Multi IP handling

[MartinKolarik]

A side effect of #447 is that probes are now able to switch between multiple IP addresses based on network conditions. This may interfere with our adoption logic in two ways:

• adoption is more complex as the user must enter the correct IP and doesn't know which one that is,
• adopted probes may switch the IP and disappear from the account.

The existing UUIDs are not enough to solve this as they don't survive restarts. Several ideas for solving this:

• have the probe report all of its IPs and identify them by "at least one IP matches" comparison - requires some dash and API changes but solves the problem and might be useful in general,
• make the UUIDs persistent - very easy change but works only for software probes; hardware probes reset their storage and would lose the value after restart, unless we also add support for this in the FW (easy but doesn't work for existing probes),
• base the UUIDs on something that doesn't change - MAC addresses?

We could also reduce the amount of switching by a node setting or entirely disable it and force IPv4 when supported but not sure if that's a good idea.

[jimaek]

have the probe report all of its IPs and identify them by "at least one IP matches" comparison - requires some dash and API changes but solves the problem and might be useful in general,

This is the best option for me. Simple and nice UX

[MartinKolarik]

Simple, not really. It's definitely the hardest of those options, but also the only one that covers all cases, so I'm leaning towards it as well. We might even combine it, as adding persistence to the UUIDs is a nice and easy improvement on top.

[alexey-yarmosh]

E.g. probe supports both IPv4 and IPv6. Yesterday it had IPv4 IP and was used for measurementA. Today it has IPv6 IP and was used for measurementB.

Now we are running two new measurements based on pervious: measurementA' and measurementB'. Ideally we want both new measurements refer to the probe. But since repeated measurements are filtering by IP, IPv4 result will be offline.

To fix this, storing of both IPs of the probe is required. So looks like it need to be implemented anyway.

[MartinKolarik]

Indeed, repeating measurements is also affected by this, but it's the same problem of probe "identity", the UUIDs could be used here as well.

[alexey-yarmosh]

UUIDs refer to device, for example laptop, which I may bring to office and back home every day and change the network. I think in repeating measurement we want to compare networks, which are identified by IPs.

[MartinKolarik]

That's an interesting thought because for the user, it's going to be the same probe in that case, but I agree for measurements, it makes sense to consider them as different.

[MartinKolarik]

Let's start with the first option here.

We need to ensure that the IPs we get actually belong to the probe, so the way I imagine it working is:

1. After a probe is connected, the API sends some new type of message to the probe. The message includes some token that the API stores for 5 minutes (in redis?).
2. The probe, after it receives this message, lists all of its non-internal IPs (1) and filters out the one that's used for the API connection. For each remaining IP, it makes a request to some new API endpoint using this IP (2) and includes the token.
3. When the API receives the request, it looks up the probe based on the token and adds the IP making the request to the list of alternative IPs (3).

Notes:

1. This is what we use in another project for similar purposes:

   _(os.networkInterfaces())
    	.toPairs()
    	.map(p => p[1])
    	.flatten()
    	.uniqBy('address')
    	.filter(address => !address.internal)
    	.map('address')
    	.value()
   

2. Via https://github.com/sindresorhus/got/blob/main/documentation/2-options.md#localaddress
3. In the DB I guess the easiest way to handle this is a json field so that we can support more than 1 alternative IP. ~~We need to track the lifetime of each IP separately - 30 days since it was last reported, same as for the primary IP.~~

---

Affected features:

• probe adoption - any IP from the list can be used
• probe reusing - any IP from the list can be used
• dashboard - alternative IPs should be shown, pending design
• one probe per IP check - should be extended to check across all IPs, not just the primary ones

[MartinKolarik]

We need to track the lifetime of each IP separately - 30 days since it was last reported, same as for the primary IP.

Actually, no, the alternative addresses should be tracked primarily on the API-side and be per connection, so after each new connection they start empty. If the probe is adopted, we should just clear the alternative IPs at that point and re-add if they are reported later.

@alexey-yarmosh ping me on Slack if I missed anything or something isn't clear.