Maybe it's the monitoring fatigue I have from trying out so many different services/systems prior to settling on Riemann+Graphite+VictorOps, but I don't really see the killer feature of OpsDash that would make me jump to it.
Not to deride the service, of course - it might be a fine product. It just seems like most of the other (modern) services available, if just a little bit more consolidated.
Looks awesome, Signed up for the beta and will give it a go with ~ 80 VMware VMs and a few bare metals boxes. I have been looking for something similar for the last year.
I just hope that
1.) it is easily extensible (as there is no way it will have an agent for all the software we run). Ideally the interface would be language agnostic (i.e. like nagios where the interface between the server and the checking code is just STDIN/STDOUT).
2.) The pricing model is reasonable once it goes out of production. We run a lot of single service VMs to isolate services. With management tools like Salt, it really makes it easy to mange, and a well tuned linux VM isn't that hungry on the RAM. It does mean that I can't afford to spend much on "per node" pricing. Especially the $10 per server per month that seems to be the norm.
3.) I hope there is a way to push stats into the dashboard from other processes.
If it were absolutely perfect in every other way, it would still have to be priced in a way that I could afford. I have no idea of how much it's going to be, and I can't put much effort into beta-testing without at least a rough estimate.
Remember that it has to beat incumbent systems and open-source systems.
Seems a pretty cool project, but I prefer to have my monitoring service completely outside of the infrastructure I need to monitor; Who would alert me if everything goes down ?
Have a lightweight monitoring server inside your regular infrastructure dedicated to monitoring the monitoring service. Have the monitoring service on the other cloud watch everything else (Including your monitoring monitor)
The site looks good. I feel that the devops world is hungry for a _full_ replacement of graphite/zabbix/reimann, etc, but the pricing hasn't been appealing enough for the best tools.
Per-host pricing schemes punish large installations and engineers who've created services that run on small, horizontally scaling instances.
In the end most of us end up with some wacky combination of different services.
With that many servers, they should throw away the big friendly icons and just serve the labels. It's not like you couldn't recolour the labels to match and attach your number-of-alerts counter.
