Table of Contents |
---|
Related Jira(s)
Jira | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
|
Jira | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
|
Assumptions
Assumption | Notes | |
---|---|---|
1 |
Issues and Decisions
# | Issue | Notes | Decision |
---|---|---|---|
1 | How fast should CPS (and DB) be able to process max heart beat failures? | is 60K really realistic if ENM goes down we should get a notification for each node do we ?! | PoC has shown 60 seconds is reasonable |
2 | Restart of NCMP | Should/Can this be handled? | As of now, there is no such case is being considered. |
3 | Does DMI Plugin provide NCMP with a health check URL during registration? Either, just rely on the default one provided with Spring boot actuator? | Document the contract. Its just the interface that matters and not the implementation. | Spring boot actuator interface |
4 | Error during cmHandle registration | If an error occurs during registration what trustlevel should the cmHandle be set to? IN eth following scenarios
| |
5 | Module sync watchdog issues/error scenarios | If cmHandle is set to none/incomplete module sync will automatically retry (Is this acceptable?) If the module sync fails we will still send a Complete message (Is this acceptable?) Registering all cmHandles could take up to 20 mins, what should happen if the last sync fails as the notification would have been sent 20 mins ago? |
Description
- Define scenarios which cause a CM Handle to go stale.
- Implement changes to support tracking of CM Handle Freshness/Staleness.
What might trigger a cmHandle to go to STALE?
- dmi plugin identifies that the device is no longer contactable.
- dmi plugin identifies that an underlying device manager managing the device (node) is out of sync with the device itself.
Requirements
Functional
# | Interface | Requirement | Additional Information | Sign-off |
---|---|---|---|---|
1 | CPS-NCMP-E-05 | The 'trustlevel' is visible on all REST methods that currently include the 'cm handle state' | existing endpoints |
|
2 | CPS-NCMP-E-05 | CM Handles can be queried (filter condition) on 'trustlevel' | using a new 'trustLevel' condition (cannot use cpsPath condition) |
|
3 | CPS-NCMP-I-01 | During registration, DMI plugin can report initial trustlevel. If the state is not 'complete', it should be considered as 'Trustlevel change' (See req 5) | Initial trust level will be backward compatible if not set, we assume trustlevel is 'complete' For a new cm-handle where the trustlevel is 'complete' this is NOT considered a chance and no notifications should be sent |
|
4 | CPS-NCMP-E-05 | Once DMI (plugin) is detected to be down the trust-level for all affected CM Handles should be set to be 'NONE'. This wil also lead to many notifcations as per req. #5 | this might lead to a high level (20K) of notifications (need to discuss capabilities) |
|
5 | CPS-NCMP-E-05.e | NCMP notification shall be sent when the trustlevel changes | Notification be sent externally based on Kafka many small or bulk: Agreed Many notifications, one for each cm-handle |
|
6 | CPS-NCMP-I-01.e | It shall be possible to report any trustlevel of one CM Handle DMI plugin can report the current trustLevel of a single cm handle id | i.e. the DMI can tell NCMP the trustLevel is 'NONE' when a node heartbeat failure is detected and 'COMPLETE' once it is restored. Again this should lead to notifications on the external interface as per req #5 |
|
Error Handling
# | Error Scenario | Expected behavior | Sign-off |
---|---|---|---|
1 | NCMP restart (all instances) | To be discussed, not sure if it can/should be handled TrustLevels should be 'NONE' and need to be restored using an audit-request (not in scope) |
Characteristics
# | Parameter | Expectation | Notes | Sign-off |
---|---|---|---|---|
1 | dmi-down detection speed | 30 seconds | ||
2 | device heartbeat frequency (message emitted by DMI plugin for each device) | 60 seconds | ||
3 | maximum supported devices (by NCMP) | 60,0000 | Given #2 and #3 this means NCMP needs to process 60,000 message / minute! | |
4 | maximum number of cm-handles down report by DMI in one request and/or per minute | 30,000 / minute | a peak can be processed within 60 seconds | |
5 | processing of all trustLevel time for DMI-Down and/or peak load by DMI | 1 second | ||
6 | If we incorporate into searches endpoints the speed should not be impacted |
Out-of-Scope
- This epic will only introduce trustLevel NONE and COMPLETE. PARTIAL and POOR may be added later as below.
- Re-registration i.e. resolving trutsLevel degradation is not in scope of this epic
High Level Interactions
draw.io Diagram | ||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
Interface | Name | Trigger | Description | Type | Endpoint or Topic | Schema |
---|---|---|---|---|---|---|
1 | HealthCheck | 30 second interval (configurable) | NCMP is to perform a health check against each of the DMI Plugins | REST | http://<dmiPluginServiceName>/manage/health This endpoint will be the standard heath check endpoint provided by spring boot actuator. We don't store it anywhere. We just document it for now. | |
2 | CMHandle trust level change | A CMHandle managed by DMI Plugin's trust level has changed | data contains {trustLevel: ENUM} event id is cmhandle id in kafka header | Kafka | kafka topic: dmi-device-heartbeat | <cloudEvents-header> id : <cmhandleId> type : org.onap.cm.events.trustlevel-notification data : { |
3 | CMHandle Query API with trustLevel Query Condition | Client Request | CmHandle is to be returned based on the values in above CMHandle Trust Map | REST |
| { |
Managing TrustLevel
DMI Plugins
- NCMP is checking every DMI Plugin for health at interface 1 every 30 seconds using the DMI Trust Map
- IF a DMI Plugin goes down, that DMI Plugin's trust level is updated to NONE in the DMI Trust Map
- The CM handles corresponding to DMI should be set to NONE.
- IF a DMI Plugin comes back up, Trust level is set back to COMPLETE for that DMI plugin only.
More details of health check URL can be accessed via:
CPS-1857 Document watchdog job impl. with health check URL
CMHandle Heartbeat
- It is the responsibility of the DMI Plugins to update NCMP about the heartbeat of CMHandle.
- Through interface 2, DMI Plugins will provide a Kafka event on the changing of trustworthiness state of a CMHandle.
- NCMP receives this event and updates the CM Handle Trust Map accordingly
- Needs to be able to handle a throughput of 60,000 State changes per minute for 2 instances
Reading Trust Level
Body of request will be in the format as below:
Code Block language text title Search Trust Level Request Body { "cmHandleQueryParameters": [ { "conditionName": "cmHandleWithTrustLevel", "conditionParameters": [ {"trustLevel": "COMPLETE"} ] } ] }
There are two end points will be subject to query:
http://<host>:<port>/ncmp/v1/ch/id-searches
http://<host>:<port>/v1/ch/searches- Interface 3
- NCMP will first check trust level query parameters to determine which trust level (NONE, COMPLETE) is being searched.
- if the target trust level is NONE
- The cm handles stored in CM Handle Trust Map having NONE will be returned.
- if the target trust level is COMPLETE
- The cm handles stored in CM Handle Trust Map having COMPLETE will be returned.
- if the target trust level is NONE