Rudder

Matthieu CERDA wrote:

Hi Christophe,

Hi Matthieu,

So you finally managed to get Tsung to cooperate :D

Yep ;-) And now everything is much funnier ;-)

Well, to begin with, can you please take a look at the webapp's logs, and also try to boost the allocated RAM to Rudder in /etc/default/jetty ?

The Webapp's log is empty :-( Do you need DEBUG information ?

I can't / want add more memory ;-)

Thanks,

Matthieu CERDA wrote:

Well, to begin with, can you please take a look at the webapp's logs, and also try to boost the allocated RAM to Rudder in /etc/default/jetty ?

Done, setup rudder 2.9.2 on my laptop with JAVA_XMX=2048 and JAVA_MAXPERMSIZE=512, I can log in with 2K hosts.

That's awesome \o/
Thank you for the feedback

Christophe Nowicki wrote:

Jonathan CLARKE wrote:

OK, so if I understand correctly, this works fine if there is enough RAM?

Non, The issue is not about RAM requirements.
We all agreed that you need enough RAM for handling a lots of nodes.
But loading all nodes when the user is logging on, it's an issue.

Then I'll close this ticket, and hope it is useful to anyone having this same problem. Documentation work is ongoing to clarify Rudder's requirements in terms of RAM for large installations.

Christophe if you disagree, just say so, and we can reopen this.

I strongly disagree.

OK, sorry for the misunderstanding.

Can we please clarify what this ticket is about then? I read the title that say "can't login when the number of hosts is too large", then you said "I can log in with 2K hosts". So I consider that problem is fixed.

If you'd like to refocus the ticket on "Rudder should not query all nodes when checking login", that's fine by me, but can we agree to change the title to that?

Jonathan CLARKE wrote:

If you'd like to refocus the ticket on "Rudder should not query all nodes when checking login", that's fine by me, but can we agree to change the title to that?

I agree.

In our tests, even will >2000 nodes, we weren't able to reproduce a case where the login was not possible, or even where LDAP requests took more than a few seconds to return WHEN (a big when) no inventories where received by the endpoint in parallele.

On the other hand, we exibited #4497, where a memory exhaustion was initiated by accumulating incoming inventories to proccess. That memory exhaustion lead to global stalling of both Rudder inventory endpoint and Rudder UI part, which look similar to what was described here.