Sql-server – Change Data Capture for Asynchronous Secondary Replica

There are several steps to configuring a server to accept ReadOnly traffic. The following link walks you through it, http://msdn.microsoft.com/en-us/library/hh710054.aspx ,but basically you need to configure each server in the AG and then set up the routing for each.

Here's the T-SQL involved:

ALTER AVAILABILITY GROUP [AG1]
 MODIFY REPLICA ON
N'COMPUTER01' WITH 
(SECONDARY_ROLE (ALLOW_CONNECTIONS = READ_ONLY));
ALTER AVAILABILITY GROUP [AG1]
 MODIFY REPLICA ON
N'COMPUTER01' WITH 
(SECONDARY_ROLE (READ_ONLY_ROUTING_URL = N'TCP://COMPUTER01.contoso.com:1433'));

ALTER AVAILABILITY GROUP [AG1]
 MODIFY REPLICA ON
N'COMPUTER02' WITH 
(SECONDARY_ROLE (ALLOW_CONNECTIONS = READ_ONLY));
ALTER AVAILABILITY GROUP [AG1]
 MODIFY REPLICA ON
N'COMPUTER02' WITH 
(SECONDARY_ROLE (READ_ONLY_ROUTING_URL = N'TCP://COMPUTER02.contoso.com:1433'));

ALTER AVAILABILITY GROUP [AG1] 
MODIFY REPLICA ON
N'COMPUTER01' WITH 
(PRIMARY_ROLE (READ_ONLY_ROUTING_LIST=('COMPUTER02','COMPUTER01')));

ALTER AVAILABILITY GROUP [AG1] 
MODIFY REPLICA ON
N'COMPUTER02' WITH 
(PRIMARY_ROLE (READ_ONLY_ROUTING_LIST=('COMPUTER01','COMPUTER02')));
GO

Sounds like you may be missing the configuration and/or routing information for the primary.

The more common and accepted way is that once a job lands on a server in an AG you start editing it:

• Add a first job step that checks whether this is the primary.
• Exit on success if it isn't.
• Otherwise continue onto the normal job step.

This has been explored in a few other SE questions (they don't specifically relate to CDC which is why you didn't find them):

• How to run SQL Job only on primary replica in AlwaysOn Availability Groups?

• SQL Server Agent Jobs and Availability Groups

The second way is to create a watchdog job using similar methods which will toggle the jobs on and off. This is probably more acceptable seeing as you're fiddling with SQL Server internal jobs. But it's up to you.

I wasn't quite happy with the samples above hardcoding AG names and such. I prefer to just have the job step operate within the database in question, and this way it can work out for itself if it even has a group and is part of it (as your jobs should continue running even if you temporarily remove them from an AG).

Set Nocount On

If  Exists (
    -- Database in an AG
    Select  Top 1 0
    From    sys.availability_databases_cluster adc
    Join    sys.availability_groups ag
    On      adc.group_id = ag.group_id
    Join    sys.dm_hadr_availability_group_states dhags
    On      ag.group_id = dhags.group_id
    Where   adc.database_name = Db_Name()
    )
And Not Exists (
    -- Database in an AG which is Primary on this instance
    Select  Top 1 0
    From    sys.availability_databases_cluster adc
    Join    sys.availability_groups ag
    On      adc.group_id = ag.group_id
    Join    sys.dm_hadr_availability_group_states dhags
    On      ag.group_id = dhags.group_id
    Where   adc.database_name =  Db_Name()
    And     Upper(dhags.primary_replica) = Upper(@@Servername)
    )
Begin
Raiserror('This is not the primary replica.', 16, 1) With Nowait
Return
End

-- Check if the database isn't accessible
If  Not Exists (
    Select  Top 1 0
    From    sys.databases
    Where   databases.name = Db_Name()
    And     state_desc = 'ONLINE'
    )
Begin
    Raiserror('The database doesn''t exist or isn''t ONLINE on this node.', 16, 1) With Nowait
    Return
End

This kind of thing is also useful for SSIS jobs and other bits that pop up over time.

Going one step further you probably want to create a PowerShell script which will go through and check your servers from time to time for AGs which have CDC jobs and which don't have a watchdog, so that you will be notified to create one.