4 Replies Latest reply on Nov 10, 2011 4:51 PM by an1310

Stickiness of data by site

an1310 Nov 9, 2011 2:48 PM

HI all,

5.0 introduced data affinity to group certain keys to certain nodes. My applications are modeling resources that have affinity to a site as well. There seems to be a logical extension of the Grouping API and the TopologyAwareConsistentHash to place the primary data owner on a node of a specific site.

For instance, I have a resource A that is located in a geographic region served by site 1. I would like the primary data owner to be located in site 1, with backups in site 2 and site 3.

WDYT?

1. Re: Stickiness of data by site

galder.zamarreno Nov 10, 2011 4:31 AM (in response to an1310)

Hmm, this is configurable at the transport level by assigning different sites accordingly:
https://docs.jboss.org/author/display/ISPN/ServerHinting

The logic behind it will look for nodes with different site ids in order to put backups there.
Actions
2. Re: Stickiness of data by site

an1310 Nov 10, 2011 8:01 AM (in response to galder.zamarreno)

Yes, I use the TopologyAware configuration liberally. What I want, though, is to "sticky" a set of keys with a certain site ID to a primary owner within the site, while preserving their backups on separate sites.
Actions
3. Re: Stickiness of data by site

markaddy Nov 10, 2011 10:07 AM (in response to an1310)

Hi, this may be of use:
https://docs.jboss.org/author/display/ISPN/Key+affinity+service
http://docs.jboss.org/infinispan/5.1/apidocs/org/infinispan/affinity/KeyAffinityServiceFactory.html
Actions
4. Re: Stickiness of data by site

an1310 Nov 10, 2011 4:51 PM (in response to markaddy)

Hi Mark - thanks for the reply, but this won't really help. The keys generated by the KeyAffinityService don't survive a rehash, and there are too many keys in my caches to regenerate (not to mention the mapping of generated keys to business keys).
Actions

Go to original post