Installation and Administration Guide for the Cisco TelePresence Exchange System Release 1.0
Split Brain Recovery
Downloads: This chapterpdf (PDF - 117.0KB) The complete bookPDF (PDF - 2.64MB) | Feedback

Split Brain Recovery

Table Of Contents

Split Brain Recovery

Diagnosing Split Brain Mode

Recovering from Split Brain Mode

Verifying Synchronization of the Database Servers

Diagnosing Corrupted DRBD Metadata

Recovering from Corrupted DRBD Metadata


Split Brain Recovery


Revised June 30, 2011

Split brain mode refers to a state in which each database server does not know the high availability (HA) role of its redundant peer, and cannot determine which server currently has the primary HA role. In split brain mode, data modifications may have been made on either node, and those changes may not be replicated to the peer. Also, neither or both nodes may be functioning in the primary HA role.

Split brain mode occurs when there is a temporary failure of the network connections between the two database servers, for example, due to one of the following occurrences:

Restart of either database server during synchronization.

Physical disconnection of the Ethernet cables from a database server.

Loss of power to one or both database servers.


Note If the current primary database server loses power, or its integrated management module (IMM) becomes unreachable by the secondary database server (for example, due to network connectivity issues), the secondary database server cannot automatically take over as primary. If the current primary database server fails under these conditions, your system may or may not enter split brain mode. In this situation, take one of the following actions:

If the primary database server comes back up, the system may enter split brain mode; proceed to the "Diagnosing Split Brain Mode" section.

If the primary database server remains down, the split brain recovery procedure is not applicable; instead, see the "Recovering from a Failed Primary Database Server" section on page 33-1.


This chapter includes the following topics:

Diagnosing Split Brain Mode

Recovering from Split Brain Mode

Verifying Synchronization of the Database Servers

Diagnosing Corrupted DRBD Metadata

Recovering from Corrupted DRBD Metadata

Diagnosing Split Brain Mode

Use this procedure to determine whether your database servers are in split brain mode.

Before You Begin

Make sure that the database servers are correctly cabled. See the "Cabling Requirements for the Database Servers" section on page 4-3.

Procedure


Step 1 Log in to the CLI of each database server.

Step 2 On each database server, enter the utils service database status command.

Step 3 If the output indicates one or more of the following conditions, the database servers are in split brain mode:

The connection state (cs) is "StandAlone."

The role (ro) values display one of the following combinations:

"Primary/Unknown" on one server and "Secondary/Unknown" on the other server.

"Secondary/Secondary" on both servers—In this particular case, if the connection state (cs) on both servers is "Connected," then the MySQL database is corrupted, and the split brain recovery procedure will not help. Instead, see the "Corrupted MySQL Database Recovery" chapter.

"Secondary/Unknown" on both servers—In this particular case, if you know that one of the database servers had a reboot during the initial synchronization process, then your database system is functioning in a mode for which the split brain recovery procedure will not help. To recover, you need to reinstall both database servers. See the "Installing and Synchronizing the Cisco TelePresence Exchange System Database Servers" section on page 5-4.

Step 4 To recover from split brain mode, proceed to the "Recovering from Split Brain Mode" section.


Example

In the following example, the connection state (cs) of one of the database servers is StandAlone, which indicates that the nodes are in split brain mode:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : secondary
The current HA role of this node                 : primary
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Running pid 2820
MySQL status                                     : Running pid  2810
Heartbeat status                                 : Running pid 3752
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs          ro               ds                 p      mounted  fstype
0:mysql  StandAlone  Primary/Unknown  UpToDate/DUnknown  r----  ext3
--------------------------------------------------------------------------------

Related Topics

Command Reference, page C-1

Recovering from Split Brain Mode

Use this procedure to recover your database servers from split brain mode.

Before You Begin

Complete the "Diagnosing Split Brain Mode" section to confirm that your system is in split brain mode.

Decide which node has the data that you want to keep. In this procedure, you will give this node the primary HA role. All data on the other node will be lost during this procedure and will not be recoverable.

If you do not know which node has the most recent or most valuable data, follow these recommendations:

If the utils service database status command output on both nodes indicates that one node currently has the primary HA role while the other node currently has the secondary HA role, you should choose the current primary node to keep as the primary database server.

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : primary
The current HA role of this node                 : primary 
The database vip address                         : 10.22.130.54
...

If the utils service database status command output on both nodes indicates that neither or both nodes have the primary HA role, choose the node that you initially installed as the primary server to keep as the primary database server.

admin: utils service database status
--------------------------------------------------------------------------------
The initial configured HA role of this node      : primary 
The current HA role of this node                 : secondary
The database vip address                         : 10.22.130.54
...

Procedure


Step 1 Log in to the CLI of the database server which has the data that you want to keep.

Step 2 Enter the utils service database drbd keep-node command to reset the server to currently function in the primary HA role.

admin: utils service database drbd keep-node 
This command will make this node as Primary
Trying to assume primary role......... [Done]
Reconnecting to MySQL......... [Done]

Step 3 Log in to the CLI of the other database server.

Step 4 Enter the utils service database drbd discard-node command to reset the server to currently function in the secondary HA role.

admin: utils service database drbd discard-node 
This command will make this node as Secondary
Trying to assume secondary role......... [Done]
Ensuring DRBD volume unmounted...
Ensuring DRBD role is Secondary...
Discarding local MySQL data..... [Done]

Synchronization begins between the two database servers.

Step 5 Proceed to the "Verifying Synchronization of the Database Servers" section.


Related Topics

Command Reference, page C-1

Verifying Synchronization of the Database Servers

Procedure


Step 1 Log in to the CLI of each database server.

Step 2 On each database server, enter the utils service database status command.

The following examples show that synchronization is in progress and proceeding successfully, because each node is aware of the HA role of its redundant peer, and the output displays the percentage of the synchronization progress. Also, the current primary database server identifies itself as the SyncSource, while the current secondary database server identifies itself as the SyncTarget.

Sample output from the current primary database server:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : primary
The current HA role of this node                 : primary 
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Running pid 10183
MySQL status                                     : Running pid  10100
Heartbeat status                                 : Running pid 20414
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs          ro                 ds                     p  mounted     fstype
...      sync'ed:    2.0%               (44104/44980)M
0:mysql  SyncSource  Primary/Secondary  UpToDate/Inconsistent  C  /mnt/mysql  ext3
--------------------------------------------------------------------------------

Sample output from the current secondary database server:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : secondary
The current HA role of this node                 : secondary 
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Not running (only runs on primary)
MySQL status                                     : Not running (only runs on primary)
Heartbeat status                                 : Running pid 17842
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs          ro                 ds                     p  mounted  fstype
...      sync'ed:    2.1%               (44060/44980)M
0:mysql  SyncTarget  Secondary/Primary  Inconsistent/UpToDate  C
--------------------------------------------------------------------------------

Note The synchronization takes approximately 40 minutes. During this time, the disk state (ds) of the current secondary server is shown as inconsistent. An inconsistent disk state indicates that the synchronization between the database servers is not complete.


Step 3 To confirm that the synchronization is complete, enter the utils service database status command on both the primary and secondary database servers.

The following examples show that synchronization is complete, because the disk state (ds) of the current secondary server is now up to date.

Sample output from the primary database server:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : primary
The current HA role of this node                 : primary 
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Running pid 10183
MySQL status                                     : Running pid  10100
Heartbeat status                                 : Running pid 20414
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs         ro                 ds                 p  mounted     fstype
0:mysql  Connected  Primary/Secondary  UpToDate/UpToDate  C  /mnt/mysql  ext3
--------------------------------------------------------------------------------

Sample output from the secondary database server:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : secondary
The current HA role of this node                 : secondary 
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Not running (only runs on primary)
MySQL status                                     : Not running (only runs on primary)
Heartbeat status                                 : Running pid 17842
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs         ro                 ds                 p  mounted  fstype
0:mysql  Connected  Secondary/Primary  UpToDate/UpToDate  C
--------------------------------------------------------------------------------



Tip If this verification procedure shows that the split brain recovery procedure did not work for either or both servers, proceed to the "Diagnosing Corrupted DRBD Metadata" section.


Diagnosing Corrupted DRBD Metadata

If, after you complete the split brain recovery procedure, the database servers still cannot connect to each other and complete synchronization, the metadata for the Distributed Replicated Block Device (DRBD) may be corrupted. The DRBD is what synchronizes the secondary database with changes that are made on the primary database.

Before You Begin

This procedure applies only after you attempt split brain recovery. (See the "Recovering from Split Brain Mode" section.)

Procedure


Step 1 Log in to the CLI of each database server.

Step 2 On each database server, enter the utils service database status command.

Step 3 The DRBD metadata is corrupted if the disk state (ds) value is "Inconsistent/Inconsistent" while the connection state (cs) is "StandAlone" or "WFConnection" on one or both servers.

Step 4 To recover from corrupted DRBD metadata, proceed to the "Recovering from Corrupted DRBD Metadata" section.


Example

In the following example, the status of one database server indicates that the nodes have corrupted DRBD metadata:

admin: utils service database status 
--------------------------------------------------------------------------------
The initial configured HA role of this node      : secondary
The current HA role of this node                 : secondary
The database vip address                         : 10.22.130.54
The database primary node name                   : ctx-db-1
The database primary node IP address             : 10.22.130.49
The database secondary node name                 : ctx-db-2
The database secondary node IP address           : 10.22.130.57
Mon status                                       : Not running (only runs on primary)
MySQL status                                     : Not running (only runs on primary)
Heartbeat status                                 : Running pid 11459
--------------------------------------------------------------------------------
drbd driver loaded OK; device status:
version: 8.3.2 (api:88/proto:86-90)
m:res    cs            ro                 ds                         p  mounted  fstype
0:mysql  WFConnection  Secondary/Unknown  Inconsistent/Inconsistent  C
--------------------------------------------------------------------------------

Related Topics

Command Reference, page C-1

Recovering from Corrupted DRBD Metadata

Before You Begin

Make sure that the database servers are correctly cabled. See the "Cabling Requirements for the Database Servers" section on page 4-3.

Complete the "Diagnosing Corrupted DRBD Metadata" section to confirm that your system has corrupted DRBD metadata.

Procedure


Step 1 Log in to the CLI of the database server which has the data that you want to keep.

This should be the same node whose data you decided to keep when you completed the procedure in the "Recovering from Split Brain Mode" section.

Step 2 Enter the utils service database drbd force-keep-node command to reset the DRBD metadata and set the server to currently function in the primary HA role.

admin: utils service database drbd force-keep-node 
This command will make this node as Primary
Trying to assume primary role......... [Done]
Overwriting peer data... [Done]

Step 3 Log in to the CLI of the other database server.

Step 4 Enter the utils service database drbd force-discard-node command to reset the DRBD metadata and set the server to currently function in the secondary HA role.

admin: utils service database drbd force-discard-node 
Shutting down Heartbeat...
Stopping High-Availability services: 
[  OK  ]
Ensuring DRBD volume unmounted...
umount: /dev/drbd0: not mounted
Taking down DRBD Resource...
Recreating DRBD meta-data...
NOT initialized bitmap
Bringing up DRBD...
Starting Heartbeat...
Starting High-Availability services: 
[  OK  ]
 [Done]

Synchronization begins between the two database servers.

Step 5 Proceed to the "Verifying Synchronization of the Database Servers" section.


Related Topics

Command Reference, page C-1