Improve reliability of chain element 'etcd'

[denis-tingaikin]

Closes #512

[zolug]

I wonder when using sdk/pkg/registry/core/chain's NewNetworkServiceEndpointRegistryClient to get a client to watch NSEs that belong to a particular NS, how should one avoid missing Delete events in case loss of nsmgr or registry handling the find? (Assuming the NSE registration/unregistration is served by a different registry and nsmgr.)

[denis-tingaikin]

I wonder when using sdk/pkg/registry/core/chain's NewNetworkServiceEndpointRegistryClient to get a client to watch NSEs that belong to a particular NS, how should one avoid missing Delete events in case loss of nsmgr or registry handling the find? (Assuming the NSE registration/unregistration is served by a different registry and nsmgr.)

True, and since it's related to the client side, I've created a separate ticket to consider the problem on the client side networkservicemesh/sdk#1671

[zolug]

Built a k8s-registry on 1.13.2 for a quick try.
Seems nses CRs including the forwarders keep getting removed from the system when reaching 60seconds lifetime.

For me its seems that register simply uses the nse returned by client.Get() for update instead of the one in the request, thus the expiration time won't be updated.

[denis-tingaikin]

@zolug Many thanks for finding this. Indeed the problem was with refreshing content of crd resources. Fix is here #527