Heal: release resources before healing

[glazychev-art]

Description

Currently we don't release all resources when heal starts.
We call Close() before heal Request but it doesn't go through all of the Path.
https://github.com/networkservicemesh/sdk/blob/main/pkg/networkservice/common/begin/event_factory.go#L98

Use-case:

Let's take a look at the death of a nsmgr case:
nsc --> nsmgr --> fwd--> nse

Steps:

1. nsc calls request on nsmgr
2. All fine we got connection.
3. nsmgr died.
4. nsc starts healing, calls close:
   Close() : nsc --->x (died) nsmgr --fwd -- nse
Close() won't reach nsmgr and resource releasing will be interrupted.
5. Call Request() from heal:
   heal Request() : nsc --> nsmgr(new) --> (interface name collision)fwd -- nse

Actual:
got interface name collision on fwd1 side.

Expected:
all fine step 5 complete successful.

Possible solutions

1. Removing collisions by duplicating

In this case we should do:

1. Interface duplicates. If we already have interface with the same name (for example kernel on NSC, because we didn't remove it on Close() inside forwarder) we can create interface named name + some suffix
2. Use /30 mask in integration-tests and use new IPs to check new connection (after heal)
   So, we believe here that the connection after healing is a completely new connection.
   Estimation: most likely, we can do it before the release

2. Improve heal (can't be done before release)

In this case we can start thinking in directions

1. Call Close() after restart. For example if nsmgr restarted, we can try to get all connections from forwarders.
2. Think about the approach that was in the previous implementation - each element monitors the others (not just NSC)
   Estimation: we can't do it before release

[denis-tingaikin]

@glazychev-art

as I can see:

option 1 can be done before release.
option 2 can't be done before release.

Is this correct?

[glazychev-art]

@denis-tingaikin
Yes

[edwarnicke]

@glazychev-art Why doesn't the forwarder recognize the connection id and treat it as a refresh?

[denis-tingaikin]

Fixed by #1167