Forum

General discussion on using the Kognitio Analytical Platform.
Contributor
Offline
Posts: 20
Joined: Thu May 23, 2013 5:11 pm

RAID cluster placement in Kognitio

by Simon.Darkin » Wed Mar 29, 2017 11:45 am

How would I commission a Kognitio database with software RAID, and have the RAID clusters span across nodes that exist in two discreet Cisco UCS environments?
Reply with quote Top
Contributor
Offline
Posts: 20
Joined: Thu May 23, 2013 5:11 pm

Re: RAID cluster placement in Kognitio

by Simon.Darkin » Wed Mar 29, 2017 11:50 am

This can be achieved by specifying node location information in the local configuration file along with a block attribute to guide the KAP software on where to place clusters. This will override the default behaviour of placing clusters in a way that keeps data local and so you should expect a performance penalty for configuring the clusters in this way..

Let’s imagine you have 5 database nodes in a UCS environment called UCS1, and a further 5 nodes in a UCS environment called UCS2. You could choose to configure this 10-node system with a cluster size of 2, pairing up nodes by taking a node from each UCS environment. This would provide the ability to restart the database and have it run from software RAID in the event of a UCS failure.

As root edit the local config file on each of the 10 nodes using 'wxviconf -l', and specify a file which will contain the node location information. Add lines similar to the following:

[includes]
location_info=/opt/kognitio/wx2/etc/wx2-location

On each node create the /opt/kognitio/wx2/etc/wx2-location file, this must be root-owned and non-writable like all other Kognitio configuration files. Within that file you can specify which attributes should be treated as defining a “block”, and which are specific to an individual node. In our example we a cluster to span nodes with the same nodenum value i.e. we want to pair node 1 from UCS1 with node 1 from UCS2 and so on:

kapnode1 (ucs=ucs1, nodenum=1) pairs to kapnode6 (ucs=ucs2, nodenum=1)
kapnode2 (ucs=ucs1, nodenum=2) pairs to kapnode7 (ucs=ucs2, nodenum=2)
kapnode3 (ucs=ucs1, nodenum=3) pairs to kapnode8 (ucs=ucs2, nodenum=3)
kapnode4 (ucs=ucs1, nodenum=4) pairs to kapnode9 (ucs=ucs2, nodenum=4)
kapnode5 (ucs=ucs1, nodenum=5) pairs to kapnode10 (ucs=ucs2, nodenum=5)

Here you would set block_attrs to ‘nodenum’ and proc_attrs to ucs, then assign values to nodenum and ucs on each node as shown in the table above

On kapnode1 you would create the following entries in /opt/kognitio/wx2/etc/wx2-location

[system]
block_attrs=nodenum
proc_attrs=ucs

[attributes]
nodenum=1
ucs=ucs1

Then repeat this on all the other nodes, modifying the values of nodenum and ucs to match the table above

Run a newsys

Navigate to the current startup directory, then review the clustermap file to check that the cluster placement is as you intended
Reply with quote Top

Who is online

Users browsing this forum: No registered users and 1 guest

cron