Page History

...

• DMI record could reach in millions of instances
• In memory solution (ie hazelcast) can handle 10s - 100s of Megabytes (add some sample calc)
  • still need persistence if we don't have enough instances
• use many small messages reduce need for amalgamation
• amalgamated string-based solution.... 

Solution Proposal

We have the below configuration managed by CPS-NCMP

1. We get a Subscription Create Request with unique subscription-id and a list of predicates(list of targets , datastore and list of filters)
   
   

   Code Block
   title Subscription Create Request
   { "subscriptionId" : "A-10", "predicates" : [ { "targets": [ch-1,ch-2], "datastore" : "ncmp-datastore:passthrough-operational", "datastore-xpath-filter" : ["p1/c1","p2/c2"] }, { "targets": [ch-3], "datastore" : "ncmp-datastore:passthrough-operational", "datastore-xpath-filter" : ["p3/c3"] } ] }

   
   
   

2. We persist the incoming request as is in our persistent store using a relevant subscription model.

3. We also maintain a distributed datastructure in Hazelcast to keep track of the request and response to the client.
   
   

   SubscriptionId	dmi-plugin	affectedCmHandles	status
   A-10	dmi-1	["ch-1","ch-2"]	PENDING
   A-10	dmi-2	["ch-3"]	PENDING

   
   

4. Read the requests from the distributed data structure and form the request for the DMI-Plugin and publish it to the internal kafka channnel.
5. If the DMI-Plugin responds back within the configured time , so the request will either be ACCEPTED or REJECTED.
   1. if whole subscription request is ACCEPTED , update the distributed data structure and then store the subscription to the ActiveSubscriptions in cps-cache.
   2. if whole subscription request is REJECTED , update the distributed data structure and send back the response to the client.

      
      

      SubscriptionId	dmi-plugin	affectedCmHandles	status
      A-10	dmi-1	["ch-1","ch-2"]	ACCEPTED
      A-10	dmi-2	["ch-3"]	REJECTED

      
      

6. We store the "ActiveSubscriptions" using a new model which will look like below.
   
   

   SubscriptionId	cmHandle	filter
   A-10	ch-1	p1/c1
   A-10	ch-1	p2/c2
   A-10	ch-2	p1/c1
   A-10	ch-2	p2/c2

   
   
   

7. Now based on whatever we have in the distributed map , we will send the response to the clients and may be clear the in-memory structure if required.

Proposed JIRAs

...