On Mon, Jan 3, 2022 at 1:19 PM Daniel McCarty <[email protected]> wrote:

> Apologies in advance if this was asked before, but I’ve looked at
> guacamole.apache.org and the usual places and haven’t seen a definitive
> answer.
>
>
>
> We have a small Guacamole installation that forwards our sales and field
> service teams to a handful of Windows demo servers via RDP.  The main guac
> server was running in AWS as a t3a.micro.  (which is an AMD EPYC 7000 @
> 2.5GHz, 2-core, 1GB RAM machine.)  After uptime of about 2 weeks we would
> start seeing datadog alerts about the service crashing and restarting, and
> about every 4 weeks the server itself would hang and need to be restarted.
> This happened for a few months.  We checked syslog, localhost_access, and
> the usual suspects, and eventually narrowed the issue down to the system
> running at about 90% available memory.  We replaced the instance with a
> t3a.small (with 2X the memory).
>
>
>
> Since then the new machine has been up for about 8 straight weeks without
> a hiccup, which is good news.  But before we mark this completely resolved,
> I just wanted to double check that we’ve gone about this the right way and
> the earlier machine was just too small?
>
>
>

How many users do you have using the instance? If you're running both
Guacamole Client (Tomcat) and guacd on the same system, then yes I'd
suspect that the t3a.micro was a bit on the small side, and that bumping to
the t3a.small is probably sufficient. You also might consider using the T4g
class - there's nothing about Guacamole that requires Intel-based
processors, and AWS's Gravitron (aarch64) processors run things just fine.
The T4g instances are a little less expensive, too, so you'll save some
money over time.

IIRC, Mike had come up with some general numbers that I can't exactly
remember off the top of my head, but I think it was something along the
lines of 1 CPU core and 4GB of RAM for every 25 users, dependent upon what
kind of workload they're doing (won't work for 25 x YouTube watchers).

-Nick

>

Reply via email to