Suhail Patel

@suhailpatel

🏦 Backend/Platform Engineer at . 🚋 Previously at 👩🏽‍🚒 He/Him 🗺️ British 🇬🇧 and Indian 🇮🇳

London, UK
Vrijeme pridruživanja: ožujak 2008.

Tweetovi

Blokirali ste korisnika/cu @suhailpatel

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @suhailpatel

  1. Prikvačeni tweet
    17. stu 2019.

    Here's the live request flow on a large subset of the 1500+ microservices on the Platform. A lot of things are involved in running a bank! If you used the app at around 3pm today, you might be represented😃

    Poništi
  2. 2. velj

    I want to start measuring my stress in DEFCON DBA levels 😅 (also this FOSDEM talk titled ‘RTFM’ on the Postgres track is superb , really fun war stories 🙌)

    Poništi
  3. 2. velj

    I really want my own robocop gopher after seeing this at FOSDEM 😍

    Poništi
  4. 2. velj

    On a lighter note, I got introduced to Danish Hip Hop last night and I’m hooked. This stuff proper kicks!

    Poništi
  5. 1. velj

    If you are relying on service mesh to take care of this, you have architected it wrong in my opinion). A lot of this isn’t service/app agnostic. Service mesh might as well be a euphemism of throw it across the network and pray it works 🔥

    Poništi
  6. 1. velj

    Completely agree, unless your service sits within a remote system (think Java Agent or Socket IPC), writing a thin layer basically means you are doomed to reinvent the protocol (and most likely screw it up) Fatter layers may provide value after some time (eg a routing proxy)

    Poništi
  7. 1. velj

    There is a fine balance. Don’t ignore the network and portray it as ‘free’. You have to invest a lot in application throttling, fallbacks in failure, proper error handling, retries where it makes sense. Engineers shouldn’t be ignoring this and pray that something will do it.

    Prikaži ovu nit
    Poništi
  8. 1. velj

    There are some benefits to microservices, easier ownership, scoped domains, theoretical individual scalability.

    Prikaži ovu nit
    Poništi
  9. 1. velj

    By the time you are doing the analysis, you are already in a trace hundreds or thousands of spans deep. You step back and lose a moment of your soul at the amount of waste. The majority of computation and data over the wire is thrown away

    Prikaži ovu nit
    Poništi
  10. 1. velj

    In my experience, no matter how much tooling you provide (tracing and monitoring et al), there is little incentive to optimise until it really starts hurting. Doing regression analysis is pretty difficult

    Prikaži ovu nit
    Poništi
  11. 1. velj

    If you abstract away the network from engineers and keeping calling it ‘free’, that probably of failure just keeps growing until someone steps back and takes a cold hard look to ask ‘why doesn’t this scale with all the money we’re throwing at it’

    Prikaži ovu nit
    Poništi
  12. 1. velj

    Not even mentioning all the compute time wasted parcelling data off. No matter what encoding you use, there is a runtime cost that only grows as your systems get used more

    Prikaži ovu nit
    Poništi
  13. 1. velj

    If you’re miraculously built an architecture which is tolerant, you will look at your cloud bill and take a weep. That line item that says ‘Data Transfer’. The network is not free

    Prikaži ovu nit
    Poništi
  14. 1. velj

    You’ve spent all your energy battling with enabling microservices and yet your failure modes are akin to a monolith

    Prikaži ovu nit
    Poništi
  15. 1. velj

    If/When that core system is exhibiting failure or a spike in latency, you now need to deal with all the additional overhead in the network. You further compound your failure till it cascades to other systems, and now they are failing

    Prikaži ovu nit
    Poništi
  16. 1. velj

    No matter how decoupled you think your systems are, there are definitely components which are core and shared amongst most of your services. This may be a service, a data store, another system etc.

    Prikaži ovu nit
    Poništi
  17. 1. velj

    Every single hop you add in your path, you reduce your chance of successfully completing a piece of work. The failure compounds

    Prikaži ovu nit
    Poništi
  18. 1. velj

    >Instead of function calls and class instantiation, they’re initiating things and throwing it over a network and hoping that it comes back This resonates so well. I have thoughts so a thread… (also thanks for pointing to this post)

    Prikaži ovu nit
    Poništi
  19. 30. sij

    Feels a bit bittersweet visiting the EU Parliament when the UK is crashing out 😩

    Poništi
  20. 29. sij

    Come see and me give our talk ‘Banking on Kubernetes, the Hard way, in Production’ 🏦 at Amsterdam! Fun failure stories promised 🙏

    Poništi
  21. 28. sij

    I’ve been using for the past two days on large and complex Go repos and it’s been a joy to work with! I definitely still want to get to the bottom of why gopls crawls to a halt in VSCode but for now, GoLand is a god send 👏

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·