Question about MCU Clustering in new version 3.1

[Naresh_R_1]

Hi Chunbo,

MCU Clustering in Version 3.1 is little confusing.

1. How many servers are required for MCU Clustering

Do we require Separate servers

1 for  Nuve and Sample another for cluster manager and another for agents??

Can we cluster with 2 mcu servers only

Also do we require separate server for Rabbitmq r can we use master MCU server as rabbitmq also?

Also configuration is little difficult than previous version

Thanks

Naresh

[Chunbo_H_Intel1]

Hi,

You will not be required to deploy multiple machines, since the whole MCU system could be run on a single server. Of course, you can run MCU on two machines. And RabittMQ can be run on primary master as you decide.

Thanks!

[Naresh_R_1]

Hi Chunbo,

I Have configured Cluster in version 3.0.1 update 2 using the reference guide.

Scenario :

We have configured 1 Master Server and started nuve, mcu and app services

2 slave servers running only agent service, and pointed rabbit mq server ip in woogen_config.js to Master server and with default port 5672 in both slave servers

When we tested this enviroment clients getting connected to only 1 slave.

Want to know how distributed system will work with multiple slaves

Thanks

Naresh

[Chunbo_H_Intel1]

Here is the documentation from our user guide:

"All workers including portals, webrtc-agents, rtsp-agents, recording-agents, audio-agents, video-agents in the cluster are scheduled by the cluster_manager with respect to the configured scheduling strategies in cluster_manager/cluster_manager.toml. For example, the configuration item "portal = last-used" means the scheduling policy of workers with purposes of "portal" are set to "last-used". The following built-in scheduling strategies are provided:

1. last-used: If more than 1 worker with the specified purpose are alive and available, the last used one will be scheduled.
2. least-used: If more than 1 worker with the specified purpose are alive and available, the one with the least work-load will be scheduled.
3. most-used: If more than 1 worker with the specified purpose are alive and available, the one with the heavist work-load will be scheduled.
4. round-robin: If more than 1 worker with the specified purpose are alive and available, they will be scheduled one by one circularly.
5. randomly-pick: If more than 1 worker with the specified purpose are alive and available, they will be scheduled randomly."

[Chunbo_H_Intel1]

Another thing is that the above policy ONLY available in v3.1. Please test in with the latest release.

[Naresh_R_1]

Hi Chunbo,

I Have Configured MCU 3.1 in 2 Server Machines still am not able to check how clustering works

In machine1 i have started the following servics

nuve, cpp, cluster-manager

In machine2 i have added rabbit host as machine1 ip-adress in each agent and controller and started following services

portal , audio-agent , video-agent, webrtc-agent, rtsp-agent, recording-agent

when  i am trying to access the room it says invalid token

please let me know for any configuration needed to work cluster

Below is the error 

ERROR: AMQPER - Connection error... { [Error: read ECONNRESET] code: 'ECONNRESET', errno: 'ECONNRESET', syscall: 'read' }
Error: read ECONNRESET
    at exports._errnoException (util.js:849:11)
    at TCP.onread (net.js:544:26)

Thanks

Naresh

[Chunbo_H_Intel1]

Hi Naresh,

According to the error output, have you set the correct RabbitMQ host and port in each component's configuration file?

Thanks!

[Naresh_R_1]

Hi Chunbo,

Can we use  Master Server as rabbit Host or we have create another server as rabbitmq server?

Thanks

Naresh

[Chunbo_H_Intel1]

Sure, you can.

Thanks!