Re-calculate limits with k8s/vertical-pod-autoscaler

[denis-tingaikin]

Our first limits were set in https://github.com/networkservicemesh/deployments-k8s/pull/727

That was good for first step but it is not looking good for now. We still see issues with limits on different systems.

Solution

Re-calcluate limits with https://github.com/kubernetes/autoscaler/tree/master/vertical-pod-autoscaler

[denis-tingaikin]

@edwarnicke Can we schedule this for v1.4.0?

[denis-tingaikin]

This looks super actual. Many users face tiny limits.

/cc @glazychev-art

[NikitaSkrynnik]

Subtasks

• [ ] Check NSMgr limits
• [ ] Check Forwarder limits
• [ ] Check Registry limits

[denis-tingaikin]

@edwarnicke

We started work with this, because many customers report about tiny limits on some components

[denis-tingaikin]

Example of the work for the vertical-pot-autoscaler

[NikitaSkrynnik]

Subtasks

• [x] Check find request perfomance in registry-k8s ~ 3h
• [x] Check NSM component limits on 10 NSC and NSE ~ 2h
• [x] Check NSM component limits on 30 NSC and NSE ~ 2h
• [x] Check NSM component limits on 60 NSC and NSE ~ 2h
• [x] Test ping time from NSC to NSE for 10, 30, 60 NSC and NSE ~ 2h
• [x] Test 10, 30, 60 NSC and NSE with disabled dataplane healing ~ 3h
• [x] Test NSCs and NSEs on one node ~ 2h

[NikitaSkrynnik]

Recommendations from VPA for 10, 20, 30, 40 NSCs and NSEs (Kernel2Ethernet2Kernel example, dataplane healing disabled)

40 NSCs and NSEs is the maximum at which all 40 pings go normally. If we deploy 50 NSCs and NSEs some pings stop working despite the fact that all requests were successful.

@edwarnicke Should we investigate the issue with 50 NSCs and NSEs now?

Here are logs from NSM with 50 NSCs and NSEs: log.zip