[atlas] [RIPE-NCC/ripe-atlas-software-probe] Software probes only connect to one controller? (#45)

Hello,

> In this case, why not have the software probes set up with a fall-back
> RIPE Atlas Anchor (controller)?

It's worth mentioning that a probe not being able to connect to its
assigned controller is the exceptional status, not the general rule.

Probes quite happily continue to measure and store results for later
delivery even if they are disconnected from the infrastructure.
Obviously they cannot receive new measurement requests while they are
disconnected. The complexity of handling the exceptional case is a high
price to pay considering that

1) the controlling infrastructure is partially hosted at the RIPE NCC,
partially at hosting providers, with enough 9s of uptime, and

2) disconnects are basically always partial (ie. always affect a subset
of the probes)

On the flip side, there are a number of benefits of making the probes
stick to the same controller as long as possible. Therefore we defined a
time interval (2 hours) in which the probes will keep on trying to
reconnect to the same server, and only ask for help if that doesn't work
out.

I hope this explains the behaviour.

Cheers,
Robert