swarm/pss: Network tests with prox handlers #1016
Conversation
| @@ -0,0 +1,426 @@ | |||
| package pss | |||
There was a problem hiding this comment.
Yes I will after all the other merges.
| // temporary function for polling a "stable" network | ||
| // stability here means no conns or drops in network within a "serenity" duration | ||
| // timeout is max time to wait for a "stable" network | ||
| func serenityNowPlease(sim *simulation.Simulation, serenity time.Duration, timeout time.Duration) bool { |
There was a problem hiding this comment.
I really dont get this :) why would you need this? just use snapshots and no discovery
There was a problem hiding this comment.
Even if we use snapshot, the connections still have to be made in the object. The health check can fire already before all these connections are made. (for example when a node only has minProxBinSize peers, its depth will be 0 and actually defined as healthy now after the kademlia depth calc changes)
If we want to reliably measure that messages end up in their correct neighborhoods, we should make sure that the network has finished forming. Otherwise the po can be checked against the wrong depth and the end result will be wrong.
There was a problem hiding this comment.
By the way, I've noticed that stability formation in the network even with mere 16 peers takes quite a bit of time. Several seconds in fact. And even then there might be seemingly spurious connects afterwards. Intuitively it doesn't seem right to me.
Of course, that belongs to problems with discovery protocol or something related, and that would be a different issue.
There was a problem hiding this comment.
now after the kademlia depth calc changes
Coming to think of it; I guess it was the same before, that networks could be healthy if there were just two known peers.
There was a problem hiding this comment.
Actually, if there was a guarantee that when after the snapshot is loaded, and before the simulation is run, that all the resulting peers from that snapshot will populate the kademlia as known, THEN perhaps waiting for the healthy state will suffice.
I would have to review that this is the case, and maybe propose in a separate PR an amendment to the snapshot loading if it is not.
There was a problem hiding this comment.
all snapshot based tests rely on this assumption and i think WaitTillHealth makes sure of that. @janos is that right?
There was a problem hiding this comment.
As we discussed in the orange-lounge, WaitTillHealth ensures the health to be true, but not that the state of snapshot is fully loaded. We are using WaitTillHealth to wait, but it may not be the correct thing to do.
nolash
force-pushed
the
pss-networktest-prox
branch
from
17760de to
b47e1cc
Compare
nolash
force-pushed
the
pss-networktest-prox
branch
from
b47e1cc to
9fc6a4b
Compare
|
solves #987 |
nolash
force-pushed
the
pss-networktest-prox
branch
from
9fc6a4b to
d310766
Compare
|
superceded by #1284 |
This PR closes #987 , adds network tests to the recently added neighborhood addressing functionality in pss.
It compares each message address to all nodes in the network and, and adds to a recipient list the nodes for whom the message address is within the nearest neighbor depth. It then checks that those recipients and only those recipients get those respective messages.
It chooses senders so that messages have to be routed as far as possible.