Log into certain nodes of an instance?

[Kalaiarasan Pus]

Do we have any facility to log into specific node of an instance?

I was able to find the nodes of an instance by checking sys_cluster_state table but unable to find any way to login into certain nodes. Is this even possible without using any chrome plugin or contacting support?

[kshitiz2]

Hi iamkalai,

Try to logout and re-login to your instance several times, you will see the logged in nodes changing for your instance account.

Regards,

Kshitiz

[DirkRedeker]

Hi Good question. Let me ask another question on that? Which experience are you asking for? If you try to forward specific users to specific nodes, I am wondering, if an instance can move me from one node to another, as the node is already selected before the user is known and logged in. Maybe it is possible to have different 'login.do' versions, without knowing, if this is even possible, and how to do? What is the reason for this requirement? Do you want to log in to check performance? Let me know about your ideas. BR Dirk

[Kalaiarasan Pus]

We are seeing random issues where transactions are getting stuck in some nodes and we get user complaint of slow performance etc. Having the ability the login to specific nodes will allow us to do atleast basic troubleshooting and also verify if the issue of slowness actually exists before opening a support ticket.

[DirkRedeker]

Hi

I get what you are trying to do. Maybe there is a different option to proactively track something like that.

I just compare your situation with admins, who want to check "hanging" VMs in a virtual computer environment.

If that machine is hanging anyhow, why should I try to log on to that machine and stress it even more.

To get feedback from active nodes, I guess it makes more sense, to get some "heartbeat" from that node, to know about the health state. 

So, I recommend to try to solve the requirement is a different way. Do you have an idea, how one node could "report" it's own health state? Maybe by running schedulers on specific "nodes", which will evaluate progress about what happens?

I am not good in system performance counters, but this will be another idea to look for.

There must be a way to diagnose the node, without logging to that node. Anyhow, logging in to that node looks like a manual task, which maybe is also not the best way.

Maybe you can put some logging information to the steps, the user are complaining about, to find out about race conditions in your instance.

I hope, this gives you some useful hints and ideas about how to approach.

Let me know, and mark my answer as correct / helpful, please.

BR

Dirk