Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Critique my planned CDN/EDGE layout on buyVM
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Critique my planned CDN/EDGE layout on buyVM

I managed to snag the third buyVM 1024 slice so now I have one in each location, and can enable anycast. Here is the plan in shitty diagram mode. I have a few questions.



  1. Should I roll a DNS server (I dont think I am at all qualified to do that) or should I use a RageDNS/Route53/Other Off the shelf option? If other, please post suggestions.
  2. Can I have a healthcheck/corosync heartbeat/other to signal the Nginx/Rproxy nodes if a backend app node goes offline?
  3. Should I consider using the anycast IP on the nginx/rproxy nodes?

Anything I should consider that I am not here?

Comments

  • LeviLevi Member

    Powerdns? What is galaraDB?

  • GalaraDB writes are only as fast greatest amount of latency between nodes, in this case Vegas and Luxemburg. Make sure you use transactions or else the app will be very slow. State snapshot transfers are another concern when a node goes down and comes back online.

    If you use the anycast IP and a node goes down you won't have any way to route around it as the anycast routing is outside of your control. The budget way to do it is using a geo routing DNS with health checks and low TTL.

    Thanked by 1jollymon
  • NeoonNeoon Community Contributor, Veteran
    edited March 2021

    For a experiment sure, I did the same just for fun, I ran tests on ripe atlas but the results where not what I did expect, so I dropped it.

    Also, you are limited to 3 pops.
    Mariadb looks decent, however, if your vps goes down, the traffic will not be rerouted, only if the hole DC goes down.

  • umiumi Member
    edited March 2021

    I was tested similar setup last summer and the main headache was the regional black holes if node went down. You simply cannot stop announcing those anycast ips for that unlucky region. The LUX,NYC and LAS locations are not that far from each other for a NYC server coupled with superfast cloudflare dns fell short to cover it. The biggest latency with this setup is APAC region with 200ms ping time to nearest node.

  • @cablepick said:
    The budget way to do it is using a geo routing DNS with health checks and low TTL.

    Yeah this is what I am thinking makes the most sense, and also lowers my overhead on managing. I know Route53 does healthchecks and geo so that seems like a way better idea. More deployment config work, but at least I will know what I am doing.

  • @umi said:
    I was tested similar setup last summer and the main headache was the regional black holes if node went down. You simply cannot stop announcing those anycast ips for that unlucky region. The LUX,NYC and LAS locations are not that far from each other for a NYC server coupled with superfast cloudflare dns fell short to cover it. The biggest latency with this setup is APAC region with 200ms ping time to nearest node.

    I take it you ditched it at some point?

  • Pony really needs an APAC location

  • 10301030 Member

    @corbpie said:
    Pony really needs an APAC location

    Singapore would be nice

    Thanked by 1lorian
  • Since you are using Anycast, you don't have to rely much on DNS except for the case where you want to route specifically or different IP or those nodes which aren't supported on anycast.

    Go with Route 53 or Flexbalancer (https://perfops.net/flexbalancer)

    Don't use 1 single provider or DDoS protector which can bring down if any mission critical usage. But for basic sites, it is fine.

    PM me if you need more help or review. :smile:

  • umiumi Member

    @jollymon said:
    I take it you ditched it at some point?

    Yep. With Rage4 you can announce anycast ip to your own servers (Chicago,Frankfurt,Singapore for example) but it will be more expensive than BuyVM, especially when traffic increase. But for dns usage it is still ok. You can make decent setup by combining buyvm and rage4 if delay of 2 minutes to update dns records is acceptable. Some dns resolvers really stick to 2 minutes TTL no matter what your dns record said.

Sign In or Register to comment.