HAproxy setting problem


I’m doing the availability test about cockroachdb using Jmeter, HAproxy.
I use jdbc request call on Jmeter.

Once the jmeter has been selected continuously with 10 threads.

On the test, When a node died, the HAproxy recognized it and distributed the existing thread connection to two other nodes.

If the dead node is restarted, the haproxy recognizes it and distributes it to the existing two nodes and the restarted node 1 when a new request is received.

However, only two existing nodes receive the request for 10 threads that were previously performed by jmeter. I think it is because I use connection pool.

I restarted haproxy, but all requests were redistributed only after forced termination. The graceful restart was not performed properly.

Can you change the haproxy setting to distribute existing connections to new nodes?


Do you know another Load Balancer open source recommended that dynamically switches server connection?

Hey @yeri

Here’s a list of other open source load balancers, but I’m not familiar with any of these besides Haproxy.

As for Haproxy, I’ll need to take a look in order to see if you can configure this to distribute connections to new nodes.



Hey @yeri

I’m not sure if this would be helpful, but it looks like Haproxy has two settings nbthread and nbproc that seem like they deal with connections, you can read more about them here.

Let me know if this helps.