2019-06-17 13:10:22 UTC - pradeep: HI Team,
How does the pulsar manage the TCP connection for the more number of the 
consumer ? Does it create a new TCP connection for every consumer?
Do the secondary consumers (not the master one who is actively receiving data) 
also hold the TCP connection in failover mode?
I have requirements to Start more than 1 lakhs consumer for 1 lakhs topic in 
failover mode, so I would be using consumer listener for parallel processing. 
Can you suggest me do pulsar will support this? Does it do TCP multiplexing 
internally?
----
2019-06-17 16:40:42 UTC - Matteo Merli: Yes, by default there will be 1 
connection per host, from a single PulsarClient instance 
----
2019-06-17 17:08:13 UTC - pradeep: Does the pulsar consumer internally operate 
in a single thread even though the application creates any number of consumer 
instance?
----
2019-06-17 17:38:20 UTC - pradeep: If one TCP connection per PulsarClient , 
then one TCP connection will be used by N number of consumers to receive 
messages from pulsar. Let's consider if all consumers are  created using one 
pulsar client. Am I right ?
----
2019-06-17 18:31:06 UTC - Matteo Merli: That’s correct
----
2019-06-17 18:31:37 UTC - Matteo Merli: By default there’s 1 connection 
per-host and 1 IO-thread to manage the connections
----
2019-06-17 18:32:06 UTC - Matteo Merli: that can be overridden in the 
PulsarClientBuilder, though typically it’s not necessary to change
----
2019-06-17 18:38:05 UTC - pradeep: ok. So If I understand correctly, Consumers 
are not holding any TCP connection with pulsar cluster, all functionality like 
consumers failover, a different type of subscription managed using the same 
single TCP connection
----
2019-06-17 19:00:26 UTC - pradeep: So if only one TCP connection per client 
(does not matter how many number of consumers), then we can create lakhs of 
consumer instance to consume messages from lakhs of topic . Only problem I do 
see is , may be a little delay in receiving message from the pulsar . Please 
correct me If I am wrong
----

Reply via email to