Monthly Update call
Me : I need 2 undivided weeks in the next 3 months to make some refactoring to our codebase, so we dont have issues in scaling.

Senior Manager (with years of tech experience almost equal to my age) :
Why would we need that, We are already hosted on AWS so we can just spin up more instances.

Me ( trying to hold the rage inside) : No it doesnt work that way, that will only hurt us more than help.

Him : What if we try with a bigger instance size. Its AWS, sure they will have a bigger instance.

Me : No, tha....
Him : (cuts me off) How about we use autoscaling and let AWS decide

Me : (Silently praying for a gun that shots across video calls.)

  • 1
    I’ve had this many times, never ends well.
  • 11
    You really need to take a piss but it takes too long? Just install more toilets or maybe a bigger one
  • 6
    Fight fire with fire:
    Sure! that will cost us! I think it might be very costly. Do you have cost estimates? no? ok - Then we should prepare some cost/benefit analysis documents! Btw, my refactor task will reduce operating costs, not increase them - can you make sure that goes into the documents?
  • 4
    @magicMirror I dont think they care about costs at all, even with the current infra, we are easily paying 40-60% more than we should.
  • 2
    @coldfire DAFUQ?
    Time to mine monero, while they pay for it....
  • 1
    Can you enlighten me why this won't work?, It's very cost inefficient but it should work no?
  • 4
    You can have a baby born in three months if you get three women to work on it instead of just one!
  • 8

    Scaling horizontally - more machines
    Scaling vertically - more resource for a machine

    Scaling horizontally is like ... Putting a knot at the end of a water hose and putting holes in the water hose.

    More holes - more water pours out.

    But... If the pressure reaches a critical point, the whole water hose just blows up.

    As an real example:

    In networking, scaling horizontally usually requires that the loadbalancer can distribute the work.

    Too many nodes and the loadbalancer will be overwhelmed by incoming connections - worst case going down under traffic, meaning none of the workers get any request.

    Other example: More nodes, but nodes take too long for requests.

    Yes you could add more nodes - but as the requests take too long, you're having a larger and larger mass of unfinished requests stuck in processing. Either the loadbalancer gives up at some point or you can't add more nodes / add nodes fast enough.

    Spreading work across several nodes as a band aid is never a good solution.

    The whole organism of loadbalancer and services needs to be balanced out or you'll end up with a doomsday machine
Add Comment