When using a standalone gateway, I feel gRPC has a problem

Camunda 8 Topics Discussion & Questions
1

version： 0.24.0-alpha1
Gateway and broker are deployed on two Linux respectively.

@ZeebeWorker(type = "node1")
public void node1(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

@ZeebeWorker(type = "node2")
public void node2(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

@ZeebeWorker(type = "node3")
public void node3(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

@ZeebeWorker(type = "node4")
public void node4(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

@ZeebeWorker(type = "node5")
public void node5(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

@ZeebeWorker(type = "node6")
public void node6(final JobClient client, final ActivatedJob job){
    client.newCompleteCommand(job.getKey()).variables(job.getVariables()).send();
}

There is nothing special about my jobworker.
But jobworker cannot get jobs in real time. Even if it is polling to get the job, it should not be 20 seconds. It is too regular. When I used the embedded gateway, I did not encounter this problem.

WechatIMG51
WechatIMG511404×678 105 KB

2

Hey @wenminglei

how do you deploy Zeebe? Do you use the helm charts?

Greets
Chris

3

Hey，@Zelldon
I did not use helm charts.
I deployed the gateway on one linux and the broker on the other linux.
My demand is that as long as there is a job, the jobworker can respond immediately, or quickly. We pay great attention to whether a process can be completed quickly.
I have seen the code that the client interacts with the gateway. (Not fully understood)
Long polling and Job Queueing, these two mechanisms seem to limit the rapid completion of a workflow to some extent.Not as fast as the listener mechanism.

4

@wenminglei that is expected, as the listener mechanism is in memory and it doesn’t need to go over the network. Now you have a distributed system. @Zelldon do we have a recommended configuration (brokers / client) for when workers needs to react fast?

5