[HN Gopher] So you wanna go on prem (2016)
       ___________________________________________________________________
        
       So you wanna go on prem (2016)
        
       Author : hackerthemonkey
       Score  : 27 points
       Date   : 2025-03-16 08:21 UTC (2 days ago)
        
 (HTM) web link (blog.lusis.org)
 (TXT) w3m dump (blog.lusis.org)
        
       | dark__paladin wrote:
       | Hugo is cool
        
       | supriyo-biswas wrote:
       | Is it still very common to use VMs for this, as opposed to
       | hosting your application in Kubernetes since it's stateless
       | anyway?
        
         | dehrmann wrote:
         | Not answering the question, but k8s doesn't entirely solve this
         | because of the DB. If you're a SaaS provider that reached this
         | scale, there's a good chance your architecture isn't just
         | stuffing everything in one Postgres instance, so shipping a VM
         | with Mongo, Postgres, Zookeeper, and Cassandra sorta makes
         | sense.
         | 
         | If you'd had ample foresight, it might have been nice to have
         | an architecture where the DB is truly abstracted away. Or not.
         | Maintaining two backend implementations is a pain and error-
         | prone.
        
       | alexpotato wrote:
       | I always think about the PagerDuty example:
       | 
       | They have their own datacenters, telco infra etc b/c if Amazon,
       | Azure or GCP goes down they want to be able to say:
       | 
       | "They all went down but PagerDuty alerts still went out to our
       | clients".
       | 
       | In other words, there are examples where managing your own infra
       | IN ADDITION to using cloud services is adding extra tolerance for
       | giant failures in the tech suppliers.
        
         | hackerthemonkey wrote:
         | Interesting. I didn't know they did. Though can't help but
         | thinking that this is all going be a lot less valuable if your
         | ovservability platform is on those mainstream cloud providers.
        
         | hn_throwaway_99 wrote:
         | The PagerDuty example seems really unique to me though.
         | 
         | That is, the _entire point_ of PagerDuty is to notify you if
         | your services go down. It seems like it would be colossally
         | stupid if PagerDuty were to say  "Sorry we weren't able to
         | notify you about your cluster on AWS going down, because we're
         | on AWS, too!"
         | 
         | For other businesses, I feel like the exact opposite incentives
         | are in play. I've seen this in practice: "Yeah, sorry boss our
         | service is down, but there is an AWS outage, so half of the
         | Internet is down - not much we can do about that!"
        
       | mrngm wrote:
       | Earlier discussion in 2016, with 102 comments:
       | https://news.ycombinator.com/item?id=11757669
        
       ___________________________________________________________________
       (page generated 2025-03-18 23:02 UTC)