configure sds with raw device
Posted in 2019
A user on Informix 12.10.FC10 / AIX 7.1 with dbspaces on raw devices asked whether Shared Disk Secondary (SDS) could be set up without cluster software like PowerHA or GPFS. One reply said SDS normally relies on a shared/cluster filesystem and warned against improvising; another reported success by keeping the shared disk out of any volume group, presenting it to both AIX hosts and disabling SCSI locking, letting Informix handle sharing with only Connection Manager for failover (requiring data to be copied to new raw disks). The poster countered that IBM docs say raw LVs need enhanced concurrent varyon, which requires PowerHA, and ended by citing an IBM technote on building concurrent volume groups without a cluster, untested. No confirmed resolution is recorded.
Topics: High Availability & Replication, Storage & Space Management, Platform-Specific Issues, Clustering, Grid & MACH11
Hello all, I am looking for advice on how to configure Informix SDS. We are using Informix 12.10.FC10 Workgroup and are running on AIX 7.1. No Informix secondary server or cluster software is configured. And DBSPACE consists of raw device, not cook device (file system). I would like to know if I can configure SDS without the cluster software like PowerHA in the above environment. Thanks, SangGyu Jeong
Hello. Informix SDS depends on a cluster filesystem, managed by your OS to work. I think this article will give you some more details: https://www-01.ibm.com/support/docview.wss?uid=swg21623842 Just remember that SDS in read/write mode need according licensing. If you will use it only for failover (read only mode) and no application connections are made to that additional node, you can use it respecting maximum limits for your Informix Edition. Hope that helps Alexandre Marini
Hello Alexandre, Thank you for your reply. As you mentioned, we will only use the SDS node for failover purposes. And now the storage used by Informix is only made up of raw devices and it is difficult to migrate to the cluster file system. We also do not have a license for software such as GPFS. So I want to know how to build an SDS for an Informix instance configured to use raw devices. Slide 23 on the link below shows that you can configure HDR and SDS without a cluster manager. Is there any way to configure SDS with only the Connection Manager without the cluster manager software? https://www.slideshare.net/IBM_Info_Management/always-on-high-availability-best- practices-for-informix
Well, any OS has features for files and folders share. But, this this like: would you trust them for a database system? In case of a failure, it could damage your primary instance, right? If it would be a different product, would you have courage enough to bring up a Orcl RAC using only linux shared filesystems? I doubt it. HTH Alexandre Marini
I'm not AIX or storage expert but this is what we did for customer (with AIX and storage experts help) You must not add that disk to any VG, then you have to present that disk to both (AIX) machines - also, on both AIX machines you have to remove "SCSI locking" for that disk. And that is all. You can access disk from both AIX machines and Informix is handling "sharing" But if you currently have raw disks on VG-s, there is no easy way to convert that to proposed soulution - you will have to move this data to "new" raw disks - if you have enough disks to make copy, you can prepare disk as mentioned above and try to dd from raw disk in VG to this new raw disk. You do not need any cluster software besides informix connection manager - it will work as expected (if prim is down it can switch autmatically to sds sec) So no need for expensive HACMP, *FS, RAC, whatever, ... just use what you have :) HTH Hrvoje
Hi. thank you for your reply. I've been looking for ways to set up a shared disk without PowerHA. The documentation in the IBM Knowledge Center states that on AIX, raw devices must be set to concurrent volumes to enable access from both primary and secondary nodes. In the document below, PowerHA seems to be required to vary on volume group in enhanced concurrent mode. https://www.ibm.com/support/knowledgecenter/en/ssw_aix_71/com.ibm.aix.cmds6/vary onvg.htm Varies the volume group on Enhanced Concurrent mode. This is only possible if the volume group is Concurrent Capable or Enhanced Concurrent Capable and the system has the PowerHA® SystemMirror® product loaded and available. If neither is true, the volume group fails the varyon operation.
Hi. Alexandre. thank you for your reply. I've been looking for ways to set up a shared raw logical volumes without PowerHA. The documentation in the IBM Knowledge Center states that on AIX, raw devices must be set to concurrent volumes to enable access from both primary and secondary nodes. In the document below, PowerHA seems to be required to vary on volume group in enhanced concurrent mode. https://www.ibm.com/support/knowledgecenter/en/ssw_aix_71/com.ibm.aix.cmds6/vary onvg.htm Varies the volume group on Enhanced Concurrent mode. This is only possible if the volume group is Concurrent Capable or Enhanced Concurrent Capable and the system has the PowerHA® SystemMirror® product loaded and available. If neither is true, the volume group fails the varion operation.
@"Mr President" and SANGGYU I am pretty much interested in your final set up. One of my customer already has HA-CMP (stupid decision I agree), and he is pretty much interested in setting SDS. Thanks bunches Eric
Hello Eric. Thank you for your interest. As I mentioned in the previous article, the storage used by Informix consists of raw devices. So I had to switch the logical volumes into enhanced concurrent mode with the varyon command to make them accessible from two hosts. I do not want to use HA Cluster software when configuring SDS whenever possible, but according to the IBM Knowledge Center article, PowerHA (formerly HACMP) is required to set up concurrent logical volumes. If you are using a file system such as GPFS, HACMP may not be needed. However, I can not test it because I do not have a GPFS license. And I know GPFS is more expensive than HACMP. In addition, the steps below show how to set up a concurrent volume group without HA Cluster. I have not tried the methods below yet. https://www-01.ibm.com/support/docview.wss?uid=isg3T1024417