Make z2 work again - and incidentally, make networks more robust to stalls. #1947
Conversation
rrw-zilliqa
force-pushed
the
users/richard/make-z2-work-again
branch
from
dc85271 to
95c2b72
Compare
| self.swarm | ||
| .behaviour_mut() | ||
| .kademlia | ||
| .add_address(peer, address.clone()); | ||
| self.swarm.behaviour_mut().kademlia.bootstrap()?; |
There was a problem hiding this comment.
Kademlia already does this periodically by itself:
https://github.com/libp2p/rust-libp2p/blob/master/protocols/kad/src/behaviour.rs#L990-L996
It also has logic to do this more sanely by triggering a bootstrap when new peers are added and avoid making more than one bootstrap at once.
Does z2 work without this change? Can you see if the automatic bootstrap is working? We can also adjust the automatic_bootstrap_throttle configuration.
There was a problem hiding this comment.
z2 doesn't work without this change, sadly...
The logic here is that, at least at startup, a node starts before the bootstrap; it fails to add the peer (because it's not up yet), and drops the bootstrap. Now it won't add the peer again, even if it comes up, and z2 then deadlocks essentially forever.
I'll see if automatic_bootstrap_throttle will work, but if it's currently set to less than a few hours, I suspect it is already triggering. We could perhaps omit the .kademlia_bootstrap(), but the .add_address() definitely isn't optional.
This also relates to one of the crashing bugs which I have utterly failed (sorry!) to tag as such and now can't find, where all peers get dropped and the network then dissociates.
There was a problem hiding this comment.
@JamesHinshelwood - OK. I've checked. You can't get away without re-adding the address, but (as expected) kademlia will eventually bootstrap itself automatically.
…ons, unles you are the bootstrap, in which case wait for someone to contact you.
rrw-zilliqa
force-pushed
the
users/richard/make-z2-work-again
branch
from
95c2b72 to
63b3e71
Compare
|
Bencher Report| Branch | users/richard/make-z2-work-again |
| Testbed | self-hosted |
Click to view all benchmark results
| Benchmark | Latency | Benchmark Result nanoseconds (ns) (Result Δ%) | Upper Boundary nanoseconds (ns) (Limit %) |
|---|---|---|---|
| process-empty/process-empty | 📈 view plot 🚷 view threshold | 8,275,300.00 (-10.11%) | 10,720,626.56 (77.19%) |
| produce-full/produce-full | 📈 view plot 🚷 view threshold | 1,718,700,000.00 (-26.50%) | 3,213,547,402.77 (53.48%) |
No description provided.