3 ms·

The article touches on it a bit, but in my experience microservices multiply operational problems (especially at startups where you don't have big, dedicated in

by nijave 3y ago

The article touches on it a bit, but in my experience microservices multiply operational problems (especially at startups where you don't have big, dedicated infrastructure teams). All of a sudden you have 5-10x things getting built in CI and deployed. You need some way to debug issues so usually distributed tracing comes up. Now that you have 10x as many of everything, you obviously want to try to centralize things like networking, authn/z, service discovery so you introduce some platforms to help with that... but someone has to run and maintain all that. That's fine if you want platform/infrastructure teams to maintain these but many startups only have a very small handful of people (10% or less of eng) handling CI/infra/release/performance/observability/networking/traffic mgmt with some title like "SRE" or "devops"

nine_k 3y ago
A typical startup with 20k DAU: "We need scale, microservices, k8s, CDNs, etc!". Stackoverflow: "We run a single .NET-based multi-tenant web app running across just nine web servers, at 5% to 10% of capacity" [1]. Hacker News: "HN still runs on one core, at least the part that serves logged-in requests, and yes this will all get better someday...it kills me that this isn't done yet but one day you will all see." [2]. HN had ~12M MAU by the end of 2022. [1]: https://www.datacenterdynamics.com/en/news/stack-overflow-still-on-prem-runs-qa-platform-off-just-nine-servers/ https://www.datacenterdynamics.com/en/news/stack-overflow-st... [2]: https://news.ycombinator.com/item?id=35157344 https://news.ycombinator.com/item?id=35157344
- rospaya 3y ago
  Does HN have the SLA and uptime that the startup signed with their clients and will half an hour of downtime sink HN?
  - nine_k 3y ago
    No. A number of startups don't either. I hazard to say that even giants like Tinder or Uber likely can have a 30 min outage and lose some revenue and goodwill, but not be hit by some exorbitant liabilities. Also, microservices add both resilience (by running many copies) and fragility (many loosely coupled moving parts). Which effect prevails, depends on many factors.
    - nijave 3y ago
      I think it's much more common to see SLAs in B2B vs B2C especially enterprise software sales (where customers are signing yearly contracts)
- nijave 3y ago
  In all fairness, HN and SO are fairly straight forward CRUD apps (esp HN) with a limited amount of 3rd party integrations. On the other end, you have companies like Zapier that integrate with 6k+ external services. With something like that in a monolith you'd be constantly redeploying to fix changes and you'd have a pretty wild dependency tree if you wanted to use official SDKs

by nijave 3y ago

nine_k 3y ago

A typical startup with 20k DAU: "We need scale, microservices, k8s, CDNs, etc!". Stackoverflow: "We run a single .NET-based multi-tenant web app running across just nine web servers, at 5% to 10% of capacity" [1]. Hacker News: "HN still runs on one core, at least the part that serves logged-in requests, and yes this will all get better someday...it kills me that this isn't done yet but one day you will all see." [2]. HN had ~12M MAU by the end of 2022. [1]: https://www.datacenterdynamics.com/en/news/stack-overflow-still-on-prem-runs-qa-platform-off-just-nine-servers/ https://www.datacenterdynamics.com/en/news/stack-overflow-st... [2]: https://news.ycombinator.com/item?id=35157344 https://news.ycombinator.com/item?id=35157344

rospaya 3y ago
Does HN have the SLA and uptime that the startup signed with their clients and will half an hour of downtime sink HN?
- nine_k 3y ago
  No. A number of startups don't either. I hazard to say that even giants like Tinder or Uber likely can have a 30 min outage and lose some revenue and goodwill, but not be hit by some exorbitant liabilities. Also, microservices add both resilience (by running many copies) and fragility (many loosely coupled moving parts). Which effect prevails, depends on many factors.
  - nijave 3y ago
    I think it's much more common to see SLAs in B2B vs B2C especially enterprise software sales (where customers are signing yearly contracts)
nijave 3y ago
In all fairness, HN and SO are fairly straight forward CRUD apps (esp HN) with a limited amount of 3rd party integrations. On the other end, you have companies like Zapier that integrate with 6k+ external services. With something like that in a monolith you'd be constantly redeploying to fix changes and you'd have a pretty wild dependency tree if you wanted to use official SDKs