This message was deleted.
# general
s
This message was deleted.
f
Hmm yeah, that is correct… At LI, D2 has a separate ZK dedicated to it, so even though the Venice client packs a D2 client, it doesn’t need to touch Venice’s ZK. But in OSS, I suppose that by default we would be making the D2 announcement on the same ZK… And there is a third ZK which is for Kafka (or Pulsar/Bookkeeper in your case). That one, again, is run as a separate one within LI, but can also be run in a different path of the same ZK.
e
How does the client uses D2? For coordination of multiple replicas of the client?
f
D2 is for service discovery. Finding physical hostnames from a logical service name.
e
I will try to remove any zookeeper references from the writer code
I have been able to remove any reference to ZooKeeper
and it seems to work properly
I need only the URL of a router, both on the reader and on the writer
f
Curious to see what references you mean! Do you have it in a branch somewhere?
does it make sense to you ?
f
Hmm… ok I see, so it is because of the D2 stuff. I think it can be good to have a mode that is D2-free (and thus fully ZK-free), but if that is a goal, then we should structure the code in a way where it is still an option to use D2.
More generally, I think it may be interesting to consider if we can make service discovery pluggable, so that we still have service disco, but with other options than D2.
e
sure thing
Now I need only to put up a POC that works 😁 when I see that I have all the pieces then it will be better to enhance the Venice code
👌 1
maybe we could create a non-Samza-based client for ingestion
f
Yeah definitely. We are probably going to work on that soon, BTW.
(At LI)
e
great
I will be happy to provide feedback/help
🤝 1
🚀 1