In progress
We will use vFW use case as the baseline to test this:
Pre-requirement: Instantiate a vFW with closed loop running.
- Error detection is very fast: less than 1 second
- Recovery:
- Kill docker container, it normally takes less than 1 minute to get the system in normal state. (SDNC, APPC will take up 5 minutes)
- Delete the pod, it normally takes much longer to get back specially for SDNC, APPC (up to 15 minutes).
- Note: Helm upgrade sometimes messed up the whole system, which will turn the system into un-useable status. However, we think this may not be a normal use case for production env.
| Time (EDT) | Categories | Sub-Categories (In Error Mode Component) | Time to Detect Failure and Repair | Pass? | Notes |
|---|---|---|---|---|---|
| VNF Onboarding and Distribution | SDC | < 5 minutes | Pass | Timing?? 30 minutes. Using a script kills those components randomly, and continue onboarding VNFs. ete-k8s.sh onap healthdist After kicking off the command; waiting for 1 minutes; killed SDC; The first one was failed; then we did redistribute, it was success. | |
| SO | < 5 minutes | Pass | After kicking off the command; waiting for 1 minutes; killed SO; The first one was failed; then we did redistribute, it was success. | ||
| A&AI | < 5 minutes | Pass |
| ||
| SDNC | < 8 minutes | Pass |
Delete SDNC pod, it took very very long time to get back, it might because of the network issues. And we got a very "weird" system, SDC gives us the following error:
| ||
| < 5 minutes | Pass |
2. Run health and preload | |||
| VNF Instantiation | SDC | < 2 seconds | Pass | Tested with manually kill the docker container | |
| VID | < 1 minute | Pass |
| ||
| SO | 5 minutes | Pass | so pod restarted as part of hard rebooting 2 k8s VMs out of 9 | ||
| A&AI | 20 minutes | Pass | restarted aai-model-loader, aai-hbase, and aai-sparky-be due to hard rebooting 2 more k8s VMs probably took extra time due to many other pods restarting at the same time and taking time to converge | ||
| SDNC | 5 minutes | Pass | sdnc pods restarted as part of hard rebooting 2 k8s VMs out of 9 | ||
| MultiVIM | < 5 minutes | Pass | deleted multicloud pods and verified that new pods that come up can orchestrate VNFs as usual | ||
Closed Loop (Pre-installed manually) | DCAE | ||||
| DMaaP | |||||
Policy (Policy documentation: Policy on OOM) | 15 minutes | Pass | Deleted dev-pdp-0. No discernible interruption to closed loop. Pod restarted in 2 minutes. Deleted dev-drools-0. Closed loop failed immediately. Pod restarted in 2 minutes. Closed loop recovered in 15 minutes. Deleted dev-pap-5c7995667f-wvrgr. No discernible interruption to closed loop. Pod restarted in 2 minutes. Deleted dev-policydb-5cddbc96cf-hr4jr. No discernible interruption to closed loop. Pod restarted in 2 minutes. | ||
| A&AI | Never (observed for > 1 hour) | Fail | Deleted aai-modelloader. Closed loop failed immediately. Even though aai-modelloader container restarted within a couple of minutes (when restarted on a VM that already has the image), closed loop never recovered. | ||
| APPC (3-node cluster) | 20 minutes | Pass | Deleted dev-appc-0. Closed loop failed immediately. dev-appc-0 pod restarted in 15 minutes. Closed loop recovered in 20 minutes. |
Requirement
Area | Priority | Min. Level | Stretch Goal | Level Descriptions (Abbreviated) |
Resiliency | High | Level 2 – run-time projects | Level 3 – run-time projects | •1 – manual failure and recovery (< 30 minutes) |