References
- - CPS-872Getting issue details... STATUS
- CPS-799 Spike: Define states and state handling for CM handle
Tasks
NCMP should be able to handle a large (10,000s!) batch of registrations at once. To make this possible we need to
Task | |
---|---|
1 | Create updated DMI Registry Yang Schema (using @yyyy-mm-dd) in changelog/db/changes/data/yang-models/dmi-registry @ 2021-12-13.yang to store Handle State |
2 | Add State, LockReason and LockReasonDetails as (Yang)Strings to the schema. Any validation or enum-limitations can be handled in the Java code. |
3 | Consider timestamp modelSyncStartTime for retry and timeout related scenarios as part of same schema update to reduce overhead of Liquibase changesets |
4 | Test/demo using CPS-Core
|
Yang Revision Proposal
- Reordered the revision tags based on reverse chronical order convention
- Added state leaf to hold the synchronization state information
- Added lock-reason leaf to hold a lock reason enum
- Added lock-reason-details leaf non enum description of lock reason. More descriptive and scenario based.
- Added model-sync-start-time to hold the start time of when the most recent sync attempt on this cmHandle was started.
- https://gerrit.onap.org/r/c/cps/+/127060/5/cps-ri/src/main/resources/changelog/db/changes/data/yang-models/dmi-registry%25402022-02-10.yang
Timestamp
modelSyncStartTime will be recorded once model sync attempt is started and will be removed once model sync is completed. CmHandle state should be changed from advised to locked during sync. Extending the query from ncmp to look for cmHandle state and modelSyncStartTime would remove need for watchdog.
Prevention of multiple NCMP attempting sync
It is possible for multiple NCMP instances to attempt to sync the same cmHandle from the singular cps instance. In order to prevent this we can implement an exclusive lock using the PESSIMISTIC_WRITE JPA locking mechanism. This will prevent more than one NCMP instance from reading, updating or deleting data related to that cmHandle.