.gsd.type              OFFLINE            OFFLINE
ora.net1.network                    ora.network.type          ONLINE             ONLINE             rhel6m1
ora.net1.network                    ora.network.type          ONLINE             ONLINE             rhel6m2
ora.oc4j                            ora.oc4j.type             ONLINE             ONLINE             rhel6m2
ora.ons                             ora.ons.type              ONLINE             ONLINE             rhel6m1
ora.ons                             ora.ons.type              ONLINE             ONLINE             rhel6m2
ora.registry.acfs                   ora.registry.acfs.type    ONLINE             ONLINE             rhel6m1
ora.registry.acfs                   ora.registry.acfs.type    ONLINE             ONLINE             rhel6m2
ora.rhel6m1.vip                     ora.cluster_vip_net1.type ONLINE             ONLINE             rhel6m1
ora.rhel6m2.vip                     ora.cluster_vip_net1.type ONLINE             ONLINE             rhel6m2
ora.scan1.vip                       ora.scan_vip.type         ONLINE             ONLINE             rhel6m2
ora.std11g2.db                      ora.database.type         ONLINE             ONLINE             rhel6m1
ora.std11g2.db                      ora.database.type         ONLINE             ONLINE             rhel6m2
ora.std11g2.myservice.svc           ora.service.type          ONLINE             ONLINE             rhel6m1
ora.std11g2.myservice.svc           ora.service.type          ONLINE             ONLINE             rhel6m2
ora.std11g2.abx.domain.net.svc      ora.service.type          ONLINE             ONLINE             rhel6m2
ora.std11g2.abx.domain.net.svc      ora.service.type          ONLINE             ONLINE             rhel6m1
After the node 2 (rhel6m2 node in this case) suffers a catastrophic failure, resources and status is as below. There are offline and failed over (vip) resources from rhel6m2.
Resource Name                       Type                      Target             State              Host
-------------                       ------                    -------            --------           ----------
ora.CLUSTER_DG.dg                   ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.DATA.dg                         ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.FLASH.dg                        ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.MYLISTENER.lsnr                 ora.listener.type         ONLINE             ONLINE             rhel6m1
ora.MYLISTENER_SCAN1.lsnr           ora.scan_listener.type    ONLINE             ONLINE             rhel6m1
ora.asm                             ora.asm.type              ONLINE             ONLINE             rhel6m1
ora.cvu                             ora.cvu.type              ONLINE             ONLINE             rhel6m1
ora.gsd                             ora.gsd.type              OFFLINE            OFFLINE
ora.net1.network                    ora.network.type          ONLINE             ONLINE             rhel6m1
ora.oc4j                            ora.oc4j.type             ONLINE             ONLINE             rhel6m1
ora.ons                             ora.ons.type              ONLINE             ONLINE             rhel6m1
ora.registry.acfs                   ora.registry.acfs.type    ONLINE             ONLINE             rhel6m1
ora.rhel6m1.vip                     ora.cluster_vip_net1.type ONLINE             ONLINE             rhel6m1
ora.rhel6m2.vip                     ora.cluster_vip_net1.type ONLINE             INTERMEDIATE       rhel6m1
ora.scan1.vip                       ora.scan_vip.type         ONLINE             ONLINE             rhel6m1
ora.std11g2.db                      ora.database.type         ONLINE             ONLINE             rhel6m1
ora.std11g2.db                      ora.database.type         ONLINE             OFFLINE
ora.std11g2.myservice.svc           ora.service.type          ONLINE             ONLINE             rhel6m1
ora.std11g2.myservice.svc           ora.service.type          ONLINE             OFFLINE
ora.std11g2.abx.domain.net.svc      ora.service.type          ONLINE             OFFLINE
ora.std11g2.abx.domain.net.svc      ora.service.type          ONLINE             ONLINE             rhel6m1
Removing of resources of the failed node begins at database resource level. There are two services running and they both have the DB instance on the failed node as a preferred instance (output is condensed)
srvctl config service -d std11g2
Service name: myservice
Service is enabled
Server pool: std11g2_myservice
Cardinality: 2
...
Preferred instances: std11g21,std11g22
Available instances:
Service name: abx.domain.net
Service is enabled
Server pool: std11g2_abx.domain.net
Cardinality: 2
...
Preferred instances: std11g21,std11g22
Available instances:
Modify the service configuration so that only the surviving instances are set as preferred instances.
$ srvctl modify service -s myservice -d std11g2 -n -i std11g21 -f
$ srvctl modify service -s abx.domain.net -d std11g2 -n -i std11g21 -f

$ srvctl config service -d std11g2
Service name: myservice
Service is enabled
Server pool: std11g2_myservice
Cardinality: 1
..
Preferred instances: std11g21
Available instances:
Service name: abx.domain.net
Service is enabled
Server pool: std11g2_abx.domain.net
Cardinality: 1
..
Preferred instances: std11g21
Available instances:

$ srvctl status service -d std11g2
Service myservice is running on instance(s) std11g21
Service abx.domain.net is running on instance(s) std11g21
Remove the database instance on the failed node
srvctl config database -d std11g2
Database unique name: std11g2
Database name: std11g2
...
Database instances: std11g21,std11g22
Disk Groups: DATA,FLASH
Mount point paths:
Services: myservice,abx.domain.net
Type: RAC
Database is administrator managed
This is done using DBCA's instance management option. If the listener has a non-default name and port then accessing the DB will fail with below message.
To fix this create a default listener (name listener and port 1521). Also if VNCR is used then remove the failed node from the registration list. Proceed to instance deletion by selecting the inactive instance on the failed node.
As node 2 is not available following warning will be issued. Click continue and proceed. During the execution various other warning will appear such as unable to remove /etc/oratab etc all of these could be ignored.
However DBCA didn't run till end, at 67% (observed through repeated runs on this 11.2.0.4 environment) following dialog box appeared. As seen on the screenshot it has no message, just an OK button. Clicking it doesn't end the DBCA session but goes to the beginning and exit the DBCA clicking cancel afterwards.
However this doesn't appear to be a failure on the DBCA to remove the instance. In fact instance is removed as subsequent instance operation only list the instance on the surviving node.
Querying the database also shows that instance 2 (std11g22 in this case) related undo tablespace and redo logs have been removed and only surviving instance related undo tablespace and redo logs are available.
SQL> select name from v$tablespace;

NAME
------------------------------
SYSTEM
SYSAUX
UNDOTBS1
TEMP
USERS
EXAMPLE
TEST

7 rows selected.

SQL> select * from v$log;

    GROUP#    THREAD#  SEQUENCE#      BYTES  BLOCKSIZE    MEMBERS ARC STATUS           FIRST_CHANGE# FIRST_TIM NEXT_CHANGE# NEXT_TIME
---------- ---------- ---------- ---------- ---------- ---------- --- ---------------- ------------- --------- ------------ ---------
         1          1       1598   52428800        512          2 NO  CURRENT               68471125 07-JUL-16   2.8147E+14
         2          1       1597   52428800        512          2 YES INACTIVE              68467762 07-JUL-16     68471125 07-JUL-16

srvctl config database -d std11g2
Database unique name: std11g2
Database name: std11g2
...
Database instances: std11g21
Disk Groups: DATA,FLASH
Mount point paths:
Services: myservice,abx.domain.net
Type: RAC
Database is administrator managed
Once the database resources are removed next step is to remove the Oracle database home entry for the failed node from the inventory.



As the node is unavailable, there's no un-installation involved. Run the inventory update command with surviving nodes. Inventory content for the Oracle home before the failed node is removed.
<HOME NAME="OraDb11g_home2" LOC="/opt/app/oracle/product/11.2.0/dbhome_4" TYPE="O" IDX="4">
   <NODE_LIST>
      <NODE NAME="rhel6m1"/>
      <NODE NAME="rhel6m2"/>
   </NODE_LIST>
</HOME>
After the inventory update
 ./runInstaller -updateNodeList ORACLE_HOME=$ORACLE_HOME "CLUSTER_NODES={rhel6m1}"
Starting Oracle Universal Installer...

Checking swap space: must be greater than 500 MB.   Actual 4095 MB    Passed
The inventory pointer is located at /etc/oraInst.loc
The inventory is located at /opt/app/oraInventory
'UpdateNodeList' was successful.

<HOME NAME="OraDb11g_home2" LOC="/opt/app/oracle/product/11.2.0/dbhome_4" TYPE="O" IDX="4">
   <NODE_LIST>
      <NODE NAME="rhel6m1"/>
   </NODE_LIST>
</HOME>
Next step is to remove the cluster resources and the node itself. If any of the node is in pin stat, unpin them. In this case both nodes are unpinned
olsnodes -s -t
rhel6m1 Active  Unpinned
rhel6m2 Inactive        Unpinned
Stop and remove the VIP resource of the failed node
# srvctl stop  vip -i rhel6m2-vip -f
# srvctl remove vip -i rhel6m2-vip -f
Remove the failed node from the cluster configuration
#  crsctl delete node -n rhel6m2
CRS-4661: Node rhel6m2 successfully deleted.
Finally remove the grid home for the failed node from the inventory. Before inventory update
<HOME NAME="Ora11g_gridinfrahome2" LOC="/opt/app/11.2.0/grid4" TYPE="O" IDX="3" CRS="true">
   <NODE_LIST>
      <NODE NAME="rhel6m1"/>
      <NODE NAME="rhel6m2"/>
   </NODE_LIST>
</HOME>
After inventory update
./runInstaller -updateNodeList ORACLE_HOME=$ORACLE_HOME "CLUSTER_NODES={rhel6m1}" CRS=TRUE
Starting Oracle Universal Installer...

Checking swap space: must be greater than 500 MB.   Actual 4095 MB    Passed
The inventory pointer is located at /etc/oraInst.loc
The inventory is located at /opt/app/oraInventory
'UpdateNodeList' was successful.

<HOME NAME="Ora11g_gridinfrahome2" LOC="/opt/app/11.2.0/grid4" TYPE="O" IDX="3" CRS="true">
   <NODE_LIST>
      <NODE NAME="rhel6m1"/>
   </NODE_LIST>
</HOME>
Validate the node removal with cluvfy
cluvfy stage -post  nodedel -n rhel6m2

Performing post-checks for node removal
Checking CRS integrity...
Clusterware version consistency passed
CRS integrity check passed
Node removal check passed
Post-check for node removal was successful.
Remove the default listener if one was created during instance remove step. The final status of resource is as below.
Resource Name                       Type                      Target             State              Host
-------------                       ------                    -------            --------           ----------
ora.CLUSTER_DG.dg                   ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.DATA.dg                         ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.FLASH.dg                        ora.diskgroup.type        ONLINE             ONLINE             rhel6m1
ora.MYLISTENER.lsnr                 ora.listener.type         ONLINE             ONLINE             rhel6m1
ora.MYLISTENER_SCAN1.lsnr           ora.scan_listener.type    ONLINE             ONLINE             rhel6m1
ora.asm                             ora.asm.type              ONLINE             ONLINE             rhel6m1
ora.cvu                             ora.cvu.type              ONLINE             ONLINE             rhel6m1
ora.gsd                             ora.gsd.type              OFFLINE            OFFLINE
ora.net1.network                    ora.network.type          ONLINE             ONLINE             rhel6m1
ora.oc4j                            ora.oc4j.type             ONLINE             ONLINE             rhel6m1
ora.ons                             ora.ons.type              ONLINE             ONLINE             rhel6m1
ora.registry.acfs                   ora.registry.acfs.type    ONLINE             ONLINE             rhel6m1
ora.rhel6m1.vip                     ora.cluster_vip_net1.type ONLINE             ONLINE             rhel6m1
ora.scan1.vip                       ora.scan_vip.type         ONLINE             ONLINE             rhel6m1
ora.std11g2.db                      ora.database.type         ONLINE             ONLINE             rhel6m1
ora.std11g2.myservice.svc           ora.service.type          ONLINE             ONLINE             rhel6m1
ora.std11g2.abx.domain.net.svc      ora.service.type          ONLINE             ONLINE             rhel6m1
Useful metalink notes
How to remove/delete a node from Grid Infrastructure Clusterware when the node has failed [ID 1262925.1]
Steps to Remove Node from Cluster When the Node Crashes Due to OS/Hardware Failure and cannot boot up [ID 466975.1]
RAC on Windows: How to Remove a Node from a Cluster When the Node Crashes Due to OS/Hardware Failure and Cannot Boot [ID 832054.1]

Related Post
Deleting a Node From 12cR1 RAC
Deleting a Node From 11gR2 RAC
Deleting a 11gR1 RAC Node