Question about Khepri

[Radu Marian]

Hi All,

We have started experimenting with the new Khepri datastore in RabbitMQ 3.13 and we have a question about it:

Is it possible to fine tune the internal RAFT parameters like for example the Khepri leader election timeout via some env var or configuration file?

We have noticed that if the Khepri leader goes down unexpectedly, the cluster can take some seconds to re-establish consensus and start working again. We would require a faster recovery time here.

Thanks,

Radu.

[Jean-Sébastien Pédron]

Hi!

RabbitMQ doesn’t have such a parameter. Khepri could pass it if it was configurable, but I don’t know who is responsible for that the first place. Could you please file an issue on GitHub so that we don’t forget about this?

You could try with RabbitMQ 4.0.x to see if something changed for your workload. Also, note that Khepri in RabbitMQ 3.13 is unsupported and you won’t be able to upgrade to 4.0.x if Khepri is enabled. I recommend that you to all your tests with RabbitMQ 4.0.x.

--

Jean-Sébastien Pédron

RabbitMQ core team