The architecture is sound. But half-assing a loadbalancer impl (vulcand) instead of embedding HAProxy is a terrible idea.
Zero Downtime Frontend Deploys with Vulcand on CoreOS
11–20 of 24 posts
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#12The architecture is sound. But half-assing a loadbalancer impl (vulcand) instead of embedding HAProxy is a terrible idea.
+1
confd + haproxy would be a better choice.
I have plans do roll the cocktail out, but running the HA prox y on each machine, much like the airbnb smartstack: http://nerds.airbnb.com/smartstack-service-discovery-cloud/
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#13We have something very similar at Balanced called Rump (routing umpire). It's written in python has a pretty cool rule based engine and a swappable backend.
It currently uses ZK and Chef, but it would be pretty interesting to use something like Etcd/Consul since no one wants to maintain a ZK cluster.
If there are individuals who want to hack on it, shoot me a msg on twitter and I'll see what I can do to speed up the process to open source it.
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#14The architecture is sound. But half-assing a loadbalancer impl (vulcand) instead of embedding HAProxy is a terrible idea.
So it's half assed if it's not HAProxy? How so?
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#15Earlier quoted context omitted.
If you're still interested in doing this for any reason, check out confd [0] for populating templates based on etcd values. [0] https://github.com/kelseyhightower/confd
Yes, confd was on my radar and looks promising. In my case, we've already got a wad of perl duct-tape out there and adding etcd was a baby step. Using confd would have been a later step. Vulcand will likely allow me to replace the whole shebang in one step.
it doesn't look like vulcan supports HTTPS, so it's a no-go for anything serious. Here's a post on my confd/nginx/docker setup for zero-downtime using etcd. http://brianketelsen.com/2014/02/25/using-nginx-confd-and-do...
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#16The architecture is sound. But half-assing a loadbalancer impl (vulcand) instead of embedding HAProxy is a terrible idea.
So it's half assed if it's not HAProxy? How so?
HAProxy is 14 years old, under active development by very competent people, used in thousands of production deployments and many hardware appliances.
Pretty much every large site you visit every day has HAProxy in the stack. It's one of the cornerstones of OSS infrastructure software.
Are you seriously asking how a 2 month old hack (vulcand) is half-assed in comparison?
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#17I'll ask about the reasoning instead of attacking the decision (since that didn't get a response).
Why not embed HAProxy?
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#18Earlier quoted context omitted.
Yes, confd was on my radar and looks promising. In my case, we've already got a wad of perl duct-tape out there and adding etcd was a baby step. Using confd would have been a later step. Vulcand will likely allow me to replace the whole shebang in one step.
it doesn't look like vulcan supports HTTPS, so it's a no-go for anything serious. Here's a post on my confd/nginx/docker setup for zero-downtime using etcd. http://brianketelsen.com/2014/02/25/using-nginx-confd-and-do...
I actually learned about confd via your post. Thanks for that.
In our case, HTTPS isn't an issue but there are good arguments for using a tried-and-true proxy. I see Vulcan as a quick way to test out docker containers for our architecture.
In the end I'll probably use confd to rewrite a VCL file for Varnish. It's just a matter of deciding how many containers and what level of sidekicking/ambassadoring I'm willing to put up with to "properly" organize things...
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#19Earlier quoted context omitted.
So it's half assed if it's not HAProxy? How so?
HAProxy is 14 years old, under active development by very competent people, used in thousands of production deployments and many hardware appliances. Pretty much every large site you visit every day has HAProxy in the stack. It's one of the cornerstones of OSS infrastructure software. Are you seriously asking how a 2 month old hack (vulcand) is half-assed in comparison?
If you're going to call something half-assed because it's not HAProxy then I for one would consider it more productive if this was framed in terms of metrics instead of something like "half-assed." Does HAProxy have a suite of metrics that can serve as benchmarks for all the projects out there that would dare to be less half-assed?
Re: Zero Downtime Frontend Deploys with Vulcand on CoreOS
#20Earlier quoted context omitted.
HAProxy is 14 years old, under active development by very competent people, used in thousands of production deployments and many hardware appliances. Pretty much every large site you visit every day has HAProxy in the stack. It's one of the cornerstones of OSS infrastructure software. Are you seriously asking how a 2 month old hack (vulcand) is half-assed in comparison?
If you're going to call something half-assed because it's not HAProxy then I for one would consider it more productive if this was framed in terms of metrics instead of something like "half-assed." Does HAProxy have a suite of metrics that can serve as benchmarks for all the projects out there that would dare to be less half-assed?
Why didn't you ask the question the other way around -- does vulcand have anything to show that it is better in any way than the current best-practice load balancer?
Reading the docs, the only thing that stands out is that it's tightly coupled with etcd, which seems like a really bad idea architecturally.